RE - MOO - VING BARRIERS WITHIN LABOR: EXPLORING CURRENT EVENTS RELATED TO DAIRY AND POULTRY LABOR MARKETS By Danielle Megan Kaminski A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for th e degree of Agricultural, Food, and Resource Economics - Doctor of Philosophy 2020 ABSTRACT RE - MOO - VING BARRIERS WITHIN LABOR: EXPLORING CURRENT EVENTS RELATED TO DAIRY AND POULTRY LABOR MARKETS By Danielle Megan Kaminski Dairy labor markets have the potential to undergo notable change in the near future due to tightening labor markets and increased activity by worker advocacy groups. Given this context, this dissertation explores three areas related to dairy labor markets. In the first essay dairy workers are surveyed to identify their preferred compensation packages to assist farmers in offering the most attractive benefits to recruit and retain workers. The final two essays move along the supply chain to see how consumers may impact dairy l abor markets particularly in relationship to ongoing demand for changes to animal welfare practices. In the second essay whether consumers are willing to pay a price premium for a label certifying working conditions is estimated. Additionally, attention is directly to answering, if consumers are willing to pay a premium (1) is it greater than or less than a premium for animal welfare labeling and (2) does it change under different information settings. Finally, ces for worker welfare practices compared to animal welfare practices. For additional robustness, the last two essays are also applied to the egg production system to explore labor effects in other animal - based agricultural production sectors. Copyright by D ANI ELLE MEGAN KAMINSKI 2020 iv ACKNOWLED GEMENTS When a project takes years to complete there are a number of people who were likely involved and essential to its completion along the way. I would like to thank a few of those people by name here. However, before I do I would like to give a gen eral word of thanks to and time, and to my friends and family who supported me along the way. My longest - term supporters have been and continue to be my parents, B ruce and Diane Kaminski. I am certain that I could not have finished this process without you. Thank you and I love you. Dr. Vincenzina Caputo invested the most amount of time directly in this project. She always encouraged me to look outside the box to e xplore new and emerging societal issues, using a well - established methodology that she has helped refine. I have learned so much about choice modeling and research methods from her aid and example. Additionally, this research would not have been possible w Maria Porter demonstrated project management, including supervision of research assistants, which I can carry with me into future projects. Thank you to Alondra, Jessica, and Maria for their research assistanc e in enumeration and translation. Dr. Melissa McKendree and Dr. Mark Skidmore provided general support and helpful comments and suggestions throughout the dissertation. The latter was also graciously provided by the additional co - authors on these papers. T hank you all. Finally, thank you to Phil Durst and Stan Moore who invited me to work with their dairy clients and introduced me to their Ag. HR work group, both experiences which enhanced my thinking on the topics of this dissertation. v TABLE OF CONTENTS LIST OF TABLES ................................ ................................ ................................ ........................ vii LIST OF FIGURES ................................ ................................ ................................ ....................... ix KEY TO ABBREVIATIONS ................................ ................................ ................................ ........ ix ESSAY 1: . 1 1.1 Introduction ................................ ................................ ................................ ........................... 1 1.2 Experimental Design ................................ ................................ ................................ ............. 4 1.2.1 Discrete Choice Experiment: Attribute Selection and Experimental Design ................ 4 1.2.2 Eliciting Risk and Time Preferences ................................ ................................ .............. 8 1.3 Data ................................ ................................ ................................ ................................ ....... 9 1.4 Estimation Methods ................................ ................................ ................................ ............ 13 1.5 Empirical Results ................................ ................................ ................................ ................ 17 1.6 Conclusion ................................ ................................ ................................ .......................... 27 APPENDIX ................................ ................................ ................................ ............................... 31 REFERENCES ................................ ................................ ................................ ......................... 40 ESSAY 2: ARE CONSU MERS WILLING TO PAY MORE FOR ANIMAL WELFARE OR WORKER WELFARE LABELS? AN APPLICATION OF EGG AND MILK PRODUCTS UNDER DIFFERENT INFORMATION SETTINGS ................................ ................................ . 45 2.1 Introduction ................................ ................................ ................................ ......................... 45 2.2 Background and Research Hypotheses ................................ ................................ ............... 48 2.3 Egg Study ................................ ................................ ................................ ............................ 51 2.3.1 Choice Experiment and Survey Desi gn ................................ ................................ ....... 51 2. 3 .2 Between - Subjects Treatments .54 2.3. 3 Data ................................ ................................ ................................ .............................. 55 2.3. 4 Data Analysis ................................ ................................ ................................ ............... 57 2.3. 5 Results ................................ ................................ ................................ .......................... 60 2.4 Milk Study ................................ ................................ ................................ .......................... 64 2.4.1 Choice Experiment and Survey Design ................................ ................................ ....... 64 2.4.2 Data ................................ ................................ ................................ .............................. 66 2.4.3 Data Analysis ................................ ................................ ................................ ............... 68 2.4.4 Results ................................ ................................ ................................ .......................... 69 2.5 Discussion and Policy Implications ................................ ................................ .................... 72 2.6 Conclusion ................................ ................................ ................................ .......................... 76 APPENDIX ................................ ................................ ................................ ............................... 78 REFERENCES ................................ ................................ ................................ ......................... 88 vi ESSAY 3: WHICH CAME FIRST: THE CHICKEN (COW) OR THE LABORER? THE U.S. .......................... 96 3.1 Introduction ................................ ................................ ................................ ......................... 96 3.2 Background and Farm Practice Identification ................................ ................................ .. 100 3.2.1 Dairy Industry ................................ ................................ ................................ ............ 101 3.2.2 Poultry Industry ................................ ................................ ................................ ......... 104 3.3 Data and Methods ................................ ................................ ................................ ............. 106 3.3.1 Online Survey ................................ ................................ ................................ ............ 108 3.3.2 Econ ometric Model ................................ ................................ ................................ .... 111 3.4 Results ................................ ................................ ................................ ............................... 114 3.4.1 Results of the Dairy Application ................................ ................................ ................ 116 3.4.2 Results of the Poultry Application ................................ ................................ ............. 120 3.5 Discussion ................................ ................................ ................................ ......................... 124 3.6 Conclusion ................................ ................................ ................................ ........................ 125 APPENDIX ................................ ................................ ................................ ............................. 128 REFERENCES ................................ ................................ ................................ ....................... 136 vii LIST OF TABLES Table 1.1 Choice Experiment Attributes and Levels 5 Table 1.2 Sample Distribution by Work Status 10 Ta ble 1.3 Percent of Respondents Earning Benefits by Employment in Dairy Sector 13 Table 1.4 Estimates from the Mixed Logit Model for the Full Sample 18 Table 1.5 Marginal WTP Estimates (Based on Full Sample (Table 1.4, Model 3) ) 22 Table 1.6 Marginal WTP Est imates as Hourly Wages of Sub - samples (Based on Table A1.3) 26 Table A1.1 Sample Distribution Comparisons 34 Table A1.2 Sub - sample Distribution Comparisons 37 Table A1.3 Estimates from the Mixed Logit Model of Sample Sub - populations 38 Table 2.1 Egg Exper imental Design: Attributes and Attribute Levels 52 Table 2.2 Information Treatments 54 Table 2.3 Egg Sample Demographics 56 Table 2.4 Estimates from the MXL - EC with correlation in WTP - space for Eggs 61 Table 2.5 Milk Experimental Design: Attributes and Att ribute Levels 65 Table 2.6 Milk Sample Demographics 66 Table 2.7. Estimates from the MXL - EC with correlation in WTP - space for Milk 69 Table A2.1 Cholesky Matrix from Model - C estimates, Eggs Study 82 Table A2.1 Cholesky Matrix from Model - L estimates, Eggs Study 82 Table A2.1 Cholesky Matrix from Model - M estimates, Eggs Study 82 Table A2.4 Cholesky Matrix from Model - C estimates, Milk Study 87 Table A2.4 Cholesky Matrix from Model - L estimates, Milk Study 87 Table A2.4 Cholesky Matrix from Model - M est imates, Milk Study 87 viii Table 3.1 Animal and Worker Welfare Practices in the Dairy Industry Included in this Study 102 Table 3.2 Animal and Worker Welfare Practices in the Poultry Industry Included in this Study 105 Table 3.3 Summary Statistics of Basic De mographics 108 Table 3.4 Latent Class Model Comparisons 115 Selected Production Practices in the Dairy Industry 117 View on the Importance of Selected Production Practices in the Poultry Industry 121 Table A3.1 Experimental Design 129 Table A3.2 Summary Statistics of Basic Demographics t - Tests of Differences in Sample Average across Dairy Latent Classes 130 Table A3.3 L Importance of Selected Production Practices in the Dairy Industry 132 Table A3.4 Summary Statistics of Basic Demographics t - Tests of Differences in Sample Average across Poultry Latent C lasses 13 3 Importance of Selected Production Practices in the Poultry Industry 135 ix LIST OF FIGURES Figure 1.1 Example Choice Task 8 Figure 1.2 Distribution of Reported Cu rrently Hourly Wage by Employment in Dairy Sector 11 Figure 1.3 Distribution of Reported Average Weekly Hours Worked by Employment in Dairy Sector 12 Figure A2.1 Example Egg Choice Question 82 Figure A2.2 Example Gallon Milk Choice Question 83 Figure 3.1 W orker and Animal Welfare Practices in the Poultry Industry BWS Sample Question 107 ix KEY TO ABBREVIATIONS BWS Best - Worst Scaling DCE Discrete choice experiment EFI Equitable Food Initiative FARM Farmers Assuring Responsible Management MNL Multinomi al logit model MXL( - EC) Mixed logit model (with error component) NAWS National Agricultural Workers Survey RUT Random utility theory USDA United States Department of Agriculture WTP Willingness - to - pay 1 ESSAY 1: REFERENCES: A CASE S TUD Y IN MICHIGAN 1.1 Introduction Many economic sectors in the United States are suffering from labor shortages, including the agricultural sector (Taylor, et al., 2012; Hertz and Zahniser 2013; Fatka 2019). Labor shortages are fostered in part due to decli ning immigration from Mexico, as these immigrants compose a sizeable portion of agricultural laborers (Charlton and Taylor, 2016). Additionally, tight labor markets increase competition for workers even among low - skilled employment sectors. Since the agric ultural sector in the United States employs only 1.3% of the workforce ( US DA ERS, n.d.) understanding the main factors driving employees to choose agricultural labor is becoming crucially important. Understanding preferred working conditions among dairy w orkers may help in alleviating the labor shortage and high turnover in the industry. Thus, this study uses a discrete choice experiment (DCE) (Hensher et al. 2005; Hensher et al. 2015) to assess current and potential dairy ferentiated by alternative compensation packages. Specifically, we limit our sample to individuals in Michigan, a high dairy production state. In addition, since prior related research has shown that contracts at other points in the supply chain are influe nced by behavioral preferences (Fischer and Wollini, 2018; Khanna, et al., 2017; Krah, et al., 2018), this study also elicits risk and time preferences to explore whether preferences for alternative compensation packages vary across different worker groups . This study contributes to the literature in three important ways. First, to the best of our preferences requiring trade - offs between compensation benefit s. While there is an extensive 2 body of literature on how farmers enter production contracts domestically (for example, Bergtold, et al., 2017; Khanna, et al., 2017; Krah, et al., 2018) and abroad (for example, Broch and Vedel, 2012; Fischer and Wollini, 20 18), which is analogous to an employment contract, there is less information on how agricultural employees enter into work relationships. Existing literature in this vein tends to be focused on rural medical professionals in developing countries (for examp le, Ubach, et al., 2003; Huicho, et al., 2012; Miranda, et al., 2012; Rockers et al., 2012; Rao, et al., 2013; Rockers et al., 2013; Kunaviktikul, et al., 2015; Scott, et al., 2015; Song et al., 2015; Efendi, et al. 2016; Smitz, et al., 2016; Takemura, et al., 2016). Only recently has literature on employment preferences in agricultural sectors been introduced by Van den Broeck, et al. (2016) in Senegal and Schuster, et al. (2017) in Peru. Research on U.S. agricultural employee work decisions has relied on secondary data to (2010) is the only paper to explore agricultural worker preferences for employment conditions in the United States using primary data. However, th eir approach is limited in that they ask for conditions that would make work more appealing in a setting that is removed from real - life employment decisions. W e help introduce a methodology still new to the field of agricultural job preferences. More spec ifically, together with Mas and Pallais (2017), this study extends the use of discrete choice experiments from other fields into the agricultural labor economics toolkit. A DCE, such as the one employed in this study, dictates that respondents make choices between alternatives, preventing respondents from rating all attributes as equally important. Our experimental design mirrors real - world decision making for job choice by asking respondents to choose between job 3 profiles described by compensation packages varying in their benefit attributes and attribute levels. Our next contribution borrows from the farmer contract literature, incorporating risk and time preferences to explore preference heterogeneity across individuals (Fischer and Wollni, 2018; Krah, e t al., 2018; Bergtold, et al., 2017; Khanna, et al., 2017; Broch and Vedel, 2012) . To our knowledge, this is the first study that explores behavioral factors (risk and time preferences) that contribute to employment contract decisions. Finally , to the best of our knowledge, this is the job preference literature by introducing a case study of a hard to reach population. Survey results in this study suggest that dairy workers work many more hours tha n non - dairy workers who are potentially interested in joining the dairy industry labor force. However, workers reveal that they would prefer working fewer hours per week on average. Additionally , our results generally future - oriented individuals value a retirement plan more than present - minded individuals. Besides this interaction with time preferences, behavioral preferences (risk and time prefere nces) do not seem to influence compensation package preferences. The next section describes the design of the choice experiment and experiments for eliciting risk and time preferences. In the third section, we describe the data. Then we outline estimation methods, followed by the empirical results. Finally, we conclude with a summary of results, implications and limitations of the study, and suggestions for potential further research. 4 1.2 Experimental Design A tablet - based survey via Qualtrics was adm inistered face - to - face to prospective, current, and past dairy workers, who had the choice of English or Spanish translations. 1 The survey had three sections. It began with questions on demographics and work history. Section two was a choice experiment on job selection. Finally, risk and time preference questions were asked. 1.2.1 Discrete Choice Experiment: Attribute Selection and Experimental Design In our DCE application, workers make a discrete choice from a set of presented job options, which contai n alternative compensation packages describe d by different attributes (benefits and hours worked) combined within each choice task or choice question. To identify the benefits and thus design the compensation packages to include in the experiment, we follo wed two steps. First, prior literature on compensation preferences was consulted to identify the types of compensation available in various industries. Then, focus groups with dairy laborers were conducted to identify the labor benefits in the dairy indust ry and thus redefine the compensation packages. In what follows, we describe in detail the procedures followed in each step. In identifying the prior literature on compensation preferences, we focus on three streams of prior literature: job retention/sati sfaction factors in U.S. agriculture , job preferences of rural medical professionals , and job preferences in international agricultural sectors. 2 From the 1 Additionally, respondents could choose to complete the survey themselves (read it) or with the aid of a research assistant (have it verbally read to them and answer orally). 2 There are a number of relevant studies here. For those on job retention/satisfaction in U.S. agriculture see Gabbard and Perloff (1997) and Nolte and Fonseca (2010). For job preferences of rural medical professionals see Ubach, et al. (2003), Huicho, et al. (2012), Miranda, et al. (2012), Rockers, et al. (2012), Rao, et al. (2013), Rockers, et al. (2013), Kunaviktikual, et al. (2015), Scott, et al (2015), Song, et al. (2015), Efendi, et al. (2016), Smitz, et al. (2016), and Takemura, et al. (2016). For job preferences in international agricultural sectors see Van den Broeck, et al. (2016) and Schuster, et al. (2017). 5 attributes used in prior works we identified the following list of attributes that appeared in at l east two streams of literature: (quality of) employer provided equipment, health insurance/care, hours worked, housing, length of contract, shift schedule, training, transportation services to/from work, treatment by employer, and wage level. Ten attribute s would be too large for an experimental design in our context, so our next task was to reduce this list. Attributes, like treatment by employer, that could not be easily quantified were removed. This was because the primary interest of this study lay in e xplaining economic decisions made by workers; by may differ in their choices because of their different views of appropriate behavior of employers. Such variation wo uld potentially contribute to measurement error. To redefine the compensation packages towards understanding the dairy context, we conducted three focus groups which helped affirm a subset of benefits identified in the existing literature and identified benefits which are specific to the dairy industry. The focus led to selection of seven attributes. The next step was to define the levels for each of the attributes. The list of attributes and their levels is reported in Table 1.1 Table 1.1 Choice Exper iment Attributes and Levels Attributes Levels Wage 15% less than current wage, Current wage, 15% more than current wage Hours (average per week) 48/56/70/84 Health insurance Present/Absent Retirement Plan Present/Absent Housing Housing Allowance /On - site housing/None Meat bonus Present/Absent Quality Incentive Present/Absent 6 the convention of previous studies in industries where the wage rate was highly variable or unknown (Ubach, et al., 2003, Rockers, et al. 2013, Kunaviktikul, et al., 2015; Scott, et al., 2015; Song, et al. 2015, and Takemura, et al., 2016) . Four levels of average hours worked per week were selected, ranging from 48 to 84 hours. 3 , 4 The following four attributes (health insurance, retirement plan, meat bonus and milk quality incentive pay) were either included (present) or excluded (absent) from each job description, with a specific description of each provided to respondents. For example, the description of the retirement plan was developed based on the employer matching system that some dairy farmers described in focus groups . A detailed description of all other attribute levels is reported in Exhibit A 1.1 of the Appendix 5 . One o f the reasons health insurance was included in this dichotomous fashion is that it is analogous to welfare benefits ( Kunaviktikul, et al., 2015; Song et al., 2015) and free health care (Van den Broeck, et al., 2016) used in prior studies. A meat bonus is t he opportunity to purchase beef at reduced market prices. As workers may be economically disadvantaged or looking to save money for the future or family in Mexico, savings on quality meat products are attractive. Finally, the quality incentive is a lump - su m payment only received if the quality of milk meets certain standards. 6 Focus groups suggested 3 The closest attributes to hours worked foun d in the literature were change in the hours worked ( Ubach, et al., 2003; Scott, et al., 2015 ) and work schedule (Huicho et al., 2012; Miranda, et al., 2012). 4 The number of hours worked per week on average were all selected to be above 40 hours to avoid conflating the number of hours worked and (overtime) pay rate. Hour levels were chosen to represent feasible shifts, for example six eight - hour shifts or seven twelve - hour shifts . 5 The descriptions of the attribute levels and the accompanying table were p rovided to all respondents during the experiment. 6 The most similar attribute in extant literature is a performance based financial award (Rockers, et al., 2013). 7 that there is high variability in whether farms offer a quality incentive and in whether workers view it as a positive (chance to increase income) or negative ( causes income to vary and not entirely within the control of the individual). Finally, the last attribute was housing which had three levels. It was either excluded from the job description or if included was defined as on - site housing or a housing allowan ce. Employer provided housing was important in the focus groups but seemed to vary in quality. Therefore, we included employer provided on - site housing and a housing allowance which provides more flexibility to the worker in terms of accommodations for the housing levels. Similar housing attributes and levels are frequent in the rural medical literature (Huicho, et al., 2012; Miranda, et al., 2012; Rockers et al., 2012; Rockers et al., 2013; Kunaviktikul, et al., 2015; Efendi, et al. 2016; Smitz, et al., 20 16; Takemura, et al., 2016 ). Considering two profiles at a time, altogether these seven attributes and their levels resulted in (2 4 x 3 2 x 4 1 ) 2 = 331,776 possible choice tasks. To reduce fatigue effects, we followed Street and Burgess (2007) and performed a D - Optimal design in two steps, which resulted in 72 questions distributed across eight blocks. Workers were randomly assigned to only one of the eight blocks. In addition, the nine questions in each block were randomized to reduce order effects. An examp le choice task is provided in Figure 1.1. 8 1.2.2 Eliciting Risk and Time Preferences After the choice experiment, respondents completed two qualitative risk and time preference questions. To illustrate, respondents indicated on an 11 - point scale a self - assessment of their preferences in response to the following questions: and beneficial for you today in order to benefit m The risk question has been used in many prior studies, most notably since 2004 by the German Socio - Economic Panel Figure 1. 1 Example Choice Task Job 1 Job 2 Opt - Out Suppose there are two job openings offered by a dairy farm. Select the job you would prefer applying to. If you would not apply to either job 9 (SOEP). 7 Likewise, the time preference question has been utilized by other researchers (see for example Falk et al. 2016; 2018). A benefit to this question format is that it is quick and low - cost. Additionally, the questions involve no math unlike lottery - framed questions, which may make the question format easier to comprehend among individuals with low levels of e ducational attainment. Finally, the questions are not context - dependent and capture general behaviors. A nationally representative validation study suggests general question framing is the best fit to a variety of risk contexts ranging from financial (hold ing stocks and incentivized lottery) to career (being self - employed) (Dohmen et al., 2011). Additionally, generality allows categorization of respondents by preferences without having to assume a particular functional form for the utility function. Respons es from these questions were used to generate risk and time dummy variables. A respondent who answered 0 - 5 (unwilling) received a value of 0 and those above 5 (willing) a 1 as in Dohmen et al. (2011, p.328, footnote 10). Therefore, when the risk variable e quals one the respondent is risk - loving (willing to take risks) and when the time variable equals one the respondent is patient or future - oriented (willing to wait for something more beneficial in the future). 1.3 Data Dairy workers are a hard to reach p opulation. Many dairy workers are immigrants, some of whom may not speak English, others who may not be authorized to work in the United States, and others who are simply distrustful of outsiders. Furthermore, dairy laborers work long hours 7 Studies that use the SOEP risk preference data include Dohmen et al., 2011 and Caliendo e t al., 2009. Studies that conduct primary data collection using this qualitative risk question include Dohmen et al., 2011 and Falk, et al., 2016; 2018. 10 and can be diff icult to convince to participate in a survey. Therefore, study participants were recruited through community leaders and extension agents. Individuals were eligible to complete our survey if they currently worked, had worked, or were interested in working current work status is available in Table 1.2. These demographics are compared to similarly known populations in Table A1.1 and Box A 1.1 of the Appendix. In the remainder of this section we begin by describing the general demographics like gender and then proceed to describe work history and current employment statistics. Table 1.2 Sample Distribution by Work Status Average (Mean) t - Test for Differences Across Groups (Yes = Significant at 0.05 level) Full Sample Working on dairy farm (1) Working not on dairy farm (2) Not working (3) 1 & 2 1 & 3 2 &3 Female 44% 22% 58% 83% Yes Yes Yes Age 33 33 33 35 No No No Born in the U.S. 17% 13% 17% 28% No No No Prior Dairy Work 54% 75% 31% 39% Yes Yes No Years Employed by Employer at Time of Survey 3 3 2 N/A No N/A N/A N 109 55 36 18 A sizeable majority of those unemployed were women, while 58% of those working outside of dairy and 22% of those working in dairy were women. While there were considerable differences in the share of women in each employment category, the groups were similar in terms of age. Most individuals were born outside the United States (mostly in Mexico). The 11 greatest share of U.S. born respondents was among individuals who are currently unemployed, although this amount was not statistically significantly different from other employment groups. Approximately three - fourths of respondents working in dairy at the time of the survey had worked for a different dair y previously to their current job, while nearly one - third who were not employed in dairy at the time but were interested in such positions had prior dairy industry experience. Individuals currently working on a dairy had been in their position for approxim ately as long as other individuals employed but not working in dairy at the time of the survey. Figure 1. 2 Distribution of Reported Currently Hourly Wage by Employment in Dairy Sector The distribution of wages for dairy workers was similar to that for people employed in other sectors but interested in or with prior experience in dairy as seen in Figure 1.2 (t - test two - sided p - value of 0.6360). Dairy workers earned $12.10 on average, while non - dairy workers earned $11.70. Respond ents working in dairy at the time of the survey had a broader distribution of weekly average hours than non - dairy workers, centered at high levels. The distribution o f 12 weekly hours for dairy workers was different to that for people employed in other sector s but interested in or with prior experience in dairy (t - test two - sided p - value of 0. 9553 ) as seen in Figure 1. 3 . Approximately 67% of workers not in dairy worked 40 hours or less per week on average, while only about 25% of dairy workers worked 40 hours o r less per week on average. The average weekly hours for those who were not working in dairy were 40 and for those who worked in dairy worked an average of 53 hours per week. In general, reported hours were consistent with reported hours from other dairy w orker studies in Idaho (Salant et al., 2017, p.22) and New York (Maloney, et al., 2016, p.26 - 8). Figure 1. 3 Distribution of Reported Average Weekly Hours Worked by Employment in Dairy Sector The greater hours worked by those in the dairy sector pays off as the income distribution is statistically greater (two - sided t - test p - value of 0.0054) than those not in the dairy sector. Calculated annual income, ignoring possible overtime pay, ranges from $4,784 - $104,000 with a mean of $32, 210 and median of $31,200 in the dairy industry. In contrast, annual income among respondents not in the dairy industry ranged from $4,776 to $54,080 with a mean of $24,782 and 13 median of $21,944. Using desired hours instead of actual hours worked, dairy wo rkers still report higher desired incomes. 8 However, the maximum desired income declines to $84,500. In addition to wage earnings, employees can receive a myriad of non - wage benefits. Table 1.3 depicts the distribution of benefits relevant to the DCE that were received by those in dairy relative to non - dairy workers. Most notable is that dairy workers receive on - site housing at much higher rates than other workers. Despite comprising a greater portion of our sample and working greater hours on average per w eek, fewer respondents working in dairy report earning overtime compared to those who are not in dairy, meaning that dairy workers may not be compensated extra for their additional hours worked compared to other sectors. Table 1.3 Percent of Respon dents Earning Benefits by Employment in Dairy Sector Benefit Percent Who Are Employed but Not Working in Dairy at Time of Survey Percent Who Are Working in Dairy at Time of Survey Overtime Pay 38.9% 20.0% Health Insurance 22.2% 16.4% Retirement F unds, Self - Paid 8.3% 3.6% On - Site Housing 21.2% 68.5% Housing Allowance 0.0% 5.6% Food 11.1% 7.2% Bonus 16.7% 23.6% Incentive Pay 8.3% 18.2% 1.4 Estimation Methods Discrete choice models are consistent with random utility theory. Accordingly, we a ssume that workers have individual employment compensation preferences and select compensation 8 We have data on the number of desired average work hours per week which we use for this calculation. We do not explicitly ask what the desired wage level is of respondents. 14 packages that maximize their utility. U tility is comprised of an observable, predictable component, , and some unexplained random component, , as represented in Equation 1. (1) where is the utility worker n derives from alternative j in choice question t. In this application, the deterministic portion of the utility function is expressed as (2) where is the vector of structural taste parameters, the multi - vector of observed variables, and is the independently and identically Gumbel distributed unobservable utility (err or term). Depending on the assumptions underlying the functional form of equation (2) and error term in equation (1), different discrete choice models can be estimated. In this application, a mixed logit model (MXL) for panel data is used to estimate the relevant parameters as it accounts for random taste variation and correlation in unobserved factors over time (Tra in, 2003). As shown in Train (2003), the probability that individual n makes a sequence of observed choices i , one for each choice task in th e assigned sequence of T choice tasks, i = ( i 1 ,..., i T ), , is represented by the following joint probability: (3) Because equation (3) lacks a closed form solution, the paramet ers of the model are estimated by simulated maximum likelihood estimation techniqu es following Train (2003). Halton draws (500) are used for the simulation rather than random draws as the former provide more efficient simulation (Bhat 2003). 15 Using equatio n (3), four models are specified. Model 1, which represents the baseline model, accounts for random taste variation. Hence, the observable portion of our underlying utility function for individual n from alternative j in choice task t is specified as follo ws: (4) where OptOut is the alternative specific constant indicating the no - application option, Wage is a continuous variable indicating the hourly wage, 9 Hours is a continuous variable representing the average number of hours worked per w eek . The remaining variables are dichotomous variables representing the presence/absence of a health insurance plan ( Health), retirement plan ( Retire), on - site housing ( OSHousing), housing allowance ( HousingA), meat bonus ( Meat), and milk quality incentive ( Quality). is the coefficient of wage, while are the coefficients representing the non - monetary attributes. The absence of health plan, retirement plan, housing accommodation, meat bonus and milk quality incentives represent the baselines. The other three mo del specifications incorporate systematic differences across workers, i.e. risk and time preferences. To illustrate, Model 2 adds to Model 1 by including two interactions between health insurance and quality incentive with risk, r . Model 3 adds to Model 1 by incorporating the interaction between retirement plan and time preferences, p . Model 4 combined models 2 and 3 by including interaction terms for both risk and time preferences, as follows: 9 if currently unemployed and with a prior job, or the federal minimum wage $7.25 if currently unemployed and never previously holding a job, by (1+ chwage), where chwage, is the change in wage respondents saw in the choice experiment question, either - 15%, 0%, or 15%. 16 (5) w here are the coefficients capturing the interaction s between the variables , , and and the risk ( and time ( preferences variables. The risk preference variable is equal to 1 if risk loving and zero otherwise, while the time preference variabl e is equal to 1 if patie nt (future oriented) and zero otherwise. All the other variables are specified as in (4). Here we remind the reader of important information about the interpretation of results (presented in Table 1.4) from equations (4) and (5) . For instance , the estimat ed population means represent the marginal utilities workers derived from the presence of the various attributes included in the experimental design (e.g., wage, hours, retirement plan, health insurance, o n - site h ousing , h ousing a llowance , m eat b onus , and milk quality incentive ). If they are positive and statistically significant, then the ir presence would increase negative and statistically significant then . For examp le, we would expect estimated population means of all our attributes, except perhaps hours , to be positive as receiving more compensation, whether in the form of wages or benefits, is generally preferred . For example, workers would prefer to have a higher wage as well to have a retirement plan as compared to not having it which is the baseline. We do not make a prediction about hours. Maloney, et al. (2016) mention that employers have moved from eight to twelve hour shifts at the request of workers for more hours, increasing salaries. However, since all hour measurements used in our study are above the standard 40 - hour work week to avoid conflation with whether overtime is or is not offered, it is possible that respondents will view more work hours as a nega tive impact on their leisure time. 17 As marginal utilities, the coefficients from the MXL are not directly meaningful. Rather, in addition to whether they positively or negatively contribute to utility, the mean coefficient estimates are useful for generati ng a cardinal rank of preferences based on estimate magnitude. However, even this ranking is limited to variables that are coded in a similar fashion and thus are on the same scale. Therefore, in our study we can create a cardinal rank of health insurance, retirement plan, on - site housing, housing allowance, meat bonus, and milk qualitive incentive as these variables are all dichotomous. We cannot include wages and hours in this rank as these two variables enter utility in a continuous fashion. Therefore, we compare the experimentally designed job attributes by looking at the marginal rate of substitution, calculated as the ratio of mean estimates of two attributes. The natural choice of common denominator is the estimated mean coefficient of the wage which then allows for us to interpret our results as the amount of money a worker is willing to pay and thus give up (if positive) or must receive (if negative) to have the numerator attribute. In other words, a positive sign of the coefficient ratio, with the wage coefficient in the denominator, indicates how much dairy workers would be willing to pay per hour to have more of the non - monetary job attribute, and a negative sign indicates how much hourly income dairy workers would be willing to accept to have mor e of the attribute. We report these calculations in Table 1.5. 1.5 Empirical Results The estimates from the MXL models are reported in Table 1.4. We remind the reader that Model 1 accounts for preference heterogeneity by allowing for random taste variati on. Models 2 through 4 add to Model 1 by also exploring differences between groups of people with different risk and time preferences. All variables were assumed to have a normal distribution except for 18 wage, which is assumed to follow a one - side triangula r distribution and interaction terms which were assumed to be nonrandom. Table 1.4 Estimates from the Mixed Logit Model for the Full Sample Variables Parameters Estimates Model 1 Model 2 Model 3 Model 4 Wage 1 0.322*** (0.047) A 0.326*** (0.049) 0.3 30*** (0.049) 0.327*** (0.048) 2 0.322*** (0.047) 0.326*** (0.049) 0.330*** (0.049) 0.327*** (0.048) Hours 2 - 0.050*** (0.007) - 0.049*** (0.008) - 0.051*** (0.008) - 0.051*** (0.008) 2 0.054*** (0.008) 0.061*** (0.007) 0.054*** (0.009) 0.055** * (0.008) Health Insurance 3 0.633*** (0.145) 0.727*** (0.242) 0.650*** (0.147) 0.697*** (0.238) 3 0.588*** (0.218) 0.472 (0.314) 0.595*** (0.216) 0.589*** (0.215) Retirement Plan 4 1. 453*** (0.249) 1.442*** (0.250) 0.986** (0.349) 0. 997*** (0.3347) 4 1.938*** (0.270) 1.929*** (0.286) 1.906*** (0.266) 1.908*** (0.265) On - site Housing 5 0.438** (0.172) 0.453*** (0.170) 0.466*** (0.176) 0.446*** (0.176) 5 0.332 (0.314) 0.218 (0.283) 0.393 (0.288) 0.392 (0.287) Hous ing Allowance 6 0.429** (0.173) 0.428** (0.177) 0.435** (0.174) 0.435** (0.174) 6 0.127 (0.466) 0.436 (0.266) 0.045 (0.517) 0.054 (0.483) 19 A Numbers in parentheses are standard errors. In Model 1 , all coefficients of the population means are statistically significant, except for the opt - out option and meat bonus. The la t ter, however, has a statistically significant random parameter (standard deviation) implying that many workers are insensitive to the meat bonus. The sign and magnitude of the mean coefficients indicate the directional impact and relative Table 1.4 (cont d) Meat Bonus 7 0.126 (0. 143) 0.142 (0.139) 0.123 (0.144) 0.121 (0.144) 7 0.563** (0.143) 0.552** (0.263) 0.569** (0.221) 0.573** (0.222) Milk Quality Incentive 8 0.527*** (0.157) 0.363 (0.258) 0.521*** (0.156) 0.366 (0.251) 8 0.858*** (0.239) 0.804*** (0.199) 0.86 4*** (0.232) 0.857*** (0.225) Opt - Out Option - 0.298 (0.605) - 0.331 (0.631) - 0.255 (0.626) - 0.268 (0.616) Interaction terms Health*Risk 9 - 0.121 (0.288) - 0.069 (0.290) Incentive*Risk 10 0.241 (0.313) 0.238 (0.308) Re tirement*Time 11 0.811* (0.444) 0.796* (0.442) Model Statistics LLF - 657.32 - 658.73 - 655.66 - 655.34 # of Parameters 16 18 17 19 # of Choices 981 981 981 981 BIC 1425 1441 1428 1442 AIC 1347 1353 1345 1349 AIC3 1363 1371 1362 1368 c rAIC 1357 1368 1357 1365 20 importance to utility. As expected, the wage coefficient is pos itive, indicating that people prefer coefficient workers prefer fewer hours per week. This result confirms findings from previous studies (Nolte and Fonseca 2010), al though counters the common notion in the industry that workers insist on working long hours. Moreover, health insurance, retirement plan, on - site housing, housing allowance, and milk quality incentive pay are all positive. Looking at the magnitude of their coefficients, the greatest preference is for a retirement plan. In contrast, hous ing attribute levels were the least preferred among the statistically significant results with on - site housing being marginally preferred to a housing allowance. Gabbard and Perloff (1997) also found that housing is ranked low compared to other benefits. All estimates of the standard deviations are statistically significant. This indicates that there are differences in preferences among sample respondents for hours, health i nsurance, retirement plan, and milk quality incentive with the greatest variability for retirement plans. This means that although some workers strongly prefer a retirement plan and are made much happier with its presence, for some workers a retirement pla n brings little or negative utility. The statistical significance of the standard deviations also indicates that hours, health insurance, retirement plan, and milk quality incentive structures will not be deemed equally appealing by all workers. Rather, on average, workers find them desirable, while a minority do not. The last three columns of table 1.4 report the estimates from Models 2 through 4, which build on Model 1 by including various risk and time preference interaction terms with the choice experi ment designed attributes. In all models (2 - 4), the interaction term between risk preferences and health insurance and milk quality are not statistically significant. This means that being risk averse versus risk loving has no differential impact on prefere nces for health insurance or milk 21 quality incentive pay. In contrast, the interaction term between being offered a retirement plan and time preference s is positive and statistically significant in Model 3 and when risk preferences are also introduced (Model 4). I t makes intuitive sense that people who are future oriented are more likely to prefer a retirement plan than respondents who are present - minded. According to the model fit criteria (Akaike Information Criteria (AIC) and 3 - Akaike Information Cr iteria (3AIC), and corrected Akaike Information Criteria (crAIC)) Model 3 also resulted in the best model fit. Most notably, Model 3 produces similar results to Model 1. To illustrate, all attribute signs and statistical significance remain unchanged; the coefficient of fewer hours per week. The relative preference rank is also the same across models 1 and 3; retirement plan is the most preferred attribute, fo llowed by health insurance, milk quality incentive, on - site housing, and housing allowance. We now take the results from Model 3 to calculate willingness to pay for each attribute. 10 The willingness to pay amounts and their 95% confidence intervals are re ported in Table 1.5. In addition to the hourly wage equivalent that is produced directly from our original analysis (reported in column 2) we convert the hourly wage to annual income amounts which are reported in column 3. 11 A positive value represents the amount a worker is willing to forego from their wage to obtain the attribute, while a negative value represents the additional amount that the worker would need to be compensated, or paid, to take the attribute. As anticipated by the prior 10 Marginal WTP for each attribute w as calculated as the r atio between the coefficient of each non - monetary attribute (Retirement Plan, Health Insurance, Milk Quality Incentive, On - Site Housing, Housing Allowance, and Hours) and the coefficient of the monetary attribute (wage). 11 Annual income amounts are based o n the willingness to pay from the hourly wage times the mean hours worked per week (47) times the number of weeks per year (52). 22 analysis and pri or expectations, all values are positive except for hours. This means that workers look favorably on and are willing to pay for all benefits except for higher hours. Table 1.5 Marginal WTP Estimates (Based on Full Sample (Table 1. 4 , Model 3) ) Margina l WTP Estimates (as Hourly Wage ) Marginal WTP Estimates (as Annual Salary ) Retirement Plan $2.99*** A (1.16) B [$0.72, $5.27] C $7,307.56 [$1,759.68, $12,897.88] Health Insurance $1.97*** (0.569) [$0.86, $3.09] $4,814.68 [$2,101.84, $7,551.96] Milk Quali ty Incentive $1.58*** (0.54) [$0.51, $2.66] $3,861.52 [$1,246.44, $6,476.60] On - Site Housing $1.41** (0.63) [$0.18, $2.65] $3,446.04 [$439.92, $6,476.60] Housing Allowance $1.32** (0.570) [$0.20, $2.44] $3,226.08 [$488.80, $5,963.36] Hours - 0.16*** ( 0.549) [ - $0.23, - $0.09] - $391.04 [ - $562.12, - 219.96] A The reported wage is for present - oriented individuals. B Numbers in parentheses are standard errors calculated using the Krinsky and Robb method in NLogit6 ; C 95% confidence interval in brackets , cal culated using the Krinsky and Robb method in NLogit6 . More specifically, results indicate that farmers should prioritize retirement plans as part of employee compensation plans, as workers are willingness to pay $2.99 per hour. Also, workers are willing to forego approximately 25% of their annual salary at the time of the survey for the presence of a retirement plan. The fact that workers are willing to forego a considerable amount of their wage is not necessarily so surprising. First, we are comparing th e presence of a retirement plan to no t having any retirement plan. As seen in Table 1.3 , few respondents currently have a retirement plan. Therefore, according to the principle of diminishing marginal 23 utility, respondents may be willing to pay greater amou nts for the initial presence of this attribute Retirement plans may also be attractive because many respondents either do not have access to banks and/or do not frequently save money. If the former, retirement plans represent access to compounded earnings that respondents do not otherwise have access to. Given the average age of our sample and assuming another 30 years of work, the future value of the amount that the employer would need to contribute today is greater than the annual salary respondents are willing to give up today. If the latter, then respondents may be willing to give up a portion of current salary to be forced to save for the future as consistent with the limited commitment savings literature (Gugerty , 2007) . While this survey was not administered to dairy owners to elicit their preferences for compensation structures, retirement plans may also be appealing from the employer perspective. To illust rate, i f the average employee contributes the maximum possible that an employer will match, then a retirement plan will cost the employer $1,792.53. 12 The annual income that workers are willing to forego for a retirement plan is nearly four times higher tha n the average amount the employer would contribute. 13 contribution they would need to reduce wages by only $0.74. 14 Thus, at no cost to the employer and a six percent reduction in average wage to the employee, workers will be appreciably happier with their employment compensation plan when it includes a retirement plan. 12 This is the average salary ($29,875.57) times the employer contribution of 6%. 13 Calculation: Marginal WTP for retirement pla n as annual salary divided by the cost of retirement plans to employers ( 7,307.56/ 1,792.53 ) . 14 Calculation: The annual cost of retirement plans to employers divided by the average number of hours worked by employees each year ( 1,792.53 / (47 * 52) ) 24 amount of wages one is willing to forego ) for a health insurance plan given recent national debates about health insurance and its provision by employers. Our results indicate that health insurance is the second most important benefit that could be offered from those included in this study. The average worker would be willing to forego $4,814.68 annually in wages for health insurance. However, while this amount is near the required deductible of $6000 as outlined in benefit description (see Exhibit A 1.1 in the Appendix), the majority are not willing to pay the full amount required. Nevertheless, it may be comforting that the deductible value is indeed within the 95% confidence interval. Altogether this suggests that workers will be happier if offered a health insurance plan by their employer but may still opt out from participating as they deem th e minimum deductible too high. The milk quality results are surprising given that risk aversion did not seem to impact preferences regarding the milk quality incentive. The average respondent is willing to forego twice the potential earnings from the milk quality incentive, even with the conservative estimate of the lowest bound of the 95% confidence interval, $1,246.44 compared to the maximum $600 possible earned a year. Perhaps workers do not consider their annual income and erroneously calculate the val ue of a $600 annual increase. Yet, recent discussions around good motivation strategies have encouraged worker choice, competency, and agency (Milligan 2020). It is also possible that a drive for these intangible factors may be factoring into the high will ingness to pay for the milk quality incentive. The milk quality incentive may be highly ranked because it represents an area over which workers have control of their compensation. Dairy work can become monotonous and having autonomy over an outcome of the job may help with motivation. Plus, workers may feel a sense of fairness from this benefit which helps disperse the premium operators receive for Grade A milk to the workers who help produce its quality. 25 As our results indicate disutility from additional h ours worked , employers would need to pay annually $391 per person per additional hour worked to make it worthwhile for the average surveyed worker to work the 4 8 th hour. This amounts to only sixteen cents per hour more, but in aggregate suggests meaningful costs. Therefore, employers should be cautious to not assign shifts over 47 hours per week without offering overtime or a higher wage rate than currently offered for risk of dissatisfying employees to look for other employment. Note that this recommendati on does not require employers to pay more than they do currently, as instead they could distribute shifts over a greater number of workers thus avoiding the concept of overtime. To further explore potential differences (heterogeneity) in job attributes eva luation among workers , we performed a subsample analysis across groups of workers. More specifically, demographic group of interest , foreign - born individuals and only respondents with (either current or past) experience working in dairy. We use the Model 3 specification since it was a better fit for our data. Demographics of these subsamples and the estimated coefficients are reported in the Appendix in Tables A1. 2 and A1.3 , respectively. In Table 1. 6 , we report the marginal WTPs for the various attributes, which allow for an easy comparison across groups while also solving issues related potential differences in scales among groups. 26 Table 1. 6 Marginal WTP Estimates as Hourly Wages of Sub - samples (Based on Table A1.3 ) Foreign - Born Individuals Respondents with Dairy Experience Retirement Plan $2.41* A ( 1.411 ) B [ $ - 0.36, $5.18 ] C $2.87* (1.627) [ - $0.32, $6.06] Health Insurance 2.19*** ( 0.650 ) [ $0.92, $3. 47 ] $1.81*** (0.661) [$0.51, $3.11] Milk Quality Incentive $1.71** ( 0.691 ) [ $0.35, $3.06 ] $2.24** (0.942) [$0.40, $4.09] On - Site Housing $1.39** ( 0.685 ) [ $0.05, $2.74 ] $1.62** (0.777) [$0.10, $3.15] Housing Allowance $1.27** ( 0.648 ) [ $0.00, $2.54 ] 1.36* (0.789) [ - $0.19, $2.91] Hours - $0.12*** ( 0.033 ) [ $ - 0.19, $ - 0.06 ] - $0.15*** (0.045) [ - $0.24, - $0.06] A The reported wage is for present - oriented individuals. B Numbers in parentheses are standard errors calculated using the Krinsky and Robb (1986) metho d in NLogit6 ; C 95% confidence interval in brackets , calculated using the Krinsky and Robb method in NLogit6 . Just as there is general consistency across model specifications, results are also robust to different subgroup analysis. For example, for all s ubgroups the sign, statistical significance, and relative ranking among significant results of wages and hours remains constant. The sign and general statistical significance of the coefficients of the foreign - born and prior dairy experience subgroups also are the same. What is perhaps most interesting from this analysis is that respondents with prior dairy experience were willing to pay more for the milk quality incentive pay than they were the retirement plan. This could be because individuals with more p rior dairy experience trust their ability to earn the incentive pay. This suggests that dairy operators who know that their workers have prior dairy experience (or considerable experience within their own company) may be highly motivated by a milk quality incentive structure. The fact that health 27 insurance remains statistically significant and highly ranked among both sub - samples suggests that a broad proportion of individuals in the dairy labor market are interested in health insurance. The former may be a result of the current national debate around health insurance and its importance in the U.S. Additionally, dairy work is highly labor intensive. While individuals may not sustain critical injuries subject to workers compensation, they may suffer from over use or other ergonomic stressors that induce them to seek pain management or other medical assistance for which they would need medical insurance to mitigate costs. Overall, as our results seem to be robust among a variety of sub - sample populations, they p rovide insight on benefits that will appeal to many types of dairy workers. 1.6 Conclusion In an environment of short supply and high turnover of dairy workers it is important to identify compensation packages that will help recruit and retain workers. We used a discrete choice experiment influenced by conversations with those in the industry and existing literature on rural compensation packages. Results from this stud y will provide new insights to dairy operators who may be struggling with labor shortages. For instance, by identifying worker preferences for job benefits, employers may be better able to attract and retain workers through the benefits they would offer. I n other words, farmers will have a better understanding of how to structure work dairy workers, they and advocacy groups, like Migrant Justice, will be better s ituated to negotiate for the most important working conditions. It is essential that both employers and negotiating employees identify potential factors that motivate and determine compensation preferences, to be 28 able to successfully match packages to empl oyees. This study assists in this process by exploring whether risk and time preferences impact compensation plan preferences. Additionally, preferences are explored across different sub - group sample populations. Finally, descriptions of current dairy work ers and their working conditions are provided which fills an information vacuum on current norms within the dairy industry which had previously been addressed only in New York (Maloney et al. 2016) and Idaho (Salant et al. 2017). Our results generally sugg est that dairy workers prefer compensation and working conditions similar to that of competing industries. A retirement plan is preferred over health insurance, housing, meat bonus, and an incentive pay system. A retirement plan provides a commitment devic e for workers to save for the long - term. Such a plan may also be particularly attractive to those who do not have a bank account with compounded interest . Additionally, in contrast to the notion that dairy workers desire as many hours as possible (Maloney, et al. 2016) , workers prefer to work fewer hours per week. This could be because in this experiment respondents chose between work week hours that were all both above their average current hourly work week and above the traditional 40 - hour work week used as a baseline for many overtime and benefit decisions. It could be that workers prefer more hours up to either 40 hours or their current weekly schedule (47 hours) but would be less happy with more hours beyond that point. It is worth further exploration w hether this is the case and if so, where the maximum utility is generated relative to the number of hours worked. Another area for continued research is exploring the observed preference heterogeneity (statistically significant standard deviation estimates in Tables 1.4 and A1.3 ) among dairy workers. Doing so would require a large (national) sample than available here. Additionally, since our results indicate that future - oriented workers are willing to forgo higher wages for a 29 retirement plan than present - o riented individuals it could be worth pondering how to identify individuals in each group and offer the most attractive plan to them. Finally, while not necessarily an area for further research, it is interesting to repeat in the discussion of heterogeneou s preferences, that workers with prior dairy experience change their relative rankings of health insurance plans and milk quality incentive pay structures. Respondents with dairy experience are willing to pay more of their current wage for the incentive pl an structure. This could be because the incentive pay structure gives them a sense of agency and control over their work, which they may not feel that they have in many other aspects of the job and which have been linked to higher levels of motivation. Th at workers who have prior dairy experience are willing to gamble with their annual income earnings via a milk quality incentive pay structure than those without dairy experience is not that surprising; however, many of our other results are surprising as t hey are largely the inverse of the frequency with which workers currently receive such benefits. As compensation plans are negotiated between workers and employers, the ultimate bundle of attributes offered will also be influenced by employer preferences a nd financial capacity. Therefore, compensation plans in the dairy industry may need to be restructured. For instance, employers should consider offering a retirement plan to their workers. This may have a particularly strong effect on recruiting more worke rs into the industry thereby addressing their labor shortage concerns. Another important implication of our study is that once operators are able to attract more workers they may want to reduce shift length and/or the number of shifts so that workers work fewer hours per week. Our results are also informative for policy makers as they affirm the addition of overtime pay of agricultural workers, an issue recently debated in the New York legislature. 30 The generalizability of these results is subjected to some limitations. For example, because of the difference in reported preferences and benefits currently accepted in the industry and the high levels of willingly foregone wages there is some need for caution in interpreting results. Despite attempts to make th e choice experiment clear to respondents by offering clear attribute descriptions and pictorial representations of attributes, it is possible that respondents overestimated their willingness to for e go wages in exchange for other benefits. Despite this, our results can provide direction to dairy farmers looking for potential solutions to their labor supply challenges. 31 APPENDIX 32 APPENDIX Exhibit A1.1 Handouts to Respondents Describing Choice Experiment Attributes 1. The health insurance plan provides for you and any immediate qualifying dependents. The premium is paid by the employer but you are responsible for the deductible, approximately $6000, with a maximum out of pocket expense of $7,250. Ther e is a 40% coinsurance for emergency room care after deductible and $40 primary doctor visit and $40 generic drug copays. 2. The employer offers voluntary enrollment into a 401K retirement plan. If you choose to participate, the employer will match 100% of t he first 6% of your salary. For example, if you make $25,000 then the employer will add the same amount that you do up to $1,500. 3. When the employer offers on - site housing, they are providing you a furnished bedroom (bed, small dresser) of your own in a ho use shared with up to 5 other workers of the same gender. Housing is available only to the worker not his/her dependents. If you quit or are fired you must move out that day. When the employer offers a housing allowance, they provide $500 towards rent/mort gage payments. You receive the payment at the end of each month, provide the employer a copy of the lease/mortgage as proof of stable housing to qualify for the allowance . The allowance is fixed, meaning it does not depend on characteristics of the housing arrangement such as the location of the shelter or the number of people residing there. 33 Exhibit A1.1 (cont 4. The meat bonus is the opportunity to purchase freshly slaughtered meat at reduce d costs. You pay the processing fee of $0.40 per pound and can request a cow be slaughtered when you per year, approximately 600lb raw. 5. Finally, the milk quality i ncentive is an additional amount of money added to payroll based on the milk quality and herd health over the prior month. Full - time milkers will receive $50 (half - time will receive $25) per month if the somatic cell count is under 150,000 for that month. Thus, a full - time employee could earn an additional $600 per year if farm performance hits quality and health standards every month. 34 Table A1.1 Sample Distribution Comparisons Average (Mean) t - Test for Differences Across Groups (Yes = Significant at 0.05 level) A This Study NAWS 2016 B Maloney, et al. (2016) Full Sample (0) Working on dairy farm (1) Working not on dairy farm (2) Not working (3) NAWS & 0 NAWS & 1 NAWS & 2 NAWS & 3 Female 23% 4% 44% 22% 58% 83% Yes No Yes Yes Age 40 31 33 33 33 35 Yes Yes Yes No Born in the U.S. 23% 0% 17% 13% 17% 28% No Yes No No Prior Dairy Work N/A N/A 54% 75% 31% 39% N/A N/A N/A N/A Years Employed by Employer at Time of Survey 0 N/A 3 3 2 N/A Yes Yes Yes N/A N 2508 205 109 55 36 18 A NY = Maloney (2016) sample B From U.S. Department of Labor (n.d.) 35 Table A1.1 (cont d) . t - Test for Differences Across Groups (Yes = Significant at 0.05 level) A NY & 0 NY & 1 NY & 2 NY & 3 Female Yes Yes Yes Yes Age Yes No No No Born in the U.S. Yes Yes Yes Yes Prior Dairy Work N/A N/A N/A N/A Years Employed by Employer at Time of Survey N/A N/A N/A N/A N A NY= Maloney (2016) sample B From U.S. Department of Labor (n.d.) 36 Box A 1.1 Discussion of Sample Comparis ons to Other Studies There are a number of differences between our sample and comparable sample populations. However, there is also quite a range in the percent female, age, and foreign - born between our two comparison samples. Often our full sample and dai ry workers at the time of the survey lie between the national agricultural workers survey (NAWS) and Maloney, et al. (2016) samples. Despite the statistical differences observed via t - tests, we maintain that our sample appears comparable to other samples o f agricultural workers. Dairy workers are a hard to reach population, contributing to the scarce amount of research about them. As such we do not know of a national or Michigan specific sample of dairy workers to compare with our non - random sample. As a re sult, we need to look at the information from multiple sources available. NAWS is a national survey of agricultural workers but is limited to crops and therefore is not inclusive of dairy workers. Maloney, et al. (2016) concentrated on Hispanic dairy worke rs in New York. There may be differences between Michigan and New York dairy labor markets. Additionally, our sample includes non - foreign - born individuals who may have different demographics than the foreign - born population. The NAWS and Maloney, et al. (2 016) samples create a wide range for demographic variables, within which our full sample and dairy workers at the time of the survey tend to fall. Thus while we again acknowledge that ours was not a random sample, similar to Maloney, et al (2016) and Salan t, et al. (2017), we have reason to believe it is representative of our study population of interest. 37 Table A1.2 Sub - sample Distribution Comparisons Average (Mean) t - Test for Differences Across Groups (Yes = Significant at 0.05 level) A Foreign - Born S ample Dairy Experience Sample NAWS 2016 C Maloney, et al. (2016) NAWS & Foreign - Born NY & Foreign - Born NAWS & Dairy Experience NY & Dairy Experience Female 37% 32% 23% 4% Yes Yes No Yes Age 34 32 40 31 Yes Yes Yes No Born in the U.S. 0% 15% 23% 0% N/A N/A No Yes Prior Dairy Work 71% 81% N/A N/A N/A N/A N/A N/A Years Employed by Employer at Time of Survey 3 3 0 N/A Yes N/A Yes N/A N 112 91 2508 205 A NY= Maloney (2016) sample B From U.S. Department of Labor (n.d.) 38 Table A1.3 Estimates from the Mixed Logit Model of Sample Sub - populations Only Foreign - Born Respondents Only Respondents with Dairy Experience Variables Parameters Model 3 Model 3 Wage 1 0.322*** ( 0.055 ) A 0.317*** (0.065) Hours 2 - 0.039*** (0.008) - 0.047*** (0.008) 2 0.047*** (0.008) 0.032*** (0.006) Health Insurance 3 0.707*** (0.164) 0.573*** (0.170) 3 0.572** (0.284) 0.510 (0.338) Retirement Plan 4 0.77 7* (0.417) 0.910** (0.409) 4 2.182*** (0.363) 1.764*** (0.338) On - site Housing 5 0.449** (0.196) 0.514** (0.203) 5 0.312 (0.269) 0.347 (0.342) Housing Allowance 6 0.410** (0.196) 0.431** (0.202) 6 0.399 (0.312) 0.041 (0.641) Meat Bonus 7 0.154 (0.158) 0.183 (0.166) 7 0.626** (0.298) 0.582** (0.260) 39 Table A1.3 (cont d ) Milk Quality Incentive 8 0.550*** (0.188) 0.710*** (0.193) 8 1.055*** (0.254) 0.894** * (0.237) No - Application 0.419 (0.720) 0.119 (0.769) Interaction term Retirement*Time 11 1.187** (0.558) 0.748 (0.547) Model Statistics LLF - 546.08 - 466.18 # of Parameters 17 17 # of Choices 819 657 BIC 1206 1043 AIC 11 26 966 3AIC 1143 983 crAIC 1141 985 a Numbers in parentheses are standard errors. 40 REFERENCES 41 REFERENCES Bergtold, J.S., Shanoyan, A., Fewell, J.E., Williams, J.R., 2017. Annual bioenergy crops for preferences for producing sweet sorghum. Energy 119, 724 731. Bhat, C.R., 2003. Simulation estimation of mixed discrete choice models using randomized and scrambled Halton sequences. Transportation Research Part B: Methodological 37, 837 855. Broch, S.W., Vedel, S.E., 2012. Using Choice Experiments to Investigate the Policy Relevance of Heterogeneity in Farmer Agri - Environmental Contract Preferences. Environ Resource Econ 51, 561 581. https://doi.org/10.1007/s10640 - 011 - 9512 - 8 Caliendo, M., Fossen, F.M., Kr itikos, A.S., 2009. Risk attitudes of nascent entrepreneurs new evidence from an experimentally validated survey. Small Bus Econ 32, 153 167. https://doi.org/10.1007/s11187 - 007 - 9078 - 6 Charlton, D., Taylor, J.E., 2016. A Declining Farm Workforce: Analysis o f Panel Data from Rural Mexico. American Journal of Agricultural Economics 98, 1158 1180. https://doi.org/10.1093/ajae/aaw018 Dohmen, T., Huffman, D., Schupp, J., Falk, A., Sunde, U., Wagner, G.G., 2011. Individual Risk Attitudes: Measurement, Determinants , and Behavioral Consequences. Journal of the European Economic Association 9, 522 550. Efendi, F., Chen, C. - M., Nursalam, N., Andriyani, N.W.F., Kurniati, A., Nancarrow, S.A., 2016. How to attract health students to remote areas in Indonesia: a discrete c hoice experiment. Int J Health Plann Mgmt 31, 430 445. https://doi.org/10.1002/hpm.2289 Falk, A., Becker, A., Dohmen, T., Enke, B., Huffman, D., Sunde, U., 2018. Global Evidence on Economic Preferences. Q J Econ 133, 1645 1692. https://doi.org/10.1093/qje/ qjy013 Falk, A., Becker, A., Dohmen, T., Huffman, D., Sunde, U., 2016. The Preference Survey Module: A Validated Instrument for Measuring Risk, Time, and Social Preferences. IZA Discussion Paper No. 9674. Fatka, J., 2019. Ag labor shortage fix offered for dairy industry [WWW Document]. Feedstuffs. URL https://www.feedstuffs.com/news/ag - labor - shortage - fix - offered - dairy - industry (accessed 6.24.19). choices: Exper imental evidence from the Ghanaian pineapple sector. Food Policy 81, 67 81. https://doi.org/10.1016/j.foodpol.2018.10.005 42 Gabbard, S.M., Perloff, J.M., 1997. The Effects of Pay and Work Conditions on Farmworker Retention. Industrial Relations 36, 474 488. https://doi.org/10.1111/0019 - 8676.00064 Associations in Kenya. Economic Development and Cultural Change 55, 251 - 282. Hensher, D.A., Rose, J.M., Greene, W.H., 2015. Applied Choice Analysis, 2nd ed. Cambridge University Press. Hensher, D.A., Rose, J.M., Greene, W.H., 2005. Applied Choice Analysis: A Primer. Cambridge University Press. Hertz, T., Zahniser, S., 2013. Is There a Farm Labor Shortage? American Journal of Agricultu ral Economics 95, 476 481. Huicho, L., Miranda, J.J., Diez - Canseco, F., Lema, C., Lescano, A.G., Lagarde, M., Blaauw, D., 2012. Job Preferences of Nurses and Midwives for Taking Up a Rural Job in Peru: A Discrete Choice Experiment. PLOS ONE 7, e50315. http s://doi.org/10.1371/journal.pone.0050315 Jans, N., Frazer Jans, J., Louviere, J., 2001. Employee choice modelling: Predicting employee behaviour under varied employment conditions. Asia Pacific Journal of Human Resources 39, 59 81. https://doi.org/10.1177/103841110103900305 Khanna, M., Louviere, J., Yang, X., 20 17. Motivations to grow energy crops: the role of crop and contract attributes. Agricultural Economics 48, 263 277. https://doi.org/10.1111/agec.12332 Krah, K., Petrolia, D.R., Williams, A., Coble, K.H., Harri, A., Rejesus, R.M., 2018. Producer Preferences for Contracts on a Risky Bioenergy Crop. Appl Econ Perspect Policy 40, 240 258. https://doi.org/10.1093/aepp/ppx034 Krinksy, I., Robb, A. 1986. One Approximating the Statistical Properties of Elasticities. The Review of Economics and Statistics 68, 715 - 71 9. Kunaviktikul, W., Chitpakdee, B., Srisuphan, W., Bossert, T., 2015. Preferred choice of work setting among nurses in Thailand: A discrete choice experiment. Nursing & Health Sciences 17, 126 133. https://doi.org/10.1111/nhs.12144 Maloney, T., Eiholzer, L., Ryan, B., 2016. Survey of Hispanic Dairy Workers in New York State 2016 (No. EB 2016 - 12). Cornell University, Ithaca, New York. Mas, A., Pallais, A., 2017. Valuing Alternative Work Arrangements. American Economic Review 107, 3722 3759. https://doi.org/ 10.1257/aer.20161500 Milligan, B. 2020. [Online Presentation] http://www.dairystrategies.com/Portals/0/leadership.presentation.video.wmv 43 Miranda, J.J., Diez - Canseco, F., Lema, C., Lescano, A.G., Lagarde, M., Blaauw, D., Huicho, L., 2012. Stated preferenc es of doctors for choosing a job in rural areas of Peru: a discrete choice experiment. PLoS ONE 7, e50567. Nolte, K.D., Fonseca, J.M., 2010. Veg etable field workers provide insight for improving farm labor retention at the U.S. - Mexican border. JAERD 2, 64 72. Rao, K.D., Ryan, M., Shroff, Z., Vujicic, M., Ramani, S., Berman, P., 2013. Rural Clinician Scarcity and Job Preferences of Doctors and Nurses in India: A Discrete Choice Experiment. PLOS ONE 8, e82984. https://doi.org/10.1371/journal.pone.0082984 Rock ers, P.C., Jaskiewicz, W., Kruk, M.E., Phathammavong, O., Vangkonevilay, P., Paphassarang, C., Phachanh, I.T., Wurts, L., Tulenko, K., 2013. Differences in preferences for rural job postings between nursing students and practicing nurses: evidence from a d Human Resources for Health 11. Rockers, P.C., Wanda, J., Laura, W., E, K.M., S, M.G., Francis, N., Kate, T., 2012. Preferences for working in rural clinics among trainee health professionals in Uganda: a discrete choice experiment . BMC Health Services Research 12, 212. Schuster, versus Revealed Employment Preferences in the Peruvian Agro - industry (Working Paper No. 2017/2), Bioeconomics Working Paper Series. Scott, A., Witt, J., Duffield, C., Kalb , G., 2015. What do nurses and midwives value about their jobs? Results from a discrete choice experiment. Journal of Health Services Research & Policy 20, 31 38. https://doi.org/10.1177/1355819614554924 Smitz, M. - F., Witter, S., Lemiere, C., Eozenou, P.H. - V., Lievens, T., Zaman, R.U., Engelhardt, Retention in Timor - Leste: Findings from a Discrete Choice Experiment. PLOS ONE 11, e0165940. https://doi.org/10.1371/journal.pone.0 165940 recruitment and retention: a discrete choice experiment. Health Policy and Planning 30, 68 77. https://doi.org/10.1093/heapol/czt098 Takemura, T., Kielmann, K. , Blaauw, D., 2016. Job preferences among clinical officers in public sector facilities in rural Kenya: a discrete choice experiment. Human Resources for Health 14, 1. https://doi.org/10.1186/s12960 - 015 - 0097 - 0 Taylor, J.E., Charlton, D., Yúnez - Naude, A., 2 012. The End of Farm Labor Abundance. Applied Economic Perspectives and Policy 34, 587 598. 44 Train, K., 2003. Discrete Choice Methods with Simulation. Cambridge University Press, New York. Ubach, C., Scott, A., French, F., Awramenko, M., Needham, G., 2003. What do hospital consultants value about their jobs? A discrete choice experiment. BMJ 326, 1432. https://doi.org/10.1136/bmj.326.7404.1432 U.S. Department of Labor, n.d. NAWS_F2Y185 [WWW Document]. Public Data files in Excel and CSV Formats. URL https://www.dol.gov/agencies/eta/national - agricultural - workers - survey/data/files - excel - csv (accessed 8.11.20). US DA ERS - Ag and Food Sectors and the Economy [WWW Document], 2019. Ag and Food Sectors and the Economy. URL https://www.ers.usda.gov/data - products/ag - and - food - statistics - charting - the - essentials/ag - and - food - sectors - and - the - economy.aspx (accessed 6.23.19). Van den Broeck, G., Van Hoyweghen, K., Maerten s, M., 2016. Employment Conditions in the Senegalese Horticultural Export Industry: A Worker Perspective. Development Policy Review 34, 301 319. https://doi.org/10.1111/dpr.12153 45 ESSAY 2: ARE CONSUME RS WI LLING TO PAY MORE FO R ANIMAL WELFARE OR WORKER WELFARE LA BELS? AN APPLICATION OF EGG AND MILK PRODUCTS UNDER DIFFE RENT INFORMATION SET TINGS 2.1 Introduction Recently, poor working conditions on farms and methods for their improvement have garnished media attention both in the United States and abroad. For example, after negotiations with dairy Vermont, signed the Milk with Dignity agreement, which stipulates minimum wor king conditions for its milk suppliers (Scheiber 2017). Contracts between suppliers (like dairy farms) with other social desirability efforts, there are additional methods to drive industry reform. Perhaps a key option is the implementation of labeling programs, which serve as a vehicle to signal quality features and production practices that are often desirable to consumers. Labeling programs on food attributes suc h as country of origin, production methods, nutritional properties, and carbon footprint , among others are exploding in modern food markets, and indeed many studies indicate that consumers are willing to pay a substantial price premium for products bearing such labels ( Akaichi, et al., 2019; Bazzani et al., 2017; Gerini, et al., 2016; Lee et al., 2015; Van Loo et al., 2015; Van Loo, et al., 2014 ; Van Wezemael et al., 2014; Caputo et al. 2013; Aprile at al., 2012; Onozaka and Mcfadden, 2011 ; Van Loo et al., 2011 ). Yet, to date, only two labels exist that certify working conditions in U.S. agriculture: the Equitable Food Initiative (EFI) and Fair Food Program. Still , their application has been limited in the applicable produce, availability in retailers, and g eographic location. For example, the EFI is limited to participating Costco Wholesale retailers and the Fair Food Program is limited to the Southeastern 46 coast. A natural question is whether the relative absence of worker welfare labels is due to their nas cent introduction to markets or because consumers are uninterested in them and what they represent. This study attempts to answer this question by exploring whether consumers are willing to pay a price premium for two animal - based food products, eggs and m ilk, bearing labels certifying working conditions. To add context to the experimental setting, decisions about products with worker welfare labels were made concurrently with decisions about products bearing animal welfare labels. Thus, we are able to expl ore the relative willingness - to - pay (WTP) for eggs and milk with worker welfare and animal welfare labels as well as the complementarity or substitution effects between them. Additionally, we provide different information treatments to assess how variation s in prior knowledge may impact WTP and label interaction. Overall, we designed two studies on eggs and milk selection, each composed by three different information treatments. This study makes a number of contributions to the fields of animal - product mar keting worker welfare labels in animal - based food production. This study provides new insights into the food choice literature, with the ultimate goal of informing pr oducers, worker advocates, and policymakers. To illustrate, assessing potential consumer demand for nascent worker welfare labeling programs will inform producers about the potential threats and trends to work condition scrutiny. As consumers grant farmers the social license to produce, it is vital that producers understand consumer expectations regarding production. A potential next step for proponents of such a labeling program is to gauge whether animal - based food products should be target products for t hese programs. As labor labels may be the next wave of social desirability labels to 47 enter the market, this contribution is timely and may assist with projections of consumer behavioral changes. The lack of prior studies in this area, also means that desp ite the plethora of studies on animal welfare systems in egg production (Lusk, 2019; Paul, et al., 2019; Heng and Peterson, 2018; Ochs, et al., 2018; Doyon and Bergeron, 2016; Gerini, et al., 2016; Heng, et al., 2013; Andersen, 2011; Norwood and Lusk, 2011 ; Chang, et al., 2010, Bennett and Blaney, 2003; animal welfare versus worker welfare labeling programs. It is feasible that there exists a trade - off between we lfare beneficiaries in animal agricultural production. For example, research suggests that hen housing systems that improve animal welfare, 15 may depress the well - being of workers due to poor air quality and ergonomics (Coalition for Sustainable Egg Supply, N.D.). Such a tradeoff poses difficult questions for all actors along the supply chain in deciding which production method, and thus welfare beneficiary, to support. Even under production methods where a pareto improvement in welfare is possible, a compar ison of willingness - to - pay for both labeling schemes informs producers of whether either or both programs are a lucrative method of product differentiation. Given that some producers may decide to concentrate on welfare 15 Karc her, et al. (2015) posits that research on the pros and cons of hen rearing systems is ongoing, and this is likely true for other animal - rearing systems as well. Nevertheless, at least one source, the Coalition for Sustainable Egg Supply (N.D.), suggests a potential trade - off regarding the benefits for animals and stressors for workers of cage - free systems relative to conventional cages. We do not believe that the motivating framing of this paper depends on there being an actual scientifically proven trade - off, but rather the potential for the consumer to believe there is one, which the cited research could currently provide. Thus, while research into the welfare of animals and workers under different rearing systems is still needed and could impact the inte rpretation of this paper in the future or suggest its repetition based on new information, we suggest that regardless of which rearing system the reader believes better for animals (or humans) , they may have faith and interest in the presented results. 48 standards for only one party, it is important for existing stakeholders of animal welfare labels to understand this potential competing or complementary label. Further, stakeholders interested in adopting either social desirability label will want to know under which conditions it is/they a re most effective. In particular, the information treatment results indicate that consumers are willing to pay a greater premium for labeled products when they have more information. This information helps marketers know which message to promote and sugges ts a return on investment to informational campaigns. The remainder of this paper describes how we came to offer these contributions. It is structured as follows; in section 2.2 we review the closest strains of literature related to consumer preferences f or animal welfare and worker welfare claims and illustrate the research hypotheses. Then we present the egg study followed by the milk study. Within each study we begin by describing the experimental and survey design, followed by the between - subjects trea tment design. Then we describe the data, outline how it will be analyzed and present the results. After both studies have been presented w e conclude the paper with a discussion of both studies, policy implications, and suggestions for further research. 2. 2 Background and Research Hypotheses How consumers value different animal welfare and worker welfare practices is explored in a cornucopia of research areas including socially acceptable credence attributes, social desirability labels, social responsibilit y labels, sustainable labels, and eco - labels. While terminology may presented with information about production practices that are not otherwise known by looking at the product. Such characteristics are called credence attributes ( Caswell and Mojduszka, 1996 ). They are often disclosed by including a label on product packaging. 49 To date, several studies have investigated consumer preferences and WTP for animal welfare label s (Akaichi, et al., 2019; Van Loo, et al., 2014; Heng, et al., 2013; Norwood and Lusk, 2011; Chang, et al. 2010). Results from these studies generally indicate that consumers are willing to pay a price premium for various products bearing such labels, alth ough the price premium may not entirely overcome the associated production costs (Akaichi, et al., 2019; Norwood and Lusk, 2011). Similarly, the few studies on consumer preferences and WTP for worker welfare labels (Drichoutis et al., 2017; Howard and Alle n, 2008) found that consumers are willing to pay a price premium for quality standards certifying worker conditions on strawberries. 16 Taken together the results from these studies suggest that consumers positively evaluate both animal and worker welfare la bels. However, none of these studies has analyzed either how consumers evaluate these labels simultaneously or the worker welfare label applied to animal - based food products. Hence, this study adds to the existing literature by exploring whether consumers are willing to pay a price premium for animal welfare and worker welfare labels on eggs and milk . We hypothesize that consumers are willing to pay a positive price premium for both labels on both animal - based food products ( H 1 ). Anticipating that consumers would be willing to pay a premium for both labels introduces two intriguing questions. For a given product, which label commands a greater premium? And, are animal welfare and worker welfare labels complements or substitutes? Existing studies have not inv estigated the potential substitution and complementarity effects between these two labels. Rather, a number of studies from the sociology literature (Howard, 2006; Howard and Allen, 2006, 2008, 2010) have concentrated on the relative ranking of various 16 F air Trade labels incorporate an element of worker welfare. However, fair trade would not necessarily apply to all workers whereas a worker welfare labeling program could. Therefore, we limit our discussion to labels solely and specifically targeted at work er welfare for a broad group of (all agricultural) workers. 50 o - across production pairs, results from these studies reveal that respondents value animal welfare more than living wages for employees, which can be assimilated to w orker welfare. Additionally, economic studies have found that animal welfare is valued more than other social desirability labels like organic and local ( Akaichi, et al., 2019; Gerini, et al., 2016; Van Loo, et al., 2014 ). Hence, this study hypothesi zes th at the consumer price premium for products with the animal welfare label will be greater than for those with the worker welfare label ( H 2 ) 17 . If the prior research exploring tradeoffs between animal welfare and worker welfare is sparse, literature on comple mentarity or substitution between them is nonexistent. Intuitively, one could imagine that people who care about the welfare of animals would also ca re about the welfare of other creatures as animals. Conversely, as in the egg example, consumers may see a trade - off (whether existing in reality or not) between improving conditions for animals or improving conditions for workers, at which point the labels would substitute for each other. Therefore, no prior expectations are formulated regarding the complement arity and substitution effects between the animal and worker welfare labels. 17 Note that our second hypothesis stems from prior studies that have asked respondents to rank the importance of food topics. We could have also turned to the theoretical sociology literature, fro m which we would likely have drawn the opposite hypothesis. more likely to support other minority groups (Kendall et al., 2006). F rom our interpretation of the underdog hypothesis and stratification theory we proffer that since interactions with humans and work, and individual and relational well - being, are more often interacted with and considered than animal well - being, that reveal ed preferences in the form of action (behavioral decision making, choice) will be stronger for worker welfare than animal welfare. This relationship might be posit ions compared to parallel improvements for animals. Thus, while we hypothesize that consumers will be willing to pay more for animal welfare labels than worker welfare labels we acknowledge that some individuals may expect the opposite relationship while s till relying upon prior research. 51 Finally, previous food choice research has established that how much consumers are willing to pay for each label is likely not only influenced by the presence or absence of oth er labels on the product but also by how much information consumers have about the label ( Caputo, 2020; McFadden and Huffman, 2017; Gifford and Bernard, 2011; Lusk, et al., 2004) . Therefore, consistently with previous studies, we expect consumer WTP to inc rease when more information about both labels is provided to them ( H 3 ). To summarize, we make no prediction about whether animal welfare and worker welfare labels are complements or substitutes. However, we do expect that consumers will prefer animal welf are labels over worker welfare labels. We also expect consumers to prefer any labeled product to the conventional product. These preferences may vary with the level of information consumers have about the labels; specifically, that those with more informat ion are willing to pay greater premiums and that they prefer worker welfare labels (which were previously unknown) even more. To test these hypotheses, we designed two discrete choice experiment studies on eggs and milk selection. In the following sections we describe the steps followed to design and analyze each of the studies as well as report the respective results. 2. 3 Egg Study 2. 3 .1 Choice Experiment and Survey Design This study uses a hypothetical choice experiment on egg selection as worker w elfare labels do not yet exist on this product. Eggs are particularly well suited for a study exploring labeling programs as the market is highly differentiated. For example, different production methods that impact animal welfare (e.g., animal welfare app roved, cage - free, certified humane, pasture - raised and organic, which has provisions for animal housing) are already advertised via package labels. Most notably, eggs are an ideal product for exploring potential consumer trade - offs for animal 52 and worker we lfare as one example of improving animal welfare for chickens, using a cage - free production system, may exacerbate worker respiratory systems and increase risk of other injuries thereby decreasing worker welfare (Coalition for Sustainable Egg Supply n.d.). As our research concentrates on the potential trade - off, or complementarity, between animal welfare and worker welfare labeling programs, which has not been explored in other contexts, we simplify our experimental design to only include these two attribu tes and their interaction, plus price. Table 2.1 reports the attributes and attribute levels selected. Table 2.1 Egg Experimental Design: Attributes and Attribute Levels Attributes Attribute Levels Accredited Cage - Free Label Absent Present Fair Labor La bel Absent Present Price $0.89 $2.29 $3.69 $5.09 The cage - free label was chosen to represent animal welfare labeling for two primary reasons. First, cage - free labels and claims have become popular in the egg industry since some retailers pledged to only source cage - free eggs or as legislation has required egg production to be cage - free as in California (Lusk, 2019). Additionally, it is the cage - free aviary system that some researchers have identified as more harmful to workers than the conventional b attery cage system (Coalition for Sustainable Egg Supply, n.d .). The Fair Labor label was a new one created to explicitly differentiate this potential new label from either of the fair food program or fair 53 trade labels and to draw attention to the worker welfare motivation/beneficiary. The price range was chosen to reflec t the different prices for eggs in the market. Prices were determined based on the low and high reported national market prices for the prior year, December 1, 2017 through November 23, 2018. Low and high prices ranged from $0.39 (Southeast White A Large) to $7.98 (Northeast Brown Organic Large) per dozen, while the weekly weighted average price ranged from $0.49 - $5.98 per dozen eggs (United States Department of Agriculture Agricultural Marketing Service, n.d .a). Additionally, these price ranges were then c ompared to those used in prior literature. Prior studies conducted in the last ten years used prices ranging from $0.50 - 4.99 for a dozen eggs (Lusk, 2019; Heng and Peterson, 2018; Gerini, et al., 2016; Heng, et al., 2013; Norwood and Lusk, 2011). Consideri ng both recent national market prices and historical prices in prior studies, price ranges were finalized to those reported in Table 2.1. An optimal in the orthogonal differences (OOD) fractional factorial design was generated following Street and Burgess (2007). The final design includes main and one - way interaction effects, with the latter being represented by the interaction between animal and worker welfare labels. The design resulted in 16 choice sets divided into two blocks, with both blocks equally likely in each treatment. Therefore, a respondent was randomly assigned to see one set of eight choice questions. An example egg choice question is presented in Figure A 2.1 in the A ppendix. The order of questions was randomized within each block to reduce order bias. contained the consent question and screening questions. It concluded with the respective following language that ensured a common understanding of terminology (e mphasis in survey): The next section proceeded by asking participants for information on their prior purchase behavior of eggs. The economic ex periment comprised the third section of the study. 54 Prior to this section respondents were randomly assigned to an information treatment group. The survey concluded with demographic questions. 2. 3 .2 Between - Subjects Treatments To explore whether consumer preferences and WTP for animal welfare and worker welfare labels varies under different information setting, we designed a between - subject experiment, whereby respondents were randomly assigned either to the Control or one of the two i nformation treatments named as follows : Label and Label & Media (summarized in Table 2.2) . Table 2.2 Information Treatments Treatment ID Treatment Name Information 1 Control None 2 Label Label Description 3 Label & Media Label Description & News Article In the Con trol respondents were faced with 8 choice questions and they were not provided with any additional information about the animal welfare and worker welfare labels. Hence, it represents the baseline knowledge that consumers would bring with them into markets at the moment of the study. In the Label treatment, prior to the choice questions participants were provided with basic information about the meaning of both animal welfare and worker welfare labels. This information is what marketers hope or assume consu mers know about the quality standards implied by a label which provides the value to the labeling program. In the Label & Media treatment, prior to the choice questions participants were faced with the same information as in the low information treatment p lus they were asked to read a news article. The news article 55 was written for this study by the authors to explore the effect of popular media on decision behavior. It included mention of both worker welfare and animal welfare labels to not disproportionate ly skew responses toward one label. In this way the explicit information was in a format similar to but perhaps more balanced or comprehensive than tactics used by advocacy parties to express their point of view in public discourses. A detailed description of all information treatments is provided in Exhibit A 2.1 of the A ppendix. Common across all treatments were the choice experiment directions. The directions included both a cheap talk and consequentiality script as prior research has suggested that doing so may reduce hypothetical bias. 18 Directions and each other treatment element (label definitions and news article) were set on a timer to encourage thoughtful reading. 2. 3 . 3 Data Data was collected through a national survey administered online by Qualtr ics from January - April 2019. A total of 536 respondents participated in our study, 179 in the Control , 178 in the Label treatment , and 179 in the Label & Media treatment. Respondents were required to be adults who had purchased eggs in the last three month s. Basic demographics of the sample by experimental treatment are reported in Table 2.3. Our sample is skewed towards traditional 18 Bias may result in reported greater willingness - to - pay in hypothetical experiments than real since the costs of choosing the purchase option are equal to zero. People may display social desirability bias where they belie ve that researchers prefer an option to be chosen over the no - buy alternative, that they are more willing to engage in with lower transaction costs (Norwood and Lusk, 2011). Reminding respondents of what the consequences would be in both a real shopping en vironment (cheap talk) and based on analysis of their responses with those from others (consequentiality) may reduce potential bias. We also capitalized the no purchase option in the directions to draw attention to this option. Nevertheless, we acknowledge that the hypothetical nature of our experiment should result in caution in extrapolating to future, real market scenarios when worker welfare labels will be introduced in food markets . 56 grocery shoppers and respondents to food surveys who are more often female (Aarnio and Lindeman, 2004 p.67; Costanigro et al., 2011 p.467 - 9). There were statistically significant differences across demographics in the three treatment groups despite random assignment. Table 2.3 Egg Sample Demographics Control Label Label & Media p - value A N= 179 N= 178 N= 179 Age (over 18) Continuous variable represents years of age 45.283 (16.434) B 43.423 (15.395) 45.583 (15.899) 0. 000 Gender 1 if respondent female; 0 otherwise 0.706 (0.457) 0.674 (0.470) 0.810 (0.393) 0.0 00 Education Low 1 if respondent does not h ave high school degree; 0 otherwise 0.028 (0.164) 0.027 (0.163) 0.011 (0.105) 0.000 Mid 1 if respondent has a high school degree but not a 0 otherwise 0.678 (0.467) 0.604 (0.489) 0.739 (0.439) 0.000 High 1 if respondent has ba or higher; 0 otherwise 0.294 (0.456) 0.368 (0.482) 0.250 (0.433) 0.000 Income Low 1 if respondent has income below $35,000 annually; 0 otherwise 0.556 (0.497) 0.571 (0.495) 0.544 (0.498) 0.039 Mid 1 if respondent has income between $35,000 and $100,000 annually; 0 otherwise 0.444 (0.497) 0.429 (0.495) 0.456 (0.498) 0.039 High 1 if respondent has income above $100,000 annually; 0 otherwise 0.122 (0.328) 0.099 (0.299) 0.122 (0.328) 0.000 57 Table 2.3 (cont d) Race Caucasian 1 if respondent is Caucasian; 0 otherwise 0.800 (0.400) 0.731 (0.444) 0.733 (0.442) 0.000 African American 1 if respondent is African America; 0 otherwise 0.094 (0.292) 0.148 (0.355) 0.156 (0.362) 0.000 U.S. Census Region Northeast 1 if resides in Northeast U.S. census region; 0 otherwise 0.183 (0.387) 0.225 (0.418) 0.189 (0.391) 0.000 Midwest 1 if resides in Midwest U.S. census region; 0 otherwise 0.306 (0.461) 0.231 (0.421) 0.217 (0.412) 0.000 Sout h 1 if resides in South U.S. census region; 0 otherwise 0.367 (0.482) 0.357 (0.479) 0.433 (0.496) 0.000 West 1 if resides in West U.S. census region; 0 otherwise 0.144 (0.352) 0.187 (0.390) 0.161 (0.368) 0.000 Political Party Democrat 1 if responde nt Democrat; 0 otherwise 0.339 (0.473) 0.319 (0.466) 0.394 (0.489) 0.000 Republican 1 if respondent Republican; 0 otherwise 0.289 (0.453) 0.291 (0.454) 0.267 (0.442) 0.020 In - group Identification Ranges from 1 (low national (American) in - group identifica tion) to 5 (high in - group id) 3.726 (0.777) 3.716 (0.808) 3.751 (0.834) 0.115 Illegal Aliens Scale - 1 if think negatively of illegal aliens; 1 if think positively of illegal aliens; 0 if neutral - 0.206 (0.941) - 0.209 (0.938) - 0.306 (0.914) 0.000 A One - wa y ANOVA B Standard deviations are in parentheses 2. 3 . 4 Data Analysis Discrete choice experiments are consistent with Random Utility Theory (RUT) which postulates that if a product is chosen from a set then it routinely provides the individual greater uti lity than the other articles in that set. As shown in Train (2009), the utility that individual n derives from alternative j at choice situation t can be expressed as follows: (1) 58 where is the observed portion of utility and is the random, unobserved component. Assuming that the are distributed, Type I Extreme Value yields the familiar M ultinomial Logit (MNL) model, which as sumes preference homogeneity in the sample . That is: all coefficients of the utility expressi on in equation (1) are the same across individuals. However, if preference heterogeneity is expected, then a mixed logit model (MXL) , which allows for a more flexible and continuous form of preference heterogeneity , should be specified. Previous food choice studies have indicated that heterogeneity in consumer preferences is a pattern in demand analysis and that MXL models best describe food choice behavior ( Tonsor, et al., 2009; Lusk and Schoeder 2004; Lusk , et al., 2003 ) . Further studies have also shown that utilities of product profiles are indeed correlated ( Scarpa et al., 2005; Scarpa et al., 2007a ). Accordingly, the data were analyzed using a MXL with Error Component (MXL - EC) as it allows us to account for i) random taste variation by allowing the coefficients of animal welfare and worker welfare labels to vary randomly over individuals and to deviate from the population mean (Train 2009), an d ii) for correlation across utilities by capturing the additional variance shared by the utility associated with the designed product alternatives but not by the opt - out alternative ( Scarpa et al., 2005; Scarpa et al., 2007a ). Further, the models were es timated in WTP - space to account for random taste variation associated with the cost coefficient (Scarpa et al., 2008) and with full - correlated random coefficients ( Caputo , et al. , 2013) . Estimations in WTP - space allow for easy comparison across treatment g roups as estimates are interpreted directly as marginal WTP values (Scarpa et al., 2008). A total of three segmented MXL - EC models were estimated, one for each treatment: Model - C, Model - L, and Model - LC for the Control , Label Information , and Label & Media 59 Information, respectively . 19 For each treatment, we specified the following indirect utility function: (2) where is the price scale parameter ( where is the is Gumbel scale parameter ); ASC is the alternative specific constant which represents the opt - out option ; Price njt is the observed price of the alternative in the choice experiment task; AW njt and WW njt are dummy variables coded 1 i f the animal welfare and worker welfare labels respectively are present and 0 if absent; is the experimentally designed interaction term between the AW and WW labels; terms represent the marginal willingness - to - pay for the resp ective attribute; is the error component d istributed normally with zero mean ; and is the error term. The label mean parameters are assumed normally distributed, while the price scale parameter is assumed log - normal distributed. 20 To explore whether consumer WTP for animal welfare and worker welfare labels are significantly influenced by information provision we employed a pooled data approach and estimated three MXL - EC models to represent each of the treatment pairs: Model - CL, Model - CLM, and Model - LLM. More specifically, Model - CL was estimated by pooling the data from Control and Label treatments, Model - CLM was estimated by pooling the data from Control and 19 A joint model restringing all the marginal WTPs for both animal wel fare and workers welfare labels to be the same across the Control , Label and Label & Media treatments was also estimated. Results from the Log - Likelihood ratio (LR) test suggests that the Marginal WTP for both labels differ across treatments as the chi - squ are statistic exceed the 1 percent critical value (p - value<0.001). This evidence indicates that comparing the marginal WTPs from the various treatments is appropriate when estimating the models separately. 20 We estimated models assuming the experimentally designed interaction between the two labels was normally distributed. However, the coefficient of the standard deviation was found never to be statistically significant. Therefore, it is assumed non - random. 60 Label & Media treatments, and Model - LLM was estimated by pooling data from the Label and Label & Media treatment groups. Following previous food choice studies (de Magistris et al, 2013; Bazzani et al., 2017, Lin et al., 2018, among others), the treatment effects were then estimated by using extended utility functions including inte raction terms between a dummy variable identifying the treatment of interest ( ) and each of the label s . For example, in the Model - LLM the utility was specified as follows: (3) The significance of and their signs represent the treatment effect on the marginal WTPs for the two labels . For Treat i n the Model - CL and Model - CLM we used the Control as baseline, while in the Model - LLM the baseline is the Label treatment. The rest of the variables are as defined in equation ( 2) . 2. 3 . 5 Results Table 2.4. reports WTP estimates from the MXL - EC mo del s with full correlated random coefficients . 21 T he first three columns report the three segmented models, which were estimated for each treatment group: Model - C ( Control ), Model - L ( Label Information ), and Model - LM ( Label & Media Information ). The last th ree columns report the estimates from the pooled models with Model - CL comparing the Control and Label Information , Model - CLM comparing the Control and Label & Media Information treatment group, and Model - LLM comparing the Label Information and Label & Medi a Information treatment groups. 21 Cholesky matrices are reported in Tables A2.1, A2.2, and A2.3 in the Appendix. 61 Table 2.4 Estimates from the MXL - EC with correlation in WTP - space for Eggs Split Sample Models Pooled Models Model - C Model - L Model - LM Model - C L Model - CLM Mode l - LLM Variables/ Parameters Animal Welfare µ 0. 72*** 1.13*** 1.28*** 0.81*** 0.85*** 1.20*** (0.19) A (0.23) (0.24) (0.18) (0.19) (0.20) 1.02*** 1.44*** 1.65*** 1.24*** 1.32*** 1.57*** (0.18) (0.18) (0.20) (0.12) (0.12) (0.13) Worker Welfare µ 0.32* 0.51*** 0.78*** 0.41*** 0.49*** 0.56*** (0.19) (0.19) (0.18) (0.14) (0.15) (0.16) 0.74*** 0.70*** 1.23*** 0.64*** 0.92*** 1.03*** (0.17) (0.14) (0.11) (0.11) (0.08) (0.10) Animal Welfare * Workers welfare µ 0.00 - 0.18 - 0.21 - 0.12 - 0.14 - 0.20 (0.28) (0.26) (0.23) (0.19) (0.17) ( 0.17) ASC µ - 4.78*** - 5.30*** - 4.08*** - 5.09*** - 4.43*** - 4.58*** (0.41) (0.48) (0.34) (0.32) (0.15) (0.28) Error Component 3.77*** 3.60*** 2.74*** 3.62*** 3.38*** 3.13*** (0.88) (0.97) (0.32) (2.06) (0.78) (0.26) Price scale parame ter µ - 0.89*** - 1.03*** - 1.03*** - 0.96*** - 0.95*** - 1.04*** (0.02) (0.03) (0.04) (0.02) (0.02) (0.02) 0.89*** 1.03*** 1.03*** 0.96*** 0.95*** 1.04*** (0.02) (0.03) (0.04) (0.02) (0.96) (0.02) Treatment Effects B Label Treatment * Anima l Welfare 0.30 (0.21) Label Treatment * Worker Welfare 0.11 (0.13) Label & Media Treatment * Animal Welfare 0.36* 0.06 (0.22) 0.26 Label & Media Treatment * Animal Welfare 0.26* 0.22 (0.15) 0.18 62 Table 2.4 (cont d) Model Statistics Number of Respondents 179 178 179 357 358 357 Number of Observations 1432 1424 1432 2856 2864 2856 Number of parameters 11 11 11 13 13 13 Log - likelihood function - 979.17 - 939. 73 - 1005.08 - 1924.04 - 1997.4752 - 1946.87 AIC/N 1.383 1.335 1.419 1.356 1.404 1.372 BIC/N 1.423 1.376 1.460 1.384 1.431 1.400 A Standard errors are in parentheses. B The treatment variable is always a dummy variable equaling 0 for the lower numbered (less information) treatment Looking at the segmented models first (Model - C, Model - L, and Model - LM), it can be noted that the coefficient on the opt - out option is always negative and statistically significant across all treatments, indicating that respondents w ould rather purchase any of the egg alternatives than not purchase. Additionally, in all treatments the standard deviation coefficient of the error component is statistically significant, suggesting that there is greater preference variability for the egg product profiles than the opt - out option. Interestingly, the interaction term is never statistically significant. This suggests that the value that consumers put o n one label, animal welfare, is independent of the value associated to the other label, worke r welfare, and vice - versa. They are separate concepts that are neither substitutes for each other nor complimentary methods of enhancing well - being for living creatures, both animal and human. Further, our results indicate that both the animal welfare and worker welfare labels are positive and significant across all treatments confirming our first hypothesis (H 1 ). Further, as conjectured (H 2 ), results reveal that consumers have a higher price premium for the animal welfare label over 63 the worker welfare labe l across all treatments. Yet, as expected, the magnitude of this premium increase s with more information for both labels, indicating that price premiums for both labels increase when consumers are provided with more information (H 3 ). To illustrate, i n the Control , the estimated mean s of the marginal WTP distribution for the animal welfare label is $ 0.72, while it is only $0.32 for the workers welfare label. These values go up in the Label treatment, where the estimated mean s of the marginal WTP distribution is $1.13 and $0.51 for the animal welfare and worker welfare labels, respectively. As expected, an even higher price premium for both labels is found in the Label & Media treatment; again, the a nimal w elfare label has the largest value estimate ($1.28) , while the w orker w elfare label ($0.79) follows. Further, t he significance of the estimated standard deviations around the mean s of the marginal WTP s for both labels suggest substantial heterogeneity with a general shift towards a higher share of positive m arginal WTPs for both labels in the Label and Label & Media treatments . For example, the share of positive marginal WTPs for the a nimal w elfare label is 7 6 % , 78 % , and 78% in the Control, Label s , and Label & Media treatments, respectively, while it is 6 7 % , 77 % , and 74 % for the w orkers w elfare label in the Control, Label s , and Label & Media treatments, respectively. Turing to the results from the pooled models (last three columns of Table 2.4 ), it can be seen that in the Model - CL the coefficients of the int eraction terms between each of the two labels and the Label treatment are not statistically significant, indicating that there is no difference in consumer valuation for both animal welfare and worker welfare labels between the Control and the Label treatm ents (the Control is the baseline). The results from the Model - CLM model, on the other hand, indicate that the marginal WTPs for both labels increase significantly when respondents are exposed to both label information and information from the news article (the Control is the baseline); the coefficients of the interaction terms between the animal welfare and 64 worker welfare labels and the Label & Media treatment is statistically significant from both labels: $0.36 and $0.26 for the animal welfare and worker welfare labels, respectively. This indicates that consumers are willing to pay $1.21 ($0.85+$0.36) for the animal welfare label and $ 0. 75 ($ 0. 49+ $ 0. 26) in the Label & Media treatment, while only $ 0. 85 and $ 0. 49 in the Control and that these differences a re statistically significant. These results are consistent with the marginal WTPs found in the segmented models. Finally, when looking at the results from the Model - LLM, it is apparent that there are no significant differences in consumer marginal WTPs for the two labels between the Label and Label & Media treatments as denoted by the insignificant coefficients of the interaction terms between the two L abel and the Label & Media treatment s ( the Label treatment is the baseline). 2. 4 Milk Study 2. 4 .1 Choice Experiment and Survey Design This study also uses a hypothetical choice experiment on gallon milk selection as worker welfare labels are not available for milk products in food markets . Milk is a frequently purchased item which is ideal for choice experim ents. Milk is a model product for exploring consumer preferences for a nascent worker welfare label, specifically, as poor working conditions in the industry have been revealed on some operations according to Migrant Justice. This suggests a worker welfare labeling program may be particularly beneficial for some dairy industry workers. Like the egg study, only three attributes were isolated so as to best explore WTP for worker welfare labeling compared to animal welfare labeling. These attributes and their levels are provided in Table 2. 5 . The animal welfare label differed from the egg study to a label that applied to milk. Specifically, the animal welfare approved label was chosen. The same author - designed Fair Labor label was used as the worker welfare la bel as in the egg study. Similarly, 65 gallon milk prices were determined by recent market prices ranging from $0.39 to $7.98 (United States Department of Agriculture Agricultural Marketing Service, n.d .b) and prior studies ranging from $1.13 to $7.00 ( Caputo , et al., 2020; Caputo, et al., 2018 ; Crespi, et al., 2015; Wolf, et al., 2011; Brooks and Lusk, 2010; Olynk, et al., 2010). Table 2. 5 Milk Experimental Design: Attributes and Attribute Levels Attributes Attribute Levels Animal Welfare Approved Label Abs ent Present Fair Labor Label Absent Present Price $1.79 $3.19 $4.59 $5.99 As in the eggs study, we employed an optimal in the orthogonal differences (OOD) fractional factorial including main and one - way interaction effects, with the latter being repr esented by the interaction between animal and worker welfare labels. The design resulted in 16 choice sets divided into two blocks of 8 questions each, and respondents were randomly assigned to one block. An example milk choice question is presented in Fig ure A2. 2 in the A ppendix . The milk study had the same four sections as the egg study . The study began with consent and screening questions plus the following language (emphasis in survey): Please keep in mind that for this survey lk produced by cows, not other animals or similarly labeled plant products, like soy milk, almond milk, etc. Then respondents indicated their prior purchase behavior of milk. An important question in this section was one that asked which package size indi viduals typically purchase. If individuals selected any size less than a 66 difference in milk attribute preferences by product size (Wolf et al., 2011). Then respond ents were randomly assigned to an information treatment. The information treatments were the same as the egg study ( Control , Label , and Label & Media , recall Table 2.2 and see Exhibit A 2.2 in the Appendix), except that the animal welfare approved label def inition replaced the cage - free one. In each treatment respondents were asked to answer 8 choice questions. The final section after the experiment contained demographic questions. 2. 4 .2 Data Data collection concurred with that for the egg study. A total o f 762 respondents participated in our study, 253 in the Control , 258 in the Label treatment , and 251 in the Label & Media treatment. Respondents were eligible to participate if they were adults who had purchased mil k in the last three months. Sample demogr aphics are presented in Table 2. 6 . There are several statistically significant observed differences between treatment groups across demographics within the sample despite random assignment . Table 2. 6 Milk Sample Demographics Control Label Label & Media p - value A N=253 N=258 N=251 Age (over 18) Continuous variable represents years of age 44.806 (14.631) A 42.874 (14.139) 42.923 (15.905) 0.000 Gender 1 if respondent female; 0 otherwise 0.758 (0.428) 0.731 (0.444) 0.742 ( 0.438) 0.013 67 Table 2.6 (cont d) Educ ation Low 1 if respondent does not have high school degree; 0 otherwise 0.032 (0.177) 0.049 (0.217) 0.027 (0.163) 0.000 Mid 1 if respondent has a high school degree but not a 0 otherwise 0.704 (0.456) 0.648 (0.478) 0.626 (0484) 0.000 High 1 if respondent has or higher; 0 otherwise 0.263 (0.441) 0.302 (0.460) 0.346 (0.476) 0.000 Income Low 1 if respondent has income below $35,000 annually; 0 other wise 0.559 (0.497) 0.533 (0.499) 0.527 (0.499) 0.006 Mid 1 if respondent has income between $35,000 and $100,000 annually; 0 otherwise 0.441 (0.497) 0.467 (0.499) 0.473 (0.499) 0.006 High 1 if respondent has income above $100,000 annually; 0 otherwise 0. 086 (0.280) 0.115 (0.320) 0.115 (0.320) 0.000 U.S. Census Region Northeast 1 if resides in Northeast U.S. census region; 0 otherwise 0.204 (0.403) 0.181 (0.385) 0.154 (0.361) 0.000 Midwest 1 if resides in Midwest U.S. census region; 0 otherwise 0. 280 (0.449) 0.280 (0.449) 0.258 (0.438) 0.032 South 1 if resides in South U.S. census region; 0 otherwise 0.387 (0.487) 0.401 (0.490) 0.170 (0.376) 0.014 West 1 if resides in West U.S. census region; 0 otherwise 0.129 (0.335) 0.137 (0.344) 0.170 (0.376) 0.000 In - group Identification Ranges from 1 (low national (American) in - group identification) to 5 (high in - group id) 3.964 (0.714 3.822 (0.709) 3.750 (0.716) 0.000 Illegal Aliens Scale - 1 if think negatively of illegal aliens; 1 if think positively of i llegal aliens; 0 if neutral - 0.430 (0.860) - 0.242 (0.930) - 0.253 (0.927) 0.000 A One - way ANOVA B Standard deviations are in parentheses 68 2.4.3 Data Analysis As for the eggs study, the milk data were analyzed using a MXL - EC model specified in WTP space. Mo re specifically, we estimated three MXL - EC segmented models (Model - C, Model - L, and Model - LM for the Control , Label , and Label & Media , respectively) and three MXL - EC pooled models: Model - CL, Model - CLM, and Model - LLM. All models were estimated assuming full correlated random coefficients. Similarly, to the eggs study, we specified the following indirect utility function for the segmented models: (4) where is the price scale parameter ( where is Gumbel scale parameter ); ASC is the alternative specific constant which represents the opt - ou t option ; Price njt is the observed price of the alternative in the choice experiment task; AW njt and WW njt are dummy variables coded 1 if the animal welfare and worker welfare labels respectively are present and 0 if absent; is th e experimentally designed interaction term between the AW and WW labels; terms represent the marginal willingness - to - pay for the respective attribute; is the error component d istributed normally with zero mean ; and is the error t erm. The label mean parameters are assumed normally distributed, while the price scale parameter is assumed log - normal distributed. The pooled models were specified as follows: (5) 69 As in the eggs study, capture the treatment effects on the marginal WTPs for the two la bels. In the Model - CL and Model - CLM we used the Control as baseline, while in the Model - LLM the baseline is the Label treatment. The rest of the variables are as defined in equation ( 4) . 2. 4 . 4 Results Table 2. 7 reports WTP estimates from the MXL - EC model s with full correlated random coefficients . 22 T he first three columns report three segmented models, which were estimated for each treatment group: Model - C ( Control ), Model - L ( Label ), and Model - LM ( Label & Media ) , while the last three columns report the estimates from the three pooled models: Model - CL, Model - C - LM, and Model - LLM. Table 2. 7 . Estimates from the MXL - EC with correlation in WTP - space for Milk Split Sample Models Pooled Models Model - C Model - L Model - LM Model - C L Model - CLM Mode l - LLM V ariables/ Parameters Animal Welfare µ 0.71*** 1.22*** 1.25*** 0.74*** 0.70*** 1.20*** (0.17) (0.18) (0.18) (0.16) (0.15) (0.15) 1.14*** 1.77*** 1.38*** 1.47*** 1.23*** 1.56*** (0.13) (0.13) (0.16) (0.09) (0.10) (0.10) Worker Welfare µ 0.71*** 0.78*** 0.96*** 0.71*** 0.70*** 0.79*** (0.15) (0.15) (0.17) (0.13) (0.14) (0.14) 1.04*** 1.44*** 1.15*** 1.25*** 1.25*** 1.44 *** (0.18) (0.11) (0.16) (0.09) (0.07) (0.07) Animal Welfare * Workers welfare µ - 0.18 - 0.07 - 0.13 - 0.14 - 0.15 - 10.80 (0.23) (0.19) (0.18) (0.15) (0.14) (0.13) ASC µ - 4.29*** - 4.41*** - 4.27*** - 4.36*** - 4.27*** - 4.30*** (0.22) (0.2 3) (0.28) (0.16) (0.17) (0.18) 22 Cholesky matrices are reported in Tables A2.4, A2.5, and A2.6 in the Appendix 70 Table 2. 7 (cont d) Error Component 2.33*** 2.09*** 3.00*** 2.25*** 2.63*** 2.56*** (0.21) (0.19) (0.27) (0.14) (0.16) (0.15) Price scale parameter µ - 0.98*** - 1.09*** - 0.93*** - 1.04*** - 0.96*** - 1.01*** (0.02) (0.03) (0.02) (0.02) (0.02) (0.02) 0.98*** 1.09*** 0.93*** 1.04 0.96*** 1.01*** (0.02) (0.03) (0.02) (0.02) (0.02) (0.02) Treatment Effects B Label Treatment * Animal Welfare 0.49* (0.20) Label Treatment * Worker Welfare 0.10 (0.15) Label & Media Treatment * Animal Welfare 0.50*** 0.04 (0.17) (0.20) Label & Media Treatment * Worker Welfare 0.24* 0.16 (0.15) (0.16) Model Statistics Number of Respondents 253 258 251 511 504 509 Number of Observations 2024 2064 2008 4088 4032 4072 Number of parameters 11 11 11 13 13 13 Log - lik elihood function - 1473.67 - 1462.68 - 1438.60 - 2944.96 - 2916.67 - 2909.35 AIC/N 1.467 1.428 1.444 1.447 1.453 1.435 BIC/N 1.498 1.458 1.475 1.467 1.474 1.455 A Standard errors are in parentheses. B The treatment variable is always a dummy variable equaling 0 for the lower numbered (less information) treatment. As with the egg study the coefficient on the opt - out option is always negative and statistically significant across all treatments, indicating that respondents rather purchase any of 71 the milk alternat ives than not purchase. Again, like the egg experiment, the experimentally designed interaction term between the two labels is never statistically significant. This means that the labels are not complements or substitutes. On the other hand, the individual label coefficients are always positive and statistically significant across all treatments, which confirms H 1 . As in the eggs study, the animal welfare label is preferred over the worker welfare label in all treatments , except in the Control where consume rs are willing to pay, on average, $0.70 for both labels. This evidence confirms our hypothesis (H 2 ). In addition, there are slight differences between the relative magnitudes across treatments of both welfare labels, except in the Control where consumers, on average, are willing to pay $0.70 for both welfare labels. For instance, consumers are willing to pay $1.22 and $1.25 for the animal welfare label in the Label and Label & M edia treatments, respectively. Similar patterns are observed for the worker we lfare label for which consumers are, on average, willing to pay $0.78 and $0.96 in the Label and Label & Media treatments, respectively. These results confirm our third hypothesis (H 3 ) ; namely, consumer WTP increases when more information about both labels is provided to them. Finally, as in the eggs study, the coefficients of the standard deviations are significant for both labels, indicating substantial heterogeneity. The magnitude of these coefficients further indicate s that the share of positive margin al WTPs for the Animal Welfare label is 73 % , 94 % , and 90% in the Control, Label s , and Label & Media treatments, respectively, while it is 75 % , 71 % , and 80 % for the w orker w elfare label in the Control, Label s , and Label & Media treatments, respectively. Ta ken together, these results suggest there are differences in premiums for each label across treatment groups, confirming our third hypothesis (H3). Hypothesis 3 is further evaluated, though, with the pooled models (Model - CL, Model - CLM, and Model - LLM) rep orted in the last 3 columns of Table 2. 7 . There also appears to be statistically significant differences in premiums for each label across the Control and the Label 72 and Label & Media treatment groups, expect for the worker welfare label between the Contro l and Label information treatments. To illustrate, the coefficients of the interaction terms between animal welfare and the Label and the Label & Media treatments are statistically significant in both the Model - CL and Model - CLM models ($0.49 and $0.50), co nfirming that prov id ing consumers with label and media information increases an existing premium for the label. As for the worker label, only the interaction term in the CLM model is statistically significant ($0.24). As in the eggs study, none of the int eraction terms in the Model - LLM are statistically significant. 2.5 Discussion and Policy Implications WTP for a label certifying working conditions on eggs and milk compared to valuation for an animal welfare label, under different information settings. A hypothetical discrete choice experiment was designed for each product to answer the following four questions: (1) How does consumer willingness - to - pay (WTP) for eggs and milk vary in the pres ence or absence of animal welfare and worker welfare labels? (2) Which labeling scheme (animal welfare or worker welfare) do consumers prefer? (3) Are these labels substitutes or complements? And (4) How do consumer preferences for animal welfare and worke r welfare change under different information settings? The results from our MXL - EC models provide insights relevant for producers and policymakers. The first question addressed is whether worker welfare labels will be viable in animal - based food markets. S ince the WTP estimates for the worker welfare label were always positive and statistically significant there is support that consumers would be willing to pay a price premium for animal - based products marketed with a worker welfare label. This suggests tha t the implementation of a labeling program on worker welfare may create new market opportunities for producers and agribusinesses. Additionally, labels certifying working 73 conditions may be a method of offsetting producer costs of improved working condition s by asking consumers to contribute to this cause. The finding that consumers are willing to pay a positive premium for a worker welfare label is consistent with Drichotis et al (2017) and Howard and Allen (2008) who conclude that European and U.S. consume rs respectively are willing to pay a price premium for strawberries with a label certifying working conditions. In addition to a positive WTP for worker welfare labels on animal - based food products in the U.S., our results indicate that consumers are will ing to pay a positive price premium for is novel is that we can identify a relative ranking between animal welfare and worker welfare labels. Respondents consis tently indicated a greater WTP for the animal welfare label than the worker welfare label which is consistent with prior studies in sociology (Howard, 2006; Howard and Allen, 2006, 2008, 2010) . We do not profess to know the mechanism behind this ranking bu t posit a number of potential factors contributing to this relationship. First, the animal welfare labels used in these experiments are currently present in the market while the worker welfare label is not. Therefore, consumers may have positive prior expe rience, trust, and expectations surrounding the existing label compared to the unknown nascent one. Additionally, consumers may associate a halo effect with the animal welfare label that is not associated with the worker welfare label. Specifically, consum ers may believe that enhanced animal welfare practices result in better animal - based food products, whereas no similar linkage between worker welfare practices and the product quality may be believed. 23 Independent of the cause behind this relationship, thi s suggests that producers and agribusinesses can achieve a greater price premium for animal welfare labels than worker welfare labels. 23 The authors would like to thank Jie Li for this suggestion at the Agricultural and Applied Economics Association (AAEA) annual meeting 2019 i n Atlanta, GA. 74 Further, our results indicate that c onsumers with the label definitions are willing to pay a greater premium than those without this prior knowledge ( Control ) as evidenced by the positive and statistically significant information treatment label interaction terms. However, there was no statistically significant observed difference in WTP between the Label and Label & Media treatment groups. This implies that as long as people know the definition of the label, additional information like a news article, does not affect their choice behavior. This result is similar to Berry, et al. (2017) who found that labeling alters purchas e behavior but that a news article had no influence. Nevertheless, despite statistical significance price premiums for both labels did nominally increase under the news article treatment compared to label definitions alone so it may be worth exploring whet her constructing a news article to better articulate and emphasize the label definition would increase the usefulness as media as a source of information propagation. Yet, the information effect should not be overlooked. Of statistically significant result s the premium ranged from $0.24 to $0.50 for milk. Market milk prices in the year prior to the choice experiment were at one point as low as $0.39 (United States Department of Agriculture Agricultural Marketing Service, n.d .b), which suggests that the info rmation treatment effect could approximately double the market price. Even at the conservative end of the highest price of $7.98 (United States Department of Agriculture Agricultural Marketing Service, N.D.b), a 3 % expressed WTP difference may lead to siza bly more revenue for a product with high purchase frequency. Finally, there was a non - statistically significant result worth acknowledging, which is that the animal welfare and worker welfare labels are independent. There was no statistical evidence that t hese labels were substitutes or complements. Thus, like consumers, producers and agribusinesses interested in adopting either label ought to evaluate each label independently. They may decide that including both labels on their product is profitable but do ing so does not 75 add any additional value than that provided by each separate label. To reiterate, when assessing each label on its own producers and agribusinesses will likely gravitate toward the animal welfare labels as consumers indicate a greater willi ngness to pay for them than worker welfare labels. Additionally, this label is already existing in the market and offers an immediate method of product differentiation. As such, some consumers may already be familiar with the label definition or more recep tive to such information which could further increase their WTP for the labeled product. Results suggest that individuals responsible for managing welfare labeling programs may want to highlight the definition of their labeling program when strategizing ab out its effectiveness. The more information about the program that consumers are aware of the more likely they are to support it financially. practical applications? A labe ling program is only one option available to policymakers. The fact that consumers were willing to pay a premium for the worker welfare label and that this premium increased with more information posits that both producers and policymakers will want to tra ck future changes in consumer attitudes toward agricultural working conditions. Producers do not want to face decreased demand or boycott, for example, due to losing the social license to produce from consumers. Additionally, policymakers could always forc e select minimum working conditions on the agricultural sector through legislation so it is worth exploring if they want external mandates and enforcement for working conditions or are happy to selectively support establishments strong in their working pra ctices by purchasing differentiated (labeled) products in the market. 76 2.6 Conclusion Recently labels certifying working conditions have entered the food market, but their application is limited geographically and by product type. For instance, to the no worker welfare labels applied to animal - based food products. A feasible next step would be to create a certification program that verifies and communicates working conditions to consumers with a product label. Results sugge st that consumers would be willing to pay a price premium for such labeled products. On the other hand, animal - based food producers may find animal welfare labeling programs more beneficial as their proliferation is greater and as consumers are willing to pay a greater premium than for worker welfare. It is possible that producers would want to participate in both animal and worker welfare labeling programs as consumers do not see these labels as substitutes, however there is also no complementarity premium to be gained by including both. A strategy that is successful in increasing the price premium is providing more information to consumers, although there may not be much additional value to using a media outlet versus simply ensuring awareness of the label definition. Despite all that was learned from this study, which is the first to explore worker welfare labeling on animal - based products and to compare worker welfare and animal welfare preferences, there remain a number of unanswered questions. For inst ance, it is unclear why consumers prefer animal welfare labels over worker welfare labels. Qualitative studies exploring this question may provide additional insights to individuals pondering worker welfare labeling programs regarding their viability and c ompetitiveness in food markets. Furthermore, another labeling system being pondered in regards to worker welfare is domestic fair trade. Future studies should investigate whether domestic fair trade may garnish a greater premium than the fair labor label i ntroduced here, or its contemporaries in the Equitable Food Initiative and Fair Food Program. Finally, as it was revealed that consumers are willing to pay a premium for 77 animal - based food products with a label certifying worker conditions, it is necessary to review the current state of working conditions in these industries, identify areas for improvement, and estimate the costs associated with implementing such changes in comparison to the potential premium. From such studies, farms may find that they have already adopted recommended practices that they may be able to capitalize on in the market if communicated to consumers either through adoption of a labeling system, or perhaps via other marketing means. More sion to farmers of the social license to produce has evolved to incorporate working conditions, and if so, which conditions are expected and how consumers want this information to be monitored and reported. 78 APPENDIX 79 A PPENDIX Exhibit A 2.1 Egg Choice Experiment Directions Egg Choice Questions (Seen by all treatments) In this section you will be faced with 8 choice questions about eggs. Before recording your responses, carefully read the following information which will ass ist you in completing the questions that follow about egg purchase decisions. Note that the button to advance is set to appear based on a timer to encourage your thoughtful reading. Each question presents two different dozen Brown Large Grade A egg produ cts and a no - purchase option. The egg products vary with regard to price and the presence or absence of two labels (Fair Labor and Accredited Cage Free). Fair Labor Accredited Cage Free All other potential egg attributes that are not explicitly reported in the product profiles (questions) are identical across options. For each choice question, please choose the egg product you would prefer to purchase. Alternatively you may choose NOT TO PURCHASE any product. Please carefully examine each option before you make a decision and choose the product that you most prefer. Before you proceed, we would like to remind you that although the egg questions are hypothetical ( that is, you will not actually hav e to pay for the product), you should answer as if you were actually buying the product at a retailer. Thus, before making your selection, consider whether you would actually be willing to pay the listed price, meaning that you would no longer have that am ount available for purchases. Also, keep in mind that the results of this survey will be available to farmers, producers, retailers, and policymakers, as well as to the wider general public of consumers. This means that this survey could affect the decisio ns of farmers, producers, retailers, and policymakers regarding new product adoption. 80 Exhibit A 2.1 (cont d) Egg Low information (Seen by Treatments 2 and 3) With the above in mind, please read carefully the meaning of t he labels that will be presented and a relevant news article. Animal products with the Fair Labor label are produced from farms that maintain an established minimum level of protection for the human rights of workers. A few examples of these rights includ e receipt of a set minimum wage and provision of proper safety equipment during production. Compliance is verified by employees from other participating farms. Animal products with the Accredited Cage Free label are produced by laying hens that are raised uncaged and are free to walk, nest, and engage in other natural behaviors. Farmers assert production methods meet such standards however there is no third party compliance verification. 81 Exhibit A 2.1 (cont d) Egg Explicit Information (Seen by Treatment 3) 82 Figure A2.1 Example Egg Choice Question Table A2.1 Cholesky Matrix from Model - C estimates, Eggs Study AW WW Err. Comp AW 1.01180 WW 0.68069 0.33489 Err.Comp. 1.93096 - 2.77504 1 .67482 *Parameters in bold are statistically significant at the 95% level or better. Table A2.2. Cholesky Matrix from Model - L estimates, Eggs Study AW WW Err. Comp AW 1.43733 WW - 0.34285 0.58470 Err.Comp. - 0.34285 - 2.13458 2.87279 * Parameter s in bold are statistically significant at the 95% level or better. Table A2.3 Cholesky Matrix from Model - LM estimates, Eggs Study AW WW Err. Comp AW 1.64782 WW - 0.52842 1.11328 Err.Comp. - 0.24443 0.86399 2.59222 * Parameters in bold are stati stically significant at the 95% level or better. 83 Figure A2. 2 Example Gallon Milk Choice Question 84 Exhibit A 2.2 Milk Choice Experiment Directions Milk Choice Experiment Directions (Seen by all treatments) Milk Choice Questions In this section you wil l be faced with 8 choice questions about milk. Before recording your responses, carefully read the following information which will assist you in completing the questions that follow about milk purchase decisions. Note that the button to advance is set to appear based on a timer to encourage your thoughtful reading. Each question presents two different milk products and a no - purchase option. The milk products vary with regard to price and the presence or absence of two labels (Fair Labor and Animal Wel fare Approved). Fair Labor Animal Welfare Approved All other potential milk attributes that are not explicitly reported in the product profiles (questions) are identical across options. For each choice qu estion, please choose the milk product you would prefer to purchase. Alternatively you may choose NOT TO PURCHASE any product. Please carefully examine each option before you make a decision and choose the product that you most prefer. Before you proceed , we would like to remind you that although the milk questions are hypothetical ( that is, you will not actually have to pay for the product), you should answer as if you were actually buying the product at a retailer. Thus, before making your selection, co nsider whether you would actually be willing to pay the listed price, meaning that you would no longer have that amount available for purchases. Also, keep in mind that the results of this survey will be available to farmers, producers, retailers, and poli cymakers, as well as to the wider general public of consumers. This means that this survey could affect the decisions of farmers, producers, retailers, and policymakers regarding new product adoption. 85 Exhibit A 2.2 (cont d) Milk Low information (Seen by Treatments 2 and 3) With the above in mind, please read carefully the meaning of the labels that will be presented and a relevant news article. Animal products with the Fair Labor label are produced from farms that ma intain an established minimum level of protection for the human rights of workers. A few examples of these rights include receipt of a set minimum wage and provision of proper safety equipment during production. Compliance is verified by employees from oth er participating farms. Animal products with the Animal Welfare Approved label are produced by animals that are raised outdoors on pasture or range for their entire lives. Farmers use additional high - welfare farming practices to meet animal care sta ndards, developed in collaboration with scientists, veterinarians, researchers, and farmers from across the globe. Compliance is verified with farm visits at least once a year by independent trained auditors. 86 Exhibit A 2.2 (cont d) Milk Explicit Information (Seen by Treatment 3) 87 Table A2.4 Cholesky Matrix from Model - C estimates, Milk Study AW WW Err. Comp AW 1.13993 WW - 0.46763 0.92896 Err.Comp. 0.41383 0.06097 2.29378 * Parameters in bold are st atistically significant at the 95% level or better. Table A2.5 Cholesky Matrix from Model - L estimates, Milk Study AW WW Err. Comp AW 1.76732 WW - 41398 1.38226 Err.Comp. - 0.31690 0.54585 1.99373 * Parameters in bold are statistically significan t at the 95% level or better. Table A2.6 Cholesky Matrix from Model - LM estimates, Milk Study AW WW Err. Comp AW 1.37926 WW 0.54110 1.34154 Err.Comp. - 1.19176 0.31598 2.73782 *Parameters in bold are statistically significant at the 95% level o r better. 88 REFERENCES 89 REFERENCES Aarnio, K., Lindeman, M., 2004. Magical food and health beliefs: a portrait of believers and functions of the beliefs. Appetite 43, 65 74. https://doi.org/10.1016/j.appet.2004.03.002 Akaichi, F., Glenk, K., Revoredo - Giha, C., 2019. Could animal welfare claims and nutritional information boost demand for organic m eat? Evidence from non - hypothetical experimental auctions. Journal of Cleaner Production 961 970. https://doi.org/10.1016/j.jclepro.2018.1 0.064 Andersen, L.M., 2011. Animal Welfare and Eggs Cheap Talk or Money on the Counter? Journal of Agricultural Economics 62, 565 584. https://doi.org/10.1111/j.1477 - 9552.2011.00310.x ood quality labels: the case of the European geographic indication and organic farming labels. International Journal of Consumer Studies 36, 158 - 165. Auger, P., Devinney, T.M., Louviere, J.J., 2007. Using Best Worst Scaling Methodology to Investigate Consu mer Ethical Beliefs Across Countries. J Bus Ethics 70, 299 326. https://doi.org/10.1007/s10551 - 006 - 9112 - 7 Bazzani, C., Caputo, V., Nayga, R.M., Canavari, M., 2017. local versus organic food using a non - hypothetical choic e experiment: Does personality matter? Food Quality and Preference 62, 144 - 154. Bazzani, C., Caputo, V., Nayga, R.M., Canavari, M., 2017. Testing Commitment Cost Theory in Choice Experiments. Economic Inquiry 55, 383 396. https://doi.org/10.1111/ecin.12377 Bennett, R.M., 1997. Farm animal welfare and food policy. Food Policy 22, 281 288. https://doi.org/10.1016/S0306 - 9192(97)00019 - 5 Bennett, R.M., Blaney, R.J.P., 2003. Estimating the benefits of farm animal welfare legislation using the contingent valuation method. Agricultural Economics 29, 85 98. https://doi.org/10.1016/S0169 - 5150(03)00037 - 9 Attribute Inferences on the Relationships between Product Claims, Percei ved Product Healthfulness, and Purchase Intentions. Journal of the Academy of Marketing Science 45, 698 719. Bhat, C.R., 2003. Simulation estimation of mixed discrete choice models using randomized and scrambled Halton sequences. Transportation Research Pa rt B: Methodological 37, 837 855. 90 Brooks, K., Lusk, J.L., 2010. Stated and Revealed Preferences for Organic and Cloned Milk: Combining Choice Experiment and Scanner Data. American Journal of Agricultural Economics 92, 1229 1241. Caliendo, M., Fossen, F.M., Kritikos, A.S., 2009. Risk attitudes of nascent entrepreneurs new evidence from an experimentally validated survey. Small Bus Econ 32, 153 167. https://doi.org/10.1007/s11187 - 007 - 9078 - 6 Caputo, V. 2020. ceptance of New Food Technologies? The case of irradiated beef in South Korea under a new labeling system and across different information regimes. Australian Journal of Agricultural and Resource Economics 59 : 1 31 . Caputo, V., Lusk, J.L., Nayga, R.M., 2020 . Am I Getting a Good Deal? Reference Dependent Decision Making When the Reference Price Is Uncertain. American Journal of Agricultural Economics 102, 132 - 153. https://doi.org/10.1093/ajae/aaz042 Caputo, V., Lusk, J.L., Nayga, R.M., 2018. Choice experiments are not conducted in a vacuum: The effects of external price information on choice behavior. Journal of Economic Behavior & Organization 145, 335 351. https://doi.org/10.1016/j.jebo.2017.11.018 Caputo, V., Nayga, R.M., Scarpa, R., 2013. Food miles or carbon emissions? Exploring labelling preference for food transport footprint with a stated choice study. 57, 465 - 482. Caputo, V., Scarpa, R., Nayga Jr., R.M., 2017. Cue versus independent food attributes: the effect of adding attributes in choice experiments. European Review of Agricultural Economics 44, 211 230. https://doi.org/https://doi - org.proxy1.cl.msu.edu/10.1093/erae/jbw022 Caputo, V. Vassilopoulos, A., Nayga, R.M., Canavari, M., 2013. Welfare Effec ts of Food Miles Labels. The Journal of Consumer Affairs 47, 311 - 327. Caswell, J . A . , and Mojduszka , E.M., Market for Quality in Food Products . American Journal of Agricultural Economics 78 , 1248 12 53 . Chang, J.B., Lusk, J.L., Norwood, F.B., 2010. The Price of Happy Hens: A Hedonic Analysis of Retail Egg Prices. Journal of Agricultural and Resource Economics 35, 406 423. Chang, J.B., Lusk, J.L., Norwood, F.B., 2009. How Closely Do Hypothetical Surveys an d Laboratory Experiments Predict Field Behavior? American Journal of Agricultural Economics 91, 518 534. Coalition for Sustainable Egg Supply, n.d. Final Research Results Report. Conner, D., Irwin, E., Simmen, M., 2018. Exploring demand for eco - labeled dai ry product attributes in Vermont. British Food Journal 120, 2857 2867. https://doi.org/10.1108/BFJ - 05 - 2018 - 0305 91 Costanigro, M., McFadden, D.T., Kroll, S., Nurse, G., 2011. An in - store valuation of local and organic apples: the role of social desirability. Agribusiness 27, 465 477. https://doi.org/10.1002/agr.20281 Crespi, J.M., Lusk, J.L., Cherry, J.B.C., Martin, L.E., Mcfadden, B.R., Bruce, A.S., 2016. Neural Activations Associated with Decision - Time and Choice in a Milk Labeling Experiment. American Journ al of Agricultural Economics 98, 74 91. https://doi.org/10.1093/ajae/aav049 de - Magistris, T., Gracia, A., Nayga, R.M., 2013. On the Use of Honesty Priming Tasks to Mitigate Hypothetical Bias in Choice Experiments. American Journal of Agricultural Economics 95, 1136 1154. Doyon, M., Bergeron, S., Cranfield, J., Tamini, L., Criner, G., 2016. Consumer Preferences for Improved Hen Housing: Is a Cage a Cage?: PRIORITY EFFECTS ON PREFERENCES FOR HEN HO U.S. ING. Canadian Journal of Agricultural Economics/Revue cana dienne 751. https://doi.org/10.1111/cjag.12113 Drichoutis, A.C., Vassilopoulos, A., Lusk, J.L., Nayga, R.M., 2017. Consumer preferences for fair labour certification. European Review of Agricultural Economics 44, 455 474. https://doi .org/10.1093/erae/jbx002 Gerini, F., Alfnes, F., Schjoll, A., 2016. Organic - and Animal Welfare - labelled Eggs: Competing for the Same Consumers? J. Agric. Econ. 67, 471 490. https://doi.org/10.1111/1477 - 9552.12154 Gifford, K., and J.C. Bernard. 2011. The E Pay for Natural and Organic Chicken. International Journal of Consumer Studies 35 , 282 89. Greene, W.H., Hensher, D.A., 2003. A latent class model for discrete choice analysis: contrasts with mixed logit. T ransportation Research Part B: Methodological 37, 681 698. Heng, Y., Peterson, H.H., 2018. Interaction Effects among Labeled Attributes for Eggs in the United States. Journal of International Food & Agribusiness Marketing 30, 236 250. https://doi.org/10.10 80/08974438.2017.1413610 Heng, Y., Peterson, H.H., Li, X., 2013. Consumer Attitudes toward Farm - Animal Welfare: The Case of Laying Hens. Journal of Agricultural and Resource Economics 38, 418 434. Howard, P.H., 2006. Central Coast consumers want more food - related information, from safety to ethics. California Agriculture 60, 14 19. Howard, P.H., Allen, P., 2010. Beyond Organic and Fair Trade? An Analysis of Ecolabel Preferences in the United States. Rural Sociology 75, 244 269. https://doi.org/10.1111/j.154 9 - 0831.2009.00009.x 92 from the United States. Renewable Agriculture and Food Systems; Cambridge 23, 235 242. https://doi.org/http://dx.doi.org.proxy1.cl.msu.edu/10 .1017/S1742170508002275 Howard, P.H., Allen, P., 2006. Beyond organic: consumer interest in new labelling schemes in the Central Coast of California. International Journal of Consumer Studies 30, 439 451. https://doi.org/10.1111/j.1470 - 6431.2006.00536.x Ke ndall, H.A., Lobao, L.M., Sharp, J.S., 2006. Public Concern with Animal Well - Being: Place, Social Structural Location, and Individual Experience*. Rural Sociology; Columbia 71, 399 428. https://doi.org/http://dx.doi.org.proxy2.cl.msu.edu/10.1526/0036011067 78070617 Karcher, D.M., Jones, D.R., Abdo, Z., Zhao, Y., Shepherd, T.A., Xin, H., 2015. Impact of commercial housing systems and nutrient and energy intake on laying hen performance and e gg quality parameters. Poultry Science 94, 485 501. https://doi.org/1 0.3382/ps/peu078 Lancaster, K.J., 1966. A New Approach to Consumer Theory. Journal of Political Economy 74, 132 157. Product: A Nonhypothetical Choice Experiment 6 3, 563 - 582. Lin, W., Ortega, D.L., Caputo, V., 2019. Are Ex - Ante Hypothetical Bias Calibration Methods Context Dependent? Evidence from Online Food Shoppers in China. Journal of Consumer Affairs 53, 520 544. https://doi.org/10.1111/joca.12202 Lusk, J.L., 2 019. Consumer preferences for cage - free eggs and impacts of retailer pledges. Agribusiness 35, 129 148. https://doi.org/10.1002/agr.21580 Lusk, J.L., House, L.O. Valli, C., Jaeger, S.R., Moore, M. J. L. Morrow, and W. B. Traill. 2004. Effect of Information about Benefits of Biotechnology on Consumer Acceptance of Genetically Modified Food: Evidence from Experimental Auctions in the United States, England, and France. European Review of Agricultural Economics 31 , 179 204 Lusk, J.L., J. Roosen, and J. Fox. , 2 003. Demand for Beef from Cattle Administered Growth Hormones or Fed Genetically Modified Corn: A Comparison of Consumers in France, Germany, the United Kingdom, and the United States. American Journal of Agricultural Economics 85 . 16 29. Lusk, J.L., Schr oeder, T.C., 2004. Are Choice Experiments Incentive Compatible? A Test with Quality Differentiated Beef Steaks. American Journal of Agricultural Economics 86, 467 482. McFadden, D., 1974. ''Conditional Logit Analysis of Qualitative Choice Behavior''. In P . Zarembka (Ed.), Frontiers in E conometrics (pp. 105 142). New York: Academic Press. 93 McFadden, J.R., and W.E. Huffman. 2017. Willingness - to - Pay for Natural, Organic, and Conventional Foods: The Effects of Information and Meaningful Labels. Food Policy 68 , 2 14 32. Norwood, F.B., Lusk, J.L., 2011. A calibrated auction - conjoint valuation method: Valuing pork and eggs produced under differing animal welfare conditions. Journal of Environmental Economics and Management 62, 80 94. https://doi.org/10.1016/j.jeem.20 11.04.001 Ochs, D.S., Wolf, C.A., Widmar, N.J.O., Bir, C., 2018. Consumer perceptions of egg - laying hen housing systems. Poultry Science 97, 3390 3396. https://doi.org/10.3382/ps/pey205 Olynk, N.J., Tonsor, G.T., Wolf, C.A., 2010. Consumer Willingness to P ay for Livestock Credence Attribute Claim Verification. Journal of Agricultural and Resource Economics 35, 261 280. Onozaka, Y., Mcfadden, D.T., 2011. Does Local Labeling Complement or Compete with Other Sustainable Labels? A Conjoint Analysis of Direct an d Joint Values for Fresh Produce Claim. American Journal of Agricultural Economics 93, 693 - 706. https://doi.org/10.1093/ajae/aar005 Paul, A.S., Lusk, J.L., Norwood, F.B., Tonsor, G.T., 2019. An experiment on the vote - buy gap with application to cage - free eggs. Journal of Behavioral and Experimental Economics 79, 102 109. https://doi.org/10.1016/j.socec.2019.02.005 Applia nce Efficiency Level. The Review of Economics and Statistics 80, 647 657. https://doi.org/10.1162/003465398557735 Rolfe, J., 1999. Ethical Rules and the Demand for Free Range Eggs. Economic Analysis and Policy 29, 187 206. https://doi.org/10.1016/S0313 - 592 6(99)50020 - 9 Rousu, M., Huffman, W.E., Shogren, J.F., Tegene, A., 2007. Effects and Value of Verifiable Information in a Controversial Market: Evidence from Lab Auctions of Genetically Modified Food. Economic Inquiry 45, 409 432. https://doi.org/10.1111/j. 1465 - 7295.2007.00017.x Scarpa R, Campbell D , Hutchinson G (2007) Benefit estimates for landscape improvements: Sequential Bayesian design and respondents rationality in a choice experiment. Land Economics 83(4), 617 634. Scarpa, R., Ferrini, S., Willis, K ., 2005. Performance of Error Component Models for Status - Quo Effects in Choice Experiments, in: Scarpa, R., Alberini, A. (Eds.), Applications of Simulation Methods in Environmental and Resource Economics, The Economics of Non - Market Goods and Resources. S pringer Netherlands, Dordrecht, pp. 247 273. 94 Scarpa, R., Thiene, M., Train, K. 2008. Utility in Willingness to Pay Space: A Tool to Address Confounding Random Scale Effects in Destination Choice to the Alps. Amer. J. Agr. Econ. 90, 994 - 1010. Scheiber, N. The New York Times. Tonsor, G.T.,Olynk, N., and Wolf, C. 2009. Consumer Preferences for Animal Welfare Attributes: The Case of Gestation Crates. Journal of Agricultural and Ap plied Economics, 41 . 713 730. Train, K.E., 2009. Discrete Choice Methods with Simulation, 2nd ed. Cambridge University Press. for Retail: Weekly Advertised Poult https://www.marketnews.usda.gov/mnp/py - report - retail?runReport=true&portal=py&startIndex=1&commodity=EGGS&class=Retail&rep Type=summary®ion=ALL&type=ALL&state=ALL&description=ALL&dataType=N one+Selected&operator=%3 C&dataValue=&repDate=12%2F01%2F2017&endDate=11 %2F23%2F2018 https://www.marketnews.usda.gov/mnp /da - report - retail?repType=wiz&run=Run&portal=da&locChoose=locState&commodityClass=allco mmodity&startIndex=1&type=retail&class=MILK&commodity=ALL&packSize=GALL ON®ion=ALL&organic=ALL&repDate=12%2F01%2F2017&endDate=11%2F23%2 F2018&compareLy=No Valente, M., Chaves, C., 2018. Perceptions and valuation of GM food: A study on the impact and importance of information provision. Journal of Cleaner Production 172, 4110 4118. https://doi.org/10.1016/j.jclepro.2017.02.042 Van Loo, E.J., Caputo, V., Nayga Jr., R.M., M willingness to pay for organic chicken breast: Evidence from choice experiment 22, 603 - 613. Van Loo, E.J., Caputo, V., Nayga Jr., R.M., Seo, H., Zhang , B., Verbeke, W., 2015. Sustainability labels on coffee: Cons umer preferences, willingness - to - pay and visual attention to attributes 118, 215 - 225. sustainability labels on meat. Food Policy 49, 137 150. https://doi.org/10.1016/j.f oodpol.2014.07.002 95 Van Wezemeal, L., Caputo, V., Nayga, R.M., Chryssochoidis, G., Verbeke, W., 2014. European consumer preferences for beef with nutrition and health claims: A multi - country investigation using discrete choice experiments 44, 167 - 176. Wolf, C.A., Tonsor, G.T., Olynk, N.J., 2011. Understanding U.S. Consumer Demand for Milk Production Attributes. Journal of Agricultural and Resource Economics 36, 326 342. 96 ESSAY 3: WHICH CAME FIRST: TH E CHICKEN (COW) OR T HE LABORER? THE DES TOWARD ANIMAL AN D WORKER WELFARE 3.1 Introduction Via their purchase behavior consumers grant a social license to food producers. If food does not meet their social acceptability requirements consumers will not purchas e the product. Over time, it seems that the number of social responsibility attributes has increased including production practices like organic and related to animal welfare, as well as organic, locally sourced and others. Recently, in addition to these v aried policy areas, expectations have been introduced regarding the treatment of agricultural workers. For example, even before moving toward less purchase tomatoes fr om farms with certified working conditions (Fair Food Program, n.d.). Aramark and Walmart made similar worker welfare commitments before select public animal welfare stances (Aramark, 2015; Walmart, 2015; Fair Food Program, n.d..). As the Fair Food Program label and its contemporaries like the Equitable Food Initiative become more common in food markets and some writers suggest there are poor conditions working conditions for agricultural workers may grow (Greenhouse, 2015) . A gap in the literature is whether the public views animal and worker welfare as complementary concepts both striving to improve the wellbeing of living creatures while monitoring and enforcing ethical production practices in agriculture or perceive a trade - off between the two types of farm practices. We seek to fill this gap. Beyond public perception, there is some evidence suggesting an explicit tradeoff between animal and worker wellbeing in the implementation of farm practices. For example, in poultry pr oduction, physiologically cage - free systems are better for hens than aviaries, but worker respiratory systems are taxed more in 97 cage - free environments (Coalition for Sustainable Egg Supply, n.d.). D espite the plethora of research on non - cage production sys tems in the poultry industry both in terms of their impact on systems (Chang, et al. 2010; Heng, et al. 2013; Lusk 2019; Lusk 2010; Malone and Lusk 2016), discuss ion o n the worker health impacts in such environments remains sparse (Coalition for Sustainable Egg Supply, n.d.). 24 More generally, compared to debates around animal welfare, attention to worker welfare in animal production is lacking. T here have been a n umber of ballot initiatives related to animal welfare practices in agriculture, such as Proposition 2 in California, but no similar ballot initiatives related to agricultural labor to date in the U.S. . Nevertheless, examples like the work by Migrant Justic e, including the adoption of the Milk with Dignity suggests some individuals opine that potential areas for improvement regarding an agr icultural worker population exist (Scheiber 2017). To address the lack of information o n cons impacting practices in the farm industry, we surveyed over 1,300 U.S. individuals. A best - worst scaling (BWS) approach was utilized to measure rences for various animal and worker welfare enhanc ing farm practices in the dairy and poultry industries. The dairy industry was selected because it is one of the most contemporaneous sectors discussing worker welfare conditions; the poultry industry was selected because of the potential explicit tradeoff between animal and worker welfare mentioned previously. Selected farm practices in these sectors focused on breaks and meals, third - party verification of conditions, and varied health 24 Note that for the purposes of this study we are assuming cage - free systems to be better for hen health. However, we do acknowledge that there is no consensus on this issue. For example, mortality rates may be higher in cage - free environmen ts and producers have a number of questions to address to make this system as productive and healthy as others ( 5 questions about cage - free hen health, welfare 2017) . 98 concerns for both an imals and workers. Additionally, two policies that simultaneously improve animal and worker welfare, training of workers (dairy) and worker to animal ratios (poultry) , were also examined. Exploring these policies with a BWS experiment allowed for a relative ranking of both worker welfare and animal welfare practices. Understanding how the U.S. public makes trade offs between animal and worker welfare enhancing practices is a question of great interest for various stakeholders, including farmers and policymakers. For instance, up to now, little attention ha s been paid to what worker welfare practices are considered most important to consumers. Furthermore , questions on whether producers should continue to prioritize animal welfare practices over worker welfare practices remain unanswered. It is important for producers to know the welfare practices most important to consumers to ensure their products comply with such standards and remain in demand given . This is because as consumers the public advocates for production pract ices with the purchases they make; they may boycott products they believe are produced unethically and/or be willing to pay a price premium for verification that specific practices are used . In addition, producers may be limited in their ability to economi cally implement all possible animal and worker welfare enhancing practices without pricing themselves out of the market. Hence, evidence on what animal and worker welfare practices are most important to consumers is crucially important to make informed far m decisions. Further, policymakers must decide on labor standards in laws s uch as the Fair Labor Standards Act . These laws also at times exclude agriculture, like the overtime pay requirement in the Fair Labor Standards Act. Thus, policymakers need to unde rstand not only broad opinion on working condition regulations, but also specifics of agricultural labor. Such decisions ensure worker safety while balancing the need for continued vitality of industry production. It is possible that 99 stronger enforcement o f worker breaks, which is one practice explored in this study, is valued by the public and shou ld be instituted. Without asking the ir constituents for their opinions on working conditions in agriculture, however, policymakers are operating with impartial i nformation. We contribute to the existing literature by introducing worker welfare enhancing practices. While there are several studies focusing on animal welfare improvements (Bennett 1997; Bennett and Blaney 2003; Ellison, et al. 2013; Heng, et al. 201 3; McKendree, et al. 2018; McKendree, et al. 2014; Wolf and Tonsor 2017; Wolf and Tonsor 2013), this is the first study to our knowledge that explores worker welfare improvements. S econd, not only does this study facilitate discussion for this nascent topi c area, but it situates that discussion in the current public discourse on related farm practices. A separate ongoing discussion in research is the need for replication studies to act as robustness checks and validate existing research. By including animal welfare enhancing practices that have been explored in previous studies, we can comment on how the relative ranking of those farm practices is maintained or varies in this study. S imilar rankings would strengthen the evidence in support of practice rankin gs while divergent rankings would suggest areas for future research in replicating animal welfare practice rankings, particularly in the context of other industry relevant topics like worker welfare or environmental sustainability. Our results indicate th at there are approximately four defined groups in each of the dairy and poultry samples with distinct preference structures. The largest group, containing over 50% of the sample population, in the dairy study ranked training programs for workers as the mos t important farm practice . For the poultry industry, the most important farm practices were treatment of sick animals and hen access to food and water. 100 The paper proceeds as follows . The first section describes the context and the farm policy selection in first the dairy industry and then the poultry industry . The second section describes the survey and research methodology, while section three summarizes the data. Next, we discuss the results beginning with the dairy study and finishing with the poultry st udy. Then we discuss the implications of the results before summarizing them in the conclusion. 3.2 Background and Farm Practice Identification P rior studies have explored preferences for animal welfare practices in the dairy (Ellison, et al. 2013; Wolf a nd Tonsor , 2017; Wolf and Tonsor , 2013) and poultry (Bennett , 1997; Bennett and Blaney , 2003; Ellison, et al. 2013; Heng, et al. 2013) in dustries. However, no prior research has improve both animal and worker welfare. This study considers this association within the context of the dairy and poultry industries . The dairy industry faces controversy surrounding the working conditions on some farms (Scheiber 2017 ; Greenhouse, 2015 ). Due in part to these n ational discussions, the National Dairy Farm Program extended their areas of concentration to include workforce development in 2018 (National Dairy Farm Program 2020). As they continue to refine what practices should be recommended i n this program and as producers decide whether it is a program they would like to participate in, it is advantageous to have public opinion data on worker welfare farm practices in the dairy industry. The poultry industry has not received the same type of national attention regarding its working conditions Proposition 2, are commonly up for debate. Therefore, the poultry industry serves as an interesting context for understanding how worker welfare concerns may fit within already well - 101 established concerns for animal welfare. Additionally, prior research suggests explicit tradeoffs may exist between animal and worker welfare in the poultry industry . For example , the Coalition for Sustainable Egg Supp ly (n.d) posits that cage - free aviary systems improve hen health but can depress worker respiratory health compared to conventional caged systems. As such we used these two industries and formulated farm animal and worker welfare practices for each. The fo llowing subsections describe the farm practice selection in each industry. 3.2.1 Dairy Industry While the initial experiment idea was inspired by Coalition for Sustainable Egg Supply (n.d.) and Scheiber (2017), the practical work of choosing the farm pr actices began with the dairy industry animal welfare practices reported in Wolf and Tonsor (2017). Using their nine animal welfare practices as a base , we brainstormed parallel worker welfare practices to elicit tradeoffs between animal welfare and worker welfare enhancements. Not all animal welfare practices were deemed to have worker welfare equivalents, while one, training of farm workers in cow handling was deemed to benefit both animals and workers. The final list was reduced to nine farm practices li sted in Table 3.1 to keep the BWS experiment manageable for participants . The nine farm practices include four animal welfare practices, four worker welfare practices, and one practice benefiting both parties. The four practices primarily benefiting a sing le party (animal or workers) can be categorized as pertaining to i) breaks, ii) third - party verification, iii) treatment of sick, and iv) health plans . 102 Table 3.1 Animal and Worker Welfare Practices in the Dairy Industry Included in this Study These farm practices were not chosen just because of their relationship to animal and worker welfare but also for their individual merits. Prior research has indicated that cow access to outdoor spaces is desirable to consumers (Van Loo et al. 2014; Wolf, et al. 2011) and it is a condition certified by some labeling programs, such as organic and animal welfare approved. As an effective means of maintaining performance, managing fatigu e and controlling the federal law in the U.S. does not require employers to provide employees lunch or coffee breaks ( Breaks and Meal Periods n.d.), and less than half of all states have adopted such policies ( Minimum Length of Meal Period Required under State Law for Adult Employees in Private Sector 1 2020). Animal Welfare Practice Worker Welfare Practice Breaks AW_breaks: All cattle must have access to outdoor exercise areas for at least 4 hours per day, weather permitting. WW_breaks: All workers are provided paid 15 minute breaks for every 4 hours worked, and a half hour (meal) break between each 4 hour shift. Third - Party Verification AW_ver: A third - party verifies that appropriate cow care and facilities are provided on farm. WW_ver: A third - party verifies that appropriate human resource management and w orking conditions are provided on farm. Treatment of Sick AW_sick: Sick cows are promptly treated or euthanized. WW_sick: Workers are paid sick time off. Health Plans AW_health: A herd health plan is developed with the help of a veterinarian. WW_h ealth: Workers are provided medical insurance. Training Training: There is a consistent training program for owners and workers focusing on principles of cow care and handling that increase both animal and worker welfare. 103 The second group of welfare practices, third party verification is not specific to a law or government actio n, say as with organic labeling being certified by the US DA, but rather is an option available in the private sector that may help strengthen consumer trust (Wolf and Tonsor 2017). Hence, they may serve as vehicle to boost the demand for food produced under certain welfare practices. To this end, third - party verification is often a component of animal welfare labeling programs, such as Certified Humane, and worker welfare labeling programs, like the Fair Food Program. There are a number of farm practices regu lated by certification programs. Many such animal welfare programs include animal health and sickness provisions. Part safety. For example, undercover dairy f arm animal welfare videos have generated outcry at the treatment of some sick and lame animals (Wolf and Tonsor 2017). The treatment of sick and health plan farm practices that we have included help address these ongoing concerns 25 . Providing paid sic k ti me off was deemed analogous to promptly treating injured or sick cows as it allows workers a degree of financial security as they proactively treat illness. Finally, as previously alluded to, the joint farm practice was worker training. Consistent training f ocusing on cow care and handling was deemed to not only enhance the humane treatment of cattle but to increase the safety and efficiency for workers. 25 The herd health plan was the only animal welfare practice included in this study but no t Wolf and Tonsor (2017). It was included in McKendree, et al. (2018). It was chosen here for its similarity to providing health insurance (which explains and reduces the costs of human health maintenance), which has been a topic hotly debated at the natio nal and state levels. 104 3.2.2 Poultry Industry Once the animal and worker welfare practices in the dairy industry were det ermined we sought analogous practices in poultry farm practices, aiding preference comparisons between animal and worker welfare across industries. Since prior studies have found that consumers are willing to pay varied amounts for animal welfare enhanceme nts for different animals ( see Clark, et al. (2017 ) f or a meta - analysis of such studies including a discussion of WTP variations across animals) it is possible that they value the relative welfare of animals and workers differently across animals as well. In particular, cage - free systems, which some studies find better for hens, have been found to depress air quality for workers, providing an expl icit trade - off between the welfare of animals versus workers (Coalition for Sustainable Egg Supply, n.d.); but, the authors are not aware of such an explicit trade - reveal a recognition of this explicit tradeoff? Which do they value more, animal or worker welfare? To explore these questions in poultry we maintained five of the farm practices from the cow study (pertaining to third - party verification , treatment of the sick and worker break s ) ( T able 3.2). 105 Table 3.2 Animal and Worker Welfare Practices in the Poultry Industry Included in this Study These five farm practices were deemed the most consistent across both the dairy and poultry industries. For example, the benefit of breaks was reimaged for its more basic benefit of ensuring a meal break for workers. Then access to feed and wate r was analogous for chicken welfare. Thus, rather than comparing WW_breaks/WW_meals to WW_breaks (Outdoor Access in Wolf and Tonsor 2017) we focus less on how time is used and more on the purpose and needs driving that break time (Feed and Water in Wolf an d Tonsor 2017). Next, we wanted to be certain to include a farm practice pair that would explore the potential tradeoff between improvements for hens and damage for workers via the cage - free system. Note that since all farm practices were phrased as improv ements to either animals or workers we did not mention the research and potential negative human effects of the AW_cage variable either in its description or as the negative of that practice for worker welfare practice, i.e. Individual battery cages are us ed Animal Welf are Practice Worker Welfare Practice Meals AW_meals: Hens have constant access to food and water WW_meals: Workers are provided paid 15 minute breaks for every 4 hours worked, and a half hour (meal) break between each 4 hour shift. Third - Party Verifi cation AW_ver: A third - party verifies that appropriate hen care and facilities are provided on farm WW_ver: A third - party verifies that appropriate human resource management practice and working conditions are provided on farm Treatment of Sick AW_sic k: Sick animals are promptly treated or euthanized WW_sick: Workers are paid sick time off Cage System AW_cage: An aviary or free - range housing system is used which does not constrain hens to individual or small - group cages. WW_ cage: Workers are provi ded proper respiratory (safety) protection (a N95 mask or respirator) Flock size Flock_size: Flock size is not increased without space and staffing capacities within determined ratios, which not only ensures hen space but also restricts the burdens on workers. 106 to reduce airborne dust and pathogens for workers. Rather, we chose to frame the worker welfare improving system as respiratory equipment, which would in all likelihood improve worker conditions in any poultry system, but particularly as a way of improv ing upon a perhaps already adopted cage - free system on behalf of workers. Finally, the joint beneficiary farm practice was replaced from Training to Flock_size . A consistent flock size to worker ratio theoretically enhances animal wellbeing by ensuring pro per care but also prevents the workload from being overwhelming for workers. 3.3 Data and Methods To determine the relative importance the U.S. general population places on the practices in tables 3.1 and 3.2 , we uti lized a BWS approach as it offers a nu mber of theoretical and practical benefits. For example, by asking not only which object in a set is the most preferred but also which is least preferred, BWS provides additional information on the preference ratings of individuals in an efficient manner, i.e. without requiring having to repeat questions with the remaining objects to ask which is least preferred (Louviere, et al., 2015 p.7). Further, BWS is preferred to other preference elicitation techniques as it requires respondents to make trade - offs be tween objects and represents a well - understood measure (Finn and Louviere 1990; Lusk and Briggeman 2009; Caputo and Lusk 2020; Wolf and Tonsor 2013). BWS requires tradeoffs as participants cannot choose that all farm practices are important , as with Likert scale questions . mean the sa 107 For each application (dairy and poultry), an object case (case 1) design was used in which each of the nine farm practices is considered a distinct object. Employment of a balanced incomplete block design generated twelve questions, each with six farm pract ices (see Figure 3.1 for an example choice question and Table A 3. 1 in the Appendix for the experimental design ). Each practice appeared eight times overall, while pairs of practices appear five times throughout the twelve questions. During the experiment, for each BW question , respondents were asked to h choice set. The order of the farm practices within each question and the question order were randomized across respondents. Figure 3.1 Worker and Anim al Welfare Practices in the Poultry Industry BWS Sample Question 108 The BWS experiment was embedded in a national online survey administered through Qualtrics from January - April, 2019. Adults who had purchased eggs and milk in the la st three months were elig ible to participate. Participants were randomly assigned to either a milk (dairy) or egg (poultry) survey, with each follow ing the same structure. The survey began with questions about prior shopping behavior for the product. Then respondents completed a d iscrete choice experiment on animal welfare and worker welfare labeling described in Essay 2 of this dissertation. The next section was the farm practice BWS task. Finally, the survey concluded with demographic questions, including three psychosocial scale s. 3.3.1 Online Survey Summary statistics of the basic demographics and variable definitions are reported in Table 3. 3 . In total 1,304 people participated with 762 in the dairy experiment and 542 in the poultry experiment. Table 3. 3 Summary Statistics o f Basic Demographics Variable Definition Dairy Sample Poultry Sample Total Sample U.S. Population Gender 1 if female; 0 otherwise 0.735 (0.441) A 0.730 (0.445) 0.733 (0.443) 0.508 B Age Young 1 if respondent 18 - 29 years; 0 otherwise 0.180 (0.384) 0.1 90 (0.393) 0.184 (0.388) Not available Mid - age 1 if respondent 30 - 64 years; 0 otherwise 0.682 (0.466) 0.673 (0.469) 0.679 (0.467) Not available Senior 1 if respondent 65 years or older; 0 otherwise 0.138 (0.345) 0.137 (0.344) 0.137 (0.344) 0.165 B Educat ion Low 1 if respondent does not have high school degree; 0 otherwise 0.042 (0.201) 0.022 (0.147) 0.034 (0.181) 0.122 B 109 Table 3. 3 (cont d) Mid 1 if respondent has a high school degree but not a bache otherwise 0.634 (0.482) 0.673 (0.469) 0.650 (0.477) 0.563 B High (4 year) degree or higher; 0 otherwise 0.324 (0.468) 0.304 (0.461) 0.316 (0.465) 0.315 B Income Low 1 if respondent has income be low $35,000 annually; 0 otherwise 0.543 (0.498) 0.557 (0.497) 0.549 (0.498) 0.279 C Mid 1 if respondent has income between $35,000 and $100,000 annually; 0 otherwise 0.457 (0.498) 0.443 (0.497) 0.451 (0.498) 0.417 C High 1 if respondent has income above $1 00,000 annually; 0 otherwise 0.119 (0.324) 0.114 (0.319) 0.117 (0.322) 0.304 C U.S. Census Region Northeast 1 if resides in Northeast U.S. census region; 0 otherwise 0.203 (0.402) 0.199 (0.400) 0.202 (0.401) 0.171 D Midwest 1 if resides in Midwest U. S. census region; 0 otherwise 0.253 (0.435) 0.251 (0.434) 0.252 (0.434) 0.208 D South 1 if resides in South U.S. census region; 0 otherwise 0.391 (0.488) 0.386 (0.487) 0.389 (0.488) 0.382 D West 1 if resides in West U.S. census region; 0 otherwise 0.152 (0 .359) 0.164 (0.371) 0.157 (0.364) 0.239 D Political Party Democrat 1 if respondent Democrat; 0 otherwise 0.356 (0.479) 0.351 (0.478) 0.354 (0.478) 0.33 E Republican 1 if respondent Republican; 0 otherwise 0.291 (0.455) 0.282 (0.451) 0.288 (0.452) 0.2 6 E In - group Identification Ranges from 1 (low national (American) in - group identification) to 5 (high in - group id) 3.803 (0.746) A 3.730 (0.807) 3.773 (0.773) N/A Illegal Aliens Scale - 1 if think negatively of illegal aliens; 1 if think positively of ille gal aliens; 0 if neutral - 0.278 (0.919) - 0.240 (0.933) - 0.262 (0.925) N/A A Numbers in parentheses are standard deviations B Data from U.S. Census Bureau (n.d) C Data from Semega, et al. (2019) D Data from National Population Totals and Components of Chan ge: 2010 - 2019 (2019) E Identification (2018) 110 There were more women who participated than men, which is a common occurrence in online research studies of primary shoppers ( Grebitus et a l., 2013; Lusk, 2011; Nocella et al., 2010 ). As this study was issued a t the same time as a food discrete choice experiment, the target Most respondents were between 30 an d 64 years of age. We fail to reject the hypotheses t hat the mean of seniors is dif ferent from that in the overall U.S. population (U.S. Census Bureau n.d.). per centage in the U.S. population (U.S. Census Bureau n.d.). Our sample i ncome was below national levels as only 27.9% of households had income below $35,000 and 30.4% had income over $100,000 in 2018 (Semega, et al. 2019). In the dairy sample the Northeast a nd Midwest were over - represented relative to the West; in the poultry sample the Midwest was over - represented relative to the West. In 2018 , 33% and 26% of registered voters identified as Democrats and Republicans respectively ( Wide Gender Gap, Growing Edu cational Divide in . These are approximately the same representation in our sample, except that our dairy sample has more Republicans represented. In addition to these basic demographics, respondents indicated their opinio n with various statements that allowed for the calculati on of three psychosocial scales: in - group identification, illegal aliens scale, and an animal welfare - worker welfare scale. The in - group identification scale was constructed from the nine questions u sed in Lyons, - group identification, to five, strong national in - group identific ation. Our sample tended to have a strong national, American in - group identification. Potentially , since many farmhands are immigrants, 111 sense of national in - concern about farm worker wel fare. More specifically, we anticipated that the higher the in - group identification score the less concerned the individual would be with worker welfare farm practices. The illegal aliens scale was constructed from the twenty questions used in Ommundsen, States citizens . - 10 to 10. We further reduced the scale to a three - level categorical variable where - 1 meant overall respondent looks negatively on illegal aliens, 0 meant the respondent was neutral toward illegal aliens, and 1 meant overall the respondent looks favorably toward illegal aliens. On average, our sample respondent was more negative toward i llegal aliens. Similar to the in - group identification scale, we hypothesized that respondents may conflate immigrant farm workers with illegal aliens and thus those with a low illegal aliens score would be less concerned with worker welfare farm practices. 3.3.2 Econometric Model A key feature of BWS is that it is consistent with random utility theory (RUT) . 26 RUT assumes that people choose the item that provides the g reatest utility. According to the RUT, t he probability that the respondent n selects it em j (as best) and k (as worst) out of J items in BWS question t is the probability that the difference in utility of the selected items ( and ) is greater than all other possible differences within each BWS question (Lusk & 26 It is also consistent with fixed utility or constant utility theory but such theories are more applicable to psychology than economics and will be ignored here, as differences are miniscule between both theories in this application (Louviere, Flynn, and Marley 2015 p.12). 112 Briggeman, 2009; Caputo and Lusk 2020). U tility is comprised of two components, the difference in utility between the j best and the k wo rst practices ( ) and a random error term ( ): (1) where is the vector of estimated parameters of the j best and k worst practices relative to a basel ine practice. In this application, we selected nine farm practices (described in tables 3.1 and 3.2) and respondents were asked to respond to twelve questions ( T = total number of best - worst questions = 12). Each choice question was represented by six - item s or farm practices , and WW_sick was chosen as the baseline as it had the lowest best - worst choice frequency. As there are policies in each choice question, respondents choose between most important - least important farm practice pairs, meaning that utility difference from chosen pair j - k is larger than the utility difference from other best - wors t pairs. The resulting model can be estimated using models that either assume preference homogeneity or allows preferences to vary across respondents. We use a latent class analysis (LCA) as recent studies have found preferences for animal welfare practi ces to differ wit hin a population when utilizing BWS (McKendree, et al. 2018). LCA models allow to identify the mean importance parameters shared by groups of individuals but which differ across groups. If political change is driven by the collective acti on of factions as argued by James Madison (17 87) in Federalist No. 10 , then latent cl ass analysis is helpful in identifying the bonding characteristics of groups with similar policy wants. Additionally, group size can be estimated such that the relative st rength of any preference group can be identi fied. This is constructed based on the in dividual probabilities of a respondent being in a specific class given their preferences. 113 Formally, the unconditional probability of a best - worst pair being selected give n the latent class s respondent n belongs to can be represented as follows: ( 2 ) Based on equation (2 ), parameters in the ob served portion of the utility can be estimated by maximizing the log - likelihood function. The estimated parameters from equation ( 2 ) are not readily interpretable. Thus, we calculate the share of preferences for each farm practice, . The forecasted p robability that a farm practice is picked as most important is equal to ( 3) As probabilities, the preference shares for all nine policies are positive and sum to 1 allowing for meaningful interpretation. As the preference share is computed with a ratio scale, if the share of preferences is twice as large for one farm practice as for another, then that farm practice is twice as preferred. This also means that if all farm practices were equally valued that we preference share is below 0.111, we can say that the policy is generally deemed less important than the other practices with shares above 0.111. In each application (dairy and poultry), following Caputo and Lusk (2020) and McKendree et al (2018) the preference shares of each practice were computed by using the Krinsky & Robb (1986) bootstrapping method employing 1,000 draws from multivariate normal distributions , which also allows for the construction of confidence intervals . Additionally, class membership probabilities can be estimated for each individual from the LCM. Participants can then be sorted into the classes identified in the LCM based on whether their 114 likelihood of being in a particular class was greater than 0.50. 27 Then demographic and psychosocial characteristics a cross classes were compared using t - tests . 3.4 Results In this section we report the estimates from the LC M s utilized to estimate the data from both th e dairy and poultry applications. In each application, the optimal number of classes in the LC models was selected in accordance with the usual information criteria for non - nested models such as the Akaike Information Criteria (AIC), modified Akaike Inform ation Criteria (3AIC), and the Bayesian Information Criteria (BIC). Additionally, attention was paid to the cluster size to ensure each group represented a sizeable portion of the population. As can be shown in Tab l e 3. 4 the criteria continue to decrease a cross the four - model specificat ion. However, beginning in the four - class model, classes with less than 10% of the population result. Therefore, based on these criteria we selected the model s considering four classes for both the dairy and poultry applicati ons, the results of ea ch are presented in the following respective subsections. 27 This cut - off allowed us to categorize all but six participants in a single class who were subsequently dropped from analysis. 115 Table 3. 4 Latent Class Model Comparisons Dairy Application Number of Latent Classes 2 3 4 5 Size of Class 1 0.326*** (0.021) 0.169*** (0.017) 0.063*** (0.010) 0.0 54*** (0.009) Size of Class 2 0.674*** (0.021) 0.256*** (0.018) 0.123*** (0.015) 0.107*** (0.013) Size of Class 3 0.576*** (0.022) 0.228*** (0.017) 0.109*** (0.013) Size of Class 4 0.586*** (0.020) 0.224*** (0.017) Size of Class 5 0.505*** (0.021 ) LLF - 29213.929 - 28749.827 - 28475.023 - 28032.319 AIC 58461.858 57551.654 57020.046 56152.638 3AIC 58478.858 57577.654 57055.046 56196.638 BIC 58433.524 57506.170 56957.157 56072.206 Poultry Application Number of Latent Classes 2 3 4 5 Size of C lass 1 0.346*** (0.023) 0.218*** (0.025) 0.095*** (0.018) 0.081*** (0.015) Size of Class 2 .654*** (0.023) 0.305*** (0.021) 0.110*** (0.016) 0.092*** (0.016) Size of Class 3 0.477*** (0.029) 0.315*** (0.023) 0.121*** (0.018) Size of Class 4 0.480*** (0.025) 0.257*** (0.022) Size of Class 5 0.448*** (0.025) LLF - 20227.557 - 19881.523 - 19659.975 - 19490.078 AIC 40489.114 39815.046 39389.950 39068.156 3AIC 40506.114 39841.046 39424.950 39112.156 BIC 40460.780 39769.562 39327.061 38987.724 116 3.4.1 Results of the Dairy Application Table 3. 5 reports the shares of preferences of the nine practices calculated using the coefficients from the LCM ( see Table A3.2 in the Appendix) and employing the Krinsky and Robb method as described in the methods secti on. Before describing the preferences of each class individually we examine the commonalities across classes. Training is deemed the most important farm practice across the two largest classes (14% in Class 1 and 27% in Class 2). Additionally, while not th e highest ranked farm practice for members of the third largest class, training was reflected as important 15% of times over the least important practice (worker sick policies). It also never ranks below the fifth (mid - point) most important practice for Cl asses 3 and 4. Therefore, training, which benefits both animals and workers, is the preferred farm practice among the U.S. public of those investigated. Promptly treating or euthanizing sick animals tends to also be a farm practice that is ranked as highly important receiving 14%, 21%, 25%, and 5% importance shares in Classes 1 - 4 respectively . In terms of rank order, treatment of sick animals is ranked once as most important, twice as second most important, and once as fourth most important. With one except ion it is always ranked higher than its worker welfare counterpart practice. Interestingly, although common in white - collar jobs, worker paid sick leave is ranked in the bottom three most important farm practices for the three largest classes, suggesting t hat implementing this farm practice is not a priority to the majority of the public. While it is helpful to know the preferences of the U.S. public as a whole, it is also informative to explore preferences across groups. 117 Table 3. 5 Latent Class Modeling Selected Production Practices in the Dairy Industry Class 1: Values practices equally Class 2: Animal welfare oriented Class 3: Concerned with treatment of sick animals Class 4: Concerned with worker he alth % of Sample in Class 58. 6% 22.8% 12.3% 6.3% Production Practice AW_Breaks 0.087 [0.072, 0.105] 0.146 [0.131, 0.165] 0.185 [0.140, 0.233] 0.028 [0.026, 0.030] A WW_Breaks 0.089 [0.073, 0.107] 0.017 [0.014, 0,019] 0.052 [0.038, 0.069] 0.140 [0.1 32, 0.149] AW_Ver 0.104 [0.08 5, 0.126] 0.132 [0.117, 0.150] 0.015 [0.011, 0.019] 0.016 [0.015, 0.017] WW _ Ver 0.108 [0.089, 0.128] 0.087 [0.076, 0.100] 0.185 [0.138, 0.243] 0.026 [0.025, 0.028] AW_Sick 0.135 [0.113, 0.160] 0.210 [0.190, 0.231] 0.245 [0.1 85, 0.317] 0.051 [0.048, 0.055] WW_Sick 0.098 [0.080, 0.118] 0.008 [0.007, 0.010] 0.045 [0.032, 0.060] 0.152 [0.143, 0.161] AW_Health 0.114 [0.096, 0.135] 0.108 [0.093, 0.123] 0.035 [0.026, 0.046] 0.014 [0.013, 0.015] WW_Health 0.125 [0.098, 0.155] 0.01 9 [0.016, 0.022] 0.088 [0.062, 0.120] 0.541 [0.525, 0.557] Training 0.140 [0.117, 0.165] 0.272 [0.251, 0.295] 0.151 [0.110, 0.196] 0.030 [0.028, 0.032] A 95% Confidence intervals derived following Krinsky and Robb (1986) are reported in brackets. Cl ass 1 is the largest class. It contains the majority of the population, comprising over 50% of respondents. Given that Class 1 ranks all practices evenly, t his group potentially poses a challenge to producers depending on their level of political activism , wh ich is open for further research. Many of the confidence intervals include 0.11 which would be the equal preference share value. On the other hand, it is also possible that this group has no clear farm practice prioritization because this is not an issue t o which they devote considerable attention. If this is the case then producers may feel relieved that the majority of the public does not deem 118 interfering in or overseeing farm practices as important. These two divergent outcomes reiterate the need for fur ther research into the behaviors of Class 1. Class 2 was the second largest membership group , with an associated class membership probability of 22.8% . Members of this group had perhaps the most intuitive preference rankings to interpret. To illustrate, th e training policy, which was presented as the only practice to improve both animal and worker welfare, has the highest share of preference, with 27 % of respondents on average considering training to be the most important policy . The next most important far m practices to C lass 2 members were the animal welfare sick and break practices , which received 21% and 15% share of respondents respectively . Interestingly, the worker welfare companions of these practices ( WW_sick and WW_break ) were the least and second least important practices, respectively. F o r example , w o rker sick has the lowest preference share of all farm practices across all classes ; less than 1% of respondents would rank it as most important . . Further, w hile worker sick leave was distinctly deeme d least important, all stand - alone worker welfare practices were ranked lower than their corresponding animal welfare practices ; 15 %, 13 % , 21%, and 11 % for verification, sick practices, and health practic es . Therefore, C lass 2 individuals can be characterized as prioritizing animal welfare above worker welfare farm practices . There is a less obvious logic to the preference rankings of C lass 3 , which has an associated class membership probability of 12.3% . On one hand an animal welfare practice, AW_sick, ranks as the most important practice ( 25 %) . On the other hand, a different animal welfare practice, AW_Ver , ranks as the least impor tant farm practice ( 2% ) . Furthermore, a worker welfare practice, WW_Ver , is tied to the second most important practice ( 1 9 %) . Thus, 119 unlike C lass 2 , we cannot conclude that C lass 3 members strongly prefer either worker or animal welfare. We also cannot say that this group ranks practices based on the welfare pair category. For example, while verification of animal welfare conditions , AW_ver, is the least important practice to this group ( 2 %) , the analogous verification of worker welfare conditions , WW_ver, is tied for the second most important practice ( 19 %) . W e refer to this class as the group. Class 4 has the lowest associated class membership probability ; 6.3%. The most important farm practice s to C lass 4 are associated with the worker welfare practices . For example, WW_ h ealth , with a preference share of 54 %, was ranked as the most important farm practice the largest share for any practice across all classes WW_sick and WW_breaks follow, with a share of prefer ences of 14% and 15.2% respectively. The lowest ranked worker welfare practice is worker verification as the third least important practice ( WW _ ver 3 %); however, this rank is still higher than the corresponding animal welfare verification practice ( A W _ ver 2 %) , suggesting that verification of farm practices is not a strong priority to this group compared to the implementation of such practices. Therefore, we refer to this class as the Since there are distinct preferen ce structures across groups, we might ask if these groups are identifiable by various demographics. That is, if we cannot easily observe these farm practice preferences, which unless an advocate or politically active it is difficult to do, can we identify members of these classes by other characteristics? The demographics of each class are summarized in Table A3.3 of the Appendix . Young individuals (under 30) are most likely to be a member of Class 1. While not statistically different from other classes, cl ass 1 members tend to be skewed toward lower to mid - range (less than $100,000) incomes. Respondents who are in 120 Group 4 of the animal welfare - worker welfare scale are also generally more likely to be in Class 1 than the other classes. Class 2 has the larges t percent representation of females in the study. It does not include many individuals from the Northeast compared to the other classes. Republicans are most likely to be in Class 2. Class 3 has fewer distinguishing characteristics, often falling within th e range of other class demographics. Class 4 is most likely to be comprised of middle - age (from 30 - 65 years old) low - income (less than $35,000) individuals. Class 4 also has the greatest share of Democrats. 3 .4.2 Results of the Poultry Application The pre ference shares are reported in Table 3. 6 and were calculated from the coefficients reported in Table A 3. 4 of the Appendix and using equation (3) . Before exploring the preferences of each class in the poultry experiment, it is worth discussing two observati ons comparing rankings across groups. The first is that the animal sick practice and animal meals practice are deemed the most and second most important farm practices for the three largest classes , respectively. Prompt treatment of sick animals receives 1 6%, 27%, and 37% of respondents from each respective class. Constant access to food and water for hens receives 16%, 22%, and 24% of respondents from each respective class. I t is clear from these results that in the poultry industry attending to animal hea lth quickly and ensuring constant access to food and water should be the priority. 28 Also consistent across classes is the ranking of the flock size practice ensuring a maximum ratio of chickens to workers. Flocksize was ranked as the fourth or fifth 28 It is worth stressing here that we make no claims about the proportion of the industr y producers already implementing this practice and therefore whether there is need for change. Our results provide commentary on the farm practices currently implemented and that could be implemented in the poultry industry. 121 most i mportant practice, towards the upper middle of rankings, for all groups. Thus, implementing this farm practice is not necessarily a priority but is also potentially less likely to cause debate and frustrations among the general U.S. public. Despite this si milar ranking across classes, our analyses indicate that preferences are indeed heterogeneous across groups, warranting an exploration of the preferences of each cohort individually. Table 3. 6 ortance of Selected Production Practices in the Poultry Industry Class 1: Verification of low importance Class 2: Animal welfare oriented Class 3: Concerned with animal health and sickness Class 4: Verification of high importance % of Sample in Class 48 .0% 31.5% 11.0% 9.5% Production Practice AW_Meals 0.159 [0.147, 0.170] 0.216 [0.192, 0.242] 0.242 [0.200, 0.287] 0.045 [0.034, 0.058] A WW_Meals 0.108 [0.100, 0.116] 0.016 [0.014, 0.019] 0.040 [0.027, 0.057] 0.148 [0.115, 0.185] AW_Ver 0.084 [0.077 , 0.092] 0.089 [0.078, 0.102] 0.009 [0.006, 0.012] 0.136 [0.112, 0.163] WW_Ver 0.087 [0.080, 0,095] 0.049 [0.042, 0.057] 0.008 [0.006, 0.011] 0.278 [0.221, 0.338] AW_Sick 0.160 [0.148, 0.172] 0.269 [0.246, 0.292] 0.373 [0.316, 0.432] 0.057 [0.039, 0.078] WW_Sick 0.116 [0.106, 0.126] 0.008 [0.007, 0.010] 0.030 [0.020, 0.043] 0.023 [0.016, 0.032] AW_Cage 0.073 [0.067, 0.079] 0.175 [0.155, 0.196] 0.069 [0.053, 0.088] 0.059 [0.033, 0.095] WW_Cage 0.137 [0.127, 0.147] 0.060 [0.052, 0.071] 0.160 [0.126, 0.20 0] 0.124 [0.102, 0.148] Flocksize 0.117 [0.105, 0.130] 0.117 [0.105, 0.130] 0.068 [0.052, 0.085] 0.130 [0.104, 0.157] A 95% Confidence intervals derived following Krinsky and Robb (1986) are reported in brackets. 122 For Class 1 (48% of the population) the preferred beneficiar y of the welfare practice is split. The two most important practices to this group are for the benefit of hens ( AW_sick and AW_meals with 16% of respondents each with the share of preferences of %) while the two least important practic es are also for animals ( AW_cage and AW_ver with 7% and 8% of respondents respectively X with the share of preferences of %) . Respondents in this class rank verification schemes low in importance , with preference shares around 8% . Other wise there seems to b e no obvious logic as to why class 1 poultry has the following farm practice ranking from most important to least important: AW_ s ick ( 16 %) , AW_ m eals ( 16 %) , WW_ c age ( 14 %) , Flocksize ( 12 %) , WW_ s ick ( 12 %) , WW_ m eals ( 11 %) , WW_ v er ( 9 %) , AW_ v er ( 8 %) , AW_ c age ( 7 % ) . 29 We refer to this class as C lass 2 (3 3 % of the population) can be described as preferring animal welfare practices to worker welfare practices. In fact, as can be seen, t he animal sick and animal meals practices w ere the most ( 27 %) and second ( 22 %) most important farm practices to C lass 2 individuals , followed by AW_ c age ( 18% ) . In contrast the worker - oriented sick and meal farm practices received the ranking of least and second least important farm practices among C lass 2 individuals , collectively comprising less than 3% of the preference shares . Therefore, w e refer to this class as 29 It is possible the reader wou ld disagree with this statement and consider class 1 uninformed and concerned about animal welfare and C lass 2 as informed and concerned about animal welfare. While we have posited that cage free systems are good for hens, this orientation was chosen mostl y for the perceived consumer belief about the system (Lusk 2019; Lusk 2010). As mentioned in footnote 14 there is also existing research indicating that cage - free systems are worse for hens. If someone believe this line of research and knew that many anima l welfare verification labels require this system, then they would actually be against both the cage - free system and such programs that require it when prioritizing hen health. We have made a judgement call that this is not the primary driver of preference rankings for this group but it is not based on evidence. Future research into C lass 2 poultry preferences is needed. 123 Class 3 (11% of the population) devotes significant attention to their perceived most important pract ices related to animal illness and access to food and water ( 37 % and 24% share of preferences, respectively ) . P ractices related to the health benefits of people in t he cage - free system ( WW_cage , 16% ) are also highly important . On the other hand, respondents in this class would, on average, devote less than one percent of preference shares to each verification practice ( 1 % for both AW_ver and WW_ver ) . W e refer to this class as Concerned with animal health and Class 4 ( 10 % of the population ) deviates strongly from the other groups in that the top two most important p ractices for other groups are in the bottom three most important farm practice for C lass 4 . Instead, C lass 4 values verification practices as most important ; w orker verification grasps a 28% preference shares and animal welfare verification a 14% preference share. This is particularly interesting as there are few such verification programs that currently exist related to worker welfare . Also important to this group is worker time for meal and snack breaks. Of all the classes , C lass 4 ranks worker welfare practices most favorably with a 15% preference share . As class 4 was the only class to rank verification practices so highly, a natural question is if these respondents live in are as where such verification systems already exist. For worker welfare this is largely limited to the South. As can be seen in Table A3.5 in the Appendix , the is n ot statistically different from the other classes. Class 4 contains the most individuals with at 2 has a greater representation of middle - aged individuals and a smaller proportion of Democrats. 124 Finally, Class 1 is younger with a more positive illegal aliens score indicating individuals in that class look more favorably on illegal aliens. 3.5 Discussion As previously mentioned, these experiments were inspir ed in part by Wolf and Tonsor (2017). Therefore, we begin our discussion of results by comparing the results from the dairy application to elicit rankings of farm practices amongst consumers. There are many differences between what consumers say the most effective and practical production practices are and the practices they are willing to pay the most for. Training was ranked the most effective of the practices re plicated in this study but only ranked third in terms of practicality and willingness - to - pay (WTP). Our findings were that the majority ranked training as the most important farm practice, perhaps suggesting that consumers evaluate importance based on effe ctiveness. However, this interpretation is contradicted when we look at AW_Sick. AW_Sick was ranked as most or second most important among three of the classes. It was ranked as most practical and had the second highest WTP estimate in Wolf and Tonsor (201 7). However, of the four farm practices used in both dairy studies, promptly treating or euthanizing sick animals was deemed the least effective practice. Perhaps most interesting is that consumers were willing to pay the greatest amount for third party ve rification among the four farm practices. Yet in our study, third party verification is deemed comparatively unimportant, which is more similar to its rank in effectiveness and practicality questions. Thus, as in other studies, these results leave us to wo nder if people actually act on the beliefs that they state are most important to them. 125 While there are still certainly areas for further research, part of the value of exploring preferences for farm practices in the dairy and poultry industries is to see how preference ranking compare across industries. Before proceeding with this comparison, recall that respondents only completed one experiment. Therefore, the individuals in the dairy experiment are not the same as those providing their rankings in the po ultry experiment. Also, Class 1 in the dairy application is not comprised of the same type of individuals as Class 1 in the poultry experiment. Keeping this information in mind, what can be learned by comparing the two experiments? First, the practice bene fiting both animals and workers is never ranked lower than fifth in most important. For the dairy industry the training practice is deemed the most important practice by the majority, while the flock size to worker ratio is a middle - of - the - pack practice. C onsistently near the top of the most important list of farm practices across industries is promptly treating or euthanizing sick animals. This might be because respondents envision a halo effect, that healthy animals leads to safer or better tasting consum ption goods. On the other hand, worker paid sick time off is typically not deemed an important farm practice. This might be because paid sick time off is unusual in hourly wage - based positions. Nevertheless, among public debate about access to medical care and concerns about community health, it is interesting that the public does not prioritize providing individuals time to recover from illness without financial distress. 3.6 Conclusion The U.S. public is placing increasing pressure on animal food product producers to treat their animals as the public deems most acceptable . Recently, there have also been discussions about rights on farms. Understanding which farm prac tices consumers deem acceptable or 126 unacceptable is important to producers who wish to maintain demand for their products and maintain their social license to produce . Additionally, insight into public opinion can assist policymakers in determining when to intervene in business affairs based on notions of social acceptability. Furthermore, in an environment where voters often demand many changes, some of which may be contradictory, it is helpful for policymakers to know which issues are of most importance to constituents. Additionally, there may be different groups of constituents with varied preferences. Understanding the preferences of these groups and the respective size of each group could also be beneficial to policymakers. In this study, we employ two b est worst scaling experiments, one each for hens and cows, to Within the national sample there were found to be four clusters of respondents within each industry. T he second large st cluster in both the dairy and poultry industry experiments valued animal welfare practices over worker welfare practices. The three remaining clusters in the dairy experiment included a group whose preferences were difficult to explain, a small group who placed high importance to worker health plans, and the majority who viewed all farm practices as approximately equally impo rtant. In the aggregate, training and promptly treating sick animals were deemed as the most important farm practic es in the dairy study . By comparison , in the poultry industry treatment of sick animals and animal access to food and water were pivotal to most individuals. These practices may be highly ranked because of a halo effect in which consumers perceive the safe ty or the quality of the food product to increas e with implementation of prompt hen treatment and reliable access to nutrients. Also notable , maintaining a specific ratio of hens to workers was also a palatable but neither highly important nor highly unimp ortant practice to implement. 127 Observations related to C lass 1 of the dairy experiment and C lass 1 of the poultry experiment suggest , small segments of the population may be driving the call for labeling programs and health insurance for workers. The forme r is consistent with prior studies which have indicated that small shares of the population are willing to pay a price premium great enough to cover implementation costs of such systems. Our findings should be considered by agribusinesses and marketers, p articularly amidst the growing proliferation of labeling programs, and their corresponding verificati on systems, in food markets. There are a number of interesting questions resulting from this novel exploration into attitudes surrounding worker welfare in agricultural sectors. Future research could focus on whether the public acts on these preferences politically and estimate any increased costs for implementation on farm . 128 APPENDIX 129 APPENDIX Table A 3. 1 Experimental Design A Farm practice order within each question and question order were randomized across respondents B Farm practices in Tables 3.1 and 3.2 were numbered from 1 to 9 from the top right cell to the right and down. Question Number Fa rm Practices A 1 9 B 1 7 8 5 2 2 4 5 2 1 6 9 3 4 8 2 6 1 7 4 2 1 3 6 9 7 5 8 2 3 5 4 1 6 4 5 7 2 9 3 7 1 4 8 9 3 7 8 5 7 6 3 8 2 9 5 9 8 7 4 6 10 9 1 5 8 3 6 11 1 4 7 5 6 3 12 4 9 3 2 8 6 130 Table A3.2 Summary Statistics of Basic Demographics t - Tests of Differences in Sample Average across Dairy Latent Classes Variable Class 1: Mid - age, low - income, Democrats Class 2: Non - distinct Class 3 : Female, non - Northeasterner Republicans Class 4: Young, low to middle income, AW - WW scale group 4 Gender 0.651 0.762 0.798 0.714 Age Young 0.023 0.179 0.129 0.215 Mid - age 0.814 0.595 0.669 0.690 Senior 0.163 0.226 0.202 0.095 Education Low 0.070 0.024 0.028 0.048 Mid 0.465 0. 571 0.657 0.653 High 0.465 0. 405 0.315 0.299 Income Low 0.767 0.690 0.567 0.484 Mid 0.233 0.310 0.433 0.516 High 0.163 0.131 0.112 0.116 U.S. Census Region Northeast 0.279 0.250 0.152 0.209 Midwest 0.209 0.226 0.270 0.257 South 0.349 0.369 0.433 0.382 West 0.163 0.154 0.146 0.151 Political Party Democrat 0.465 0.310 0.281 0.385 Republican 0.186 0.310 0.340 0.273 In - group Identification 3.897 3.858 3.881 3.757 Illegal Aliens Scale - 0.116 - 0.357 - 0.478 - 0.204 131 Table A3.2 (cont d) t - Test for Differences across Classes (yes = significant at 0.10 level) Variable 1 & 2 1 & 3 1 & 4 2 & 3 2 & 4 3 & 4 Gender No Yes No No No Yes Age Young Yes Yes Yes No No Yes Mid - age Yes Yes Yes No No No Senior No No No No Yes Yes Education Low No No No No No No Mid No Yes Yes No No No High No Yes Yes No Yes No Income Low No Yes Yes Yes Yes Yes Mid No Yes Yes Yes Yes Yes High No No No No No No U.S. Census Region Northeast No Yes No Yes No Ye s Midwest No No No No No No South No No No No No No West No No No No No No Political Party Democrat Yes Yes No No No Yes Republican No Yes No No No Yes In - group Identification No No No No No Yes Illegal Aliens Scale No Yes No No No Yes 132 Table A3.3 Importance of Selected Production Practices in the Dairy Industry Class 1: Mid - age, low - income, Democrats Class 2: Non - distinct Class 3 : Female, non - Northeaster ner Republicans Class 4: Young, low to middle income, AW - WW scale group 4 Production Practice AW_Breaks - 1.698*** (0.187) A 1.428*** (0.150) 2. 869 *** (0.119) - 0.110** (0.044) WW_Breaks - 0.080 (0.155) 0.147 (0.145) 0.694*** (0.081) - 0.089** (0.041) AW_Ver - 2.230*** (0.201) - 1.117*** (0.138) 2.765*** (0.122) 0.071 (0.045) WW_Ver - 1.757*** (0.197) - 1.431*** (0.152) 2.347*** (0.110) 0.095** (0.045) AW_Sick - 1.085*** (0.194) 1.700*** (0.151) 3.228*** (0.116) 0.320*** (0.044) WW_Sick Baseline Farm Pra ctice (0.00) AW_Health - 2.393*** (0.215) - 0.238* (0.137) 2.566*** (0.113) 0.154*** (0.045) WW_Health 1.269*** (0.178) 0.678*** (0.131) 0.811*** (0.074) 0.241*** (0.042) Training - 1.615*** (0.213) 1.220*** (0.140) 3.487*** (0.122) 0.360*** (0.046) Membe rship Percent 6.3%*** 12.3%*** 22.8%*** 58.6%*** Log Likelihood - 28475.023 Akaike Information Criterion 57020.046 Bayesian Information Criterion 56957.157 A Standard errors are in parentheses. Asterisks (***, **, *) indicate significance at the 0.01, 0 .05, and 0.10 levels, respectively. 133 Table A3.4 Summary Statistics of Basic Demographics t - Tests of Differences in Sample Average across Poultry Latent Classes Average Variable Class 1: Highly educated Class 2: Female Class 3: Mid - age, non - Democr a ts Class 4: Young and more pro I.A. Gender 0.816 0.845 0.747 0.674 Age Young 0. 0 81 0.119 0.118 0.270 Mid - age 0.673 0.593 0.724 0.660 Senior 0.245 0.288 0.159 0.069 Education Low 0.000 0.000 0.012 0.039 Mid 0.530 0.695 0.694 0.676 High 0.46 9 0.305 0.294 0.286 Income Low 0.510 0.508 0.571 0.571 Mid 0.490 0.492 0.429 0.429 High 0.102 0.085 0.118 0.124 U.S. Census Region Northeast 0.184 0.136 0.188 0.220 Midwest 0.204 0.288 0.271 0.239 South 0.449 0.407 0.353 0.390 West 0.163 0 .169 0.188 0.151 Political Party Democrat 0.469 0.322 0.282 0.382 Republican 0.306 0.288 0.341 0.236 In - group Identification 3.617 3.885 3.885 3.617 Illegal Aliens Scale - 0.327 - 0.373 - 0.465 - 0.050 134 Table A3.4 (cont d) t - Test for Differences across Classes (yes = significant at 0.10 level) Variable 1 & 2 1 & 3 1 & 4 2 & 3 2 & 4 3 & 4 Gender No No Yes Yes Yes No Age Young No No Yes No Yes Yes Mid - age No No No Yes No No Senior No No Yes Yes Yes Yes Education Low No No Yes No Yes Yes Mid Yes Yes No No Yes No High Yes Yes No No Yes No Income Low No No No No No No Mid No No No No No No High No No No No No No U.S. Census Region Northeast No No No No No No Midwest No No No No No No South No No No No No No West No No No No No No Political Party Democrat No Yes No No No Yes Republican No No No No No Yes In - group Identification Yes Yes Yes No No Yes Illegal Aliens Scale No No Yes No Yes Yes 135 Table A3.5 Production Practices in the Poultry Industry Class 1: Highly educated Class 2: Female Class 3: Mid - age, non - Democrats Class 4: Young and more pro I.A. Production Practice AW_Meals 0.704*** (0.155) A 2.083*** (0.219) 3.255*** (0.119) 0.313*** (0.060) WW_Meals 1.886*** (0.264) 0.284 (0.188) 0.655*** (0.070) - 0.072 (0.058) AW_Ve r 1.806*** (0.190) - 1.250*** (0.228) 2.371 *** (0.107) - 0.321*** (0.068) WW_Ver 2.513*** (0.265) - 1.321*** (0.217) 1.775*** (0.098) - 0.283*** (0.062) AW_Sick 0.903*** (0.157) 2.522*** (0.229) 3.475*** (0.105) 0.323*** (0.063) WW_Sick Baseline Farm Practi ce (0.00) AW_Cage 0.921*** (0.174) 0.826*** (0.221) 3.047*** (0.116) - 0.468*** (0.067) WW_Cage 1.707*** (0.206) 1.671*** (0.202) 1.983*** (0.093) 0.169*** (0.063) Flocksize 1.756*** (0.206) 0.819*** (0.224) 2.643*** (0.115) - 0.429*** (0.063) Membership Percent 0.095%*** 11.0%*** 31.5% *** 48.0%*** Log Likelihood - 19659.975 Akaike Information Criterion 39389.950 Bayesian Information Criterion 39327.061 A Standard errors are in parentheses. Asterisks (***, **, *) indicate significance at the 0.01, 0.05 , and 0.10 levels, respectively. 136 REFERENCES 137 REFERENCES 5 questions about cage - free hen health, welfare, 2017 Egg Industry. Aramark Establishes Animal Welfare Policy [WWW Document], n.d. URL https://www.aramark.com/about - us/news/ aramark - general/aramark - establishes - animal - welfare - policy (accessed 7.29.19). Auger, P., Devinney, T.M., Louviere, J.J., 2007. Using Best - Worst Scaling Methodology to Investigate Consumer Ethical Beliefs across Countries. Journal of Business Ethics 70, 299 326. Bennett, R.M., 1997. Farm animal welfare and food policy. Food Policy 22, 281 288. https://doi.org/10.1016/S0306 - 9192(97)00019 - 5 Bennett, R.M., Blaney, R.J.P., 2003. Estimating the benefit s of farm animal welfare legislation using the contingent valuation method. Agricultural Economics 29, 85 98. https://doi.org/10.1016/S0169 - 5150(03)00037 - 9 Breaks and Meal Periods | U.S. Departm ent of Labor [WWW Document], n.d. URL https://www.dol.gov/general/topic/workhours/breaks (accessed 7.28.20). Caputo, V., Lusk, J.L., 2020. What agricultural and food policies do U.S. consum er prefer? A best - worst scaling approach. Agricultural Economics 51, 75 93. Chang, J.B., Lusk, J.L., Norwood, F.B., 2010. The Price of Happy Hens: A Hedonic Analysis of Retail Egg Prices. Journal of Agricultural and Resource Economics 35, 406 423. Clark, B ., Stewart, G.B., Panzone, L.A., Kyriazakis, I., Frewer, L.J., 2017. Citizens, consumers and farm animal welfare: A meta - analysis of willingness - to - pay studies. Food Policy 68, 112 127. https://doi.org/10.1016/j.foodpol.2017.01.006 Coalition for Sustainabl e Egg Supply, n.d. Final Research Results Report. Ellison, B., Brooks, K., Mieno, T., 2017. Which livestock production claims matter most to consumers? Agric Hum Values 34, 819 831. https://doi.org/1 0.1007/s10460 - 017 - 9777 - 9 Finn, A., Louviere, J., 1990. Shopping - center patronage models: Fashioning a consideration set segmentation solution. Journal of Business Research, Special Issue Cutting Edge Research in Retailing 21, 259 275. https://doi.org/10.1 016/0148 - 2963(90)90032 - 9 Grebitus, C., Lusk, J. L., and Nayga, R. M. 2013. Effect of distance of transportation on willingness to pay for food. Ecological economics, 88:67 - 75. 138 Greenhouse, S. , 201 5 . Farm Labor Groups Make Progress on Wages and Working Condi tions . The New York Times. Heng, Y., Peterson, H.H., Li, X., 2013. Consumer Attitudes toward Farm - Animal Welfare: The Case of Laying Hens. Journal of Agricultural and Resource Economics 38, 418 434. Kendall, H.A., Lobao, L.M., Sharp, J.S., 2006. Public Con cern with Animal Well - Being: Place, Social Structural Location, and Individual Experience. Rural Sociology; Columbia 71, 399 428. http://dx.doi.org.proxy2.cl.msu.edu/10.1526/0036 01106778070617 Krinksy, I., Robb, A. 1986. One Approximating the Statistical Properties of Elasticities. The Review of Economics and Statistics 68, 715 - 719. Louviere, J.J., Flynn, T.N., Marley, A.A.J., 2015. Best - worst Scaling: Theory, Methods and Applica tions. Cambridge University Press. Lusk, J.L., 2010. The Effect of Proposition 2 on the Demand for Eggs in California. Lusk, J. L. 2011. External validity of the food values scale. Food Quality and Preference, 22(5): 452 - 462. Lusk, J.L., 2019. Consumer pre ferences for cage - free eggs and impacts of retailer pledges. Agribusiness 35, 129 148. https://doi.org/10.1002/agr.21580 Lusk, J.L., Briggeman, B.C., 2009. Food Values. American Journal of Agricultural Econ omics 91, 184 196. Lusk, J.L., Tonsor, G.T., 2016. How Meat Demand Elasticities Vary with Price, Income, and Product Category. Applied Economic Perspectives and Policy 38, 673 711. https://doi.org/10.1093/aepp/ppv050 Lyons, P.A., Coursey, L.E., Kenworthy, J.B., 2013. National Identity and Group Narcissism as Predictors of Intergroup Attitudes Toward Undocumented Latino Immigrants in the United States. Hispanic Journal of Behavioral Sciences 35, 323 335. https://doi.org/10.1177/0739986313488090 Madison, J. 1787. Federalist No. 10: The Same Subject Continued: The Union as a Safeguard Against Domestic Faction and Insurrection. New York Daily Advertiser . Malone, T., Lusk, J.L., 2016. Putting the Chicken Be fore the Egg Price: An Ex Post Analysis of 532. Marley, A.A.J., Louviere, J.J., 2005. Some probabilistic models of best, worst, and best worst choices. Journal of Mathema tical Psychology, Special Issue Honoring Jean - Claude Falmagne: Part 1 49, 464 480. https://doi.org/10.1016/j.jmp.2005.05.003 139 McKendree, M.G.S., Croney, C.C., Widmar, N.J.O., 2014. Effects of demographic factors and information sources on United States cons umer perceptions of animal welfare. J Anim Sci 92, 3161 3173. https://doi.org/10.2527/jas.2014 - 6874 McKendree, M.G.S., Tonsor, G.T., Wolf, C.A., 2018. Animal Welfare Perceptions of the U.S. Public and C ow - Calf Producers. Journal of Agricultural and Applied Economics 50, 544 578. https://doi.org/10.1017/aae.2018.14 Minimum Length of Meal Period Required under State Law for Adult Employees in Private Sector 1 | U.S. Department of Labor [WWW Document], 2020 . URL https://www.dol.gov/agencies/whd/state/meal - breaks (accessed 7.28.20). National Population Totals and Components of Change: 2010 - 2019: Annual Estimates of the Resident Population for the United States, Regions, States, and Puerto Rico: April 1, 2010 to July 1, 2019 (No. NST - EST2019 - 01), 2019. U.S Census Bureau. Nocella, G., Hubbard, L., and Scarpa, R. 2010. Farm animal welfare, consumer willingness to pay, and trust: results of a cross - national survey. Applied economic perspectives and policy, 32(2):275 - 297. Ommundsen, R., Morch, S., Hak, T., Larsen, K.S., Veer, K.V.D., 2002. Attitudes toward illegal immigra tion: A cross - national methodological comparison. The Journal of Psychology; Provincetown 136, 103 10. Partners Fair Food Program [WWW Document], n.d. URL https://www.fairfoodprogram.org/partners/ (accessed 7.29.19). Regmi, P., Deland, T.S., Steibel, J.P ., Robison, C.I., Haut, R.C., Orth, M.W., Karcher, D.M., 2015. Effect of rearing environment on bone growth of pullets. Poult Sci 94, 502 511. https://doi.org/10.3382/ps/peu041 Regmi, P., Nelson, N., Steibe l, J.P., Anderson, K.E., Karcher, D.M., 2016. Comparisons of bone properties and keel deformities between strains and housing systems in end - of - lay hens. Poult. Sci. 95, 2225 2234. https://doi.org/10.3382/ps /pew199 The New York Times. Semega, J., Kollar, M., Creamer, J., Mohanty, A., 2019. Income and Poverty in the United States: 2018 (No. P60 - 266). U.S Census Bureau . Stone, J., Costanigro, M., Goemans, C., 2018. Public Opinion on Colorado Water Rights Transfers: Are Policy Preferences Consistent with Concerns over Impacts? Journal of Agricultural and Resource Economics 43, 403 422. https://doi.org/10.22004/ag.econ.276502 Times. 140 Tucker P, 2003. The impact of rest breaks upon accident risk, fatigue and performance: a review. Work & Stress 17, 123 137. https://doi.org/10.1080/0267837031000155949 U.S. Census Bureau, n.d. U.S. Census Bureau QuickFacts: United States [WWW Document]. URL https://www.census.gov/quickfacts/fact/table/ U.S. /PST045219 (accessed 7.24.20). sustainability labels on meat. Food Policy 49, 137 150. https://doi.org/10.1016/j.foodpol.2014.07.002 Walmart U.S. Announces New Animal Welfare and Antibiotics Positions [WWW Document], n.d. . Corporate - U.S. . URL https://corporate.walmart.com/newsroom/ 2015/05/22/walmart - u - s - announces - new - animal - welfare - and - antibiotics - positions (accessed 7.29.19). Research Center. Wolf, C.A., Tonsor, G.T., 2017. Cow Welfare in the U.S . Dairy Industry: Willingness - to - Pay and Willingness - to - Supply. Journal of Agricultural and Resource Economics; Logan 42, 164 179. Wolf, C.A., Tonsor, G.T., 2013. Dairy Farmer Policy Preferences. Journal of Agricultural and Resource Economics 38, 220 234. Wolf, C.A., Tonsor, G.T., Olynk, N.J., 2011. Understanding U.S. Consumer Demand for Milk Production Attributes. Journal of Agricultural and Resource Econo mics 36, 326 342.