Consumer willingness to pay for shelf life of high-temperature, short-time-pasteurized fluid milk: Implications for smart labeling and food waste reduction

Food waste in the United States was valued at $285 billion in 2019, representing 70% of all food surplus; dairy and eggs alone represented 15.90% of food surplus. Milk is the fifth most consumed beverage in the United States, and therefore its contribution to food waste has significant economic and environmental ramifications. Smart labels that provide precise spoilage information for fluid milk may help reduce food waste in fluid milk, but it is unclear if consumers will accept or pay for this novel technology. This paper examines consumer preferences for high temperature, short time pasteurized fluid milk shelf life and smart date labels and tests how information about the environmental impact of fluid milk food waste affects consumers’ acceptance and will-ingness to pay. We used a choice-based conjoint study administered in an online survey, along with a between-subject experiment to measure preferences under different information treatments about the environmental impact of food waste. Our results suggest that consumers' valuations of extended shelf life and an ecolabel is positive; however, using the smart label creates disu-tility for consumers, thereby hindering acceptance of new labeling technology that may facilitate food waste reduction in the milk industry. These findings imply that retailers should find alternative means to enhance the communication of precise shelf life information and its role in reducing food waste.


INTRODUCTION
Food waste is a major problem globally and nationally, with approximately one-third of the world's food supply ending up lost or wasted (Blakeney, 2019).In the United States alone, surplus food (i.e., unsold or uneaten food that would most likely be discarded in landfills, fed to animals, or donated to charity) accounts for 35% of the 229 million tons (208 million tonnes) of food available (ReFED, 2021).In 2019, dairy and eggs accounted for 15.9% of surplus food in the United States, ranking third after produce (34.3%) and prepared foods (18.8%;ReFED, 2020).Among dairy products, fluid milk has been estimated in the past to account for 66.9% of dairy food loss by weight (Buzby et al., 2014).As the fifth most consumed beverage in the United States (Statista, 2019b;Wolf et al., 2020), fluid milk accounts for 60.8% of total dairy department sales (Statista, 2021); thus, food loss of this magnitude may also significantly affect retailers' profitability.
Understanding consumer behavior is critical for food waste prevention (Quested et al., 2013;Roe et al., 2018;Bolos et al., 2019).The largest amount of food waste comes from households (37.2%), followed by consumer-facing businesses (13.1% in retailers and 15.8% in foodservice outlets), which significantly affects the environment, as food production requires large amounts of energy and water (US EPA, 2020).Many drivers of food waste have been proposed, such as spoilage and date-labeling confusion.Spoilage ranks as the third most cited reason to dispose of food (US EPA, 2020).Many US consumers dispose of edible food due to package date labels, which consumers often misconstrue as an indication that food is unsafe after that date (Gunders, 2017).Conversely, when shoppers choose products nearing spoilage that they consume right away, they effectively prevent waste at the retail level (Qi and Roe, 2016;Yu and Jaenicke, 2021).
Consumers' confusion about the meaning of product shelf life also contributes to food waste.For instance, consumers perceive product freshness as continuously declining until it reaches zero (i.e., loses value) by the end of product shelf life (i.e., date on the label; Wu et al., 2017).However, product shelf life is an estimate based on the producers' processing characteristics and assumptions on shipping and storage conditions.Therefore, if dynamic environmental data are collected from the shipping point to the final consumer, it is feasible that the use of technology (e.g., smart labels) could help to make accurate and dynamic changes to shelf life labels.
Smart labels incorporate technology that extends the functionality and content of tags or packaging beyond traditional print methods.Some applications include the theoretical extension of shelf life of food products and the enhancement of food traceability (Dobrucka and Cierpiszewski, 2014).Retailers also find them useful to dynamically update prices, shelf life, and other environmentally sensitive information (McFarlane et al., 2012).By providing timely product information, intelligent food packaging can also increase consumers' value perception and help reduce food waste (Skinner, 2015;Roe et al., 2018;ReFED, 2021).Given the costs associated with using smart labels and the low margins for fluid milk, producers also need to consider the extent to which customers are willing to accept and pay for this technology.
The purpose of this study is to examine consumer preferences for shelf life and smart date labels in HTSTpasteurized fluid milk and to test how information about the environmental impact of fluid milk food waste affects consumers' acceptance and willingness to pay (WTP).We focus on HTST milk because it is widely accepted by US consumers, and HTST pasteurization is the most common method of fluid milk pasteurization in the United States (IDFA, 2023).Therefore, as US consumers have been reported to show negative reactions to ultrapasteurized (UP) milk (Chapman and Boor, 2001), which typically has a shelf life of 60 to 90 d, shelf life extension of HTST milk could play an important role in fulfilling the demands of some customers and consumers for (non-UP) extended shelf life fluid milk products (Barbano et al., 2006;Koutchma and Barnes, 2013).Although COVID-19 may have led to further consumer interest in extended shelf life fluid milk products (Adams et al., 2021), it is also expected that some consumers may continue to negatively view extended shelf life products.
We implemented an online choice-based conjoint (CBC) study together with a between-subject experiment to measure preferences under different information treatments about the environmental impact of food waste.We used this to estimate consumers' WTP for HTST fluid milk with precise spoilage information provided via a smart label (e.g., a quick response code connected to temperature sensors to predict spoilage).By randomly providing information about food waste related to milk consumption, we also assessed how targeted environmental messaging moderates consumers' preferences for HTST fluid milk with a smart label coupled to a food waste-related ecolabel.
To achieve these objectives, we tested 5 main hypotheses related to consumers' valuation of shelf life, smart labeling technology, and food waste-related ecolabelling for HTST fluid milk products.Consumers prioritize price and freshness-the time when food's attributes (e.g., appearance and taste) are at their best-when choosing perishable products, and their perceptions of freshness are directly influenced by the framing and presence of shelf life date labels (e.g., code dates and expiration dates; Li et al., 2020).Most date labels reflect a minimum quality standard set by the manufacturer, rather than a food safety standard, but many consumers believe that products abruptly lose quality after that date and are no longer fit for consumption (Theotokis et al., 2012;Roe et al., 2018).The inconsistency of current labeling practices (e.g., "sell by," "best by," "use by," "best before") also generates confusion (Neff et al., 2019;ReFED, 2021;Yu and Jaenicke, 2021).Therefore, most consumers prefer perishable products with a longer shelf life (Wu et al., 2017).Following this logic, we therefore predict the following.
Hypothesis 1 (H1): Consumers are willing to pay more for HTST milk with longer shelf life as indicated by the days remaining until the date on the product date label.
Product traceability systems-such as smart packaging-can improve product safety and quality and enhance consumer trust (Lavelli, 2013;Tarjan et al., 2014), while also enabling marketers to target relevant consumer trends (e.g., sustainability, transparency; Tsiros and Heilman, 2005).Smart labels, such as quick response (QR) code labels, can therefore contribute to more efficient supply chain management (Dobrucka and Cierpiszewski, 2014), and designing an intuitive technology platform that provides consumers with relevant information encourages their adoption of QR code labels (Kim and Woo, 2016).Hence, we predict the following.
Hypothesis 2 (H2): Consumers are willing to pay more for HTST milk with precise spoilage information provided by smart label technology compared with milk with a static date label.
Ecolabels are tools that differentiate companies from their competitors (Hamilton and Zilberman, 2006) while informing consumers of companies' sustainability efforts (Parguel et al., 2011;Meis and Kashima, 2017).However, some factors that could limit consumption of products with ecolabels are price perception and WTP (Prieto-Sandoval et al., 2016;Yokessa and Marette, 2019;Meis-Harris et al., 2021).Previous research has found that consumers' WTP increases in the presence of ecolabels.Yokessa and Marette (2019) presented a list of studies that evaluated consumers' WTP for products (i.e., mainly food) with ecolabels.They found that consumers' WTP for green products (i.e., those with an ecolabel) was greater than that for equivalent regular products, regardless of exogenous characteristics (e.g., product type, country, type of ecolabel; Yokessa and Marette, 2019).We therefore predict the following.
Hypothesis 3 (H3): Consumers are willing to pay more for HTST milk with a food waste-related ecolabel compared with milk without an ecolabel.
Hypothesis 4 (H4): Consumers are willing to pay more for HTST milk with a smart label coupled to a food waste-related ecolabel compared with smartlabeled milk without a coupled ecolabel.
Ecolabels aim to persuade consumers to choose more sustainable products by providing information regarding products' sustainable features (Yokessa and Marette, 2019;Meis-Harris et al., 2021).However, communicating sustainable features could be challenging because ecolabel effectiveness tends to rely on consumers' sensitivity toward and involvement with environmental problems-such as food waste (Yokessa and Marette, 2019).Likewise, most eco-friendly outcomes could seem abstract (i.e., distant from the self) and intangible (White et al., 2019).Retailers and researchers have used targeted information treatments to influence consumer behavior by promoting products with particular attributes using material printed on the packaging or shared through other means (Scozzafava et al., 2020).For instance, White et al. (2019) suggested implementing analogies (i.e., comparing sustainable outcomes to relatable experiences) to promote a shift toward sustainable consumer behavior.An additional effort to enhance messaging around food waste could be combining ecolabels with targeted, relatable information pertaining to food waste (i.e., analogies) and a way to access relevant product information (i.e., smart labels with precise spoilage dates; Kim and Woo, 2016;Yokessa and Marette, 2019).Lau et al. (2022) conducted an 8-wk in-store experiment to determine consumer acceptance of QR-code smart labels that provide access to more precise fluid milk date labeling, along with food waste-related messaging, and found that some consumer segments would adopt this technology.Thus, we predict the following.
Hypothesis 5 (H5): Providing consumers with relatable information about food waste associated with milk consumption increases their WTP for HTST milk with: (a) a food waste-related ecolabel, and (b) a smart label coupled to a food waste-related ecolabel.
Prior studies have examined preferences among fluid milk consumers and found evidence of substantial variation (Wolf et al., 2011) and distinct market segments.
Harwood and Drake (2018) identified 4 distinct groups of US milk consumers that vary in terms of price sensitivity, preference for specific product attributes (e.g., locally farmed, organic, pasture raised, taste, brand), and demographic characteristics.Similarly, Wu et al. (2020) identified 4 segments of milk consumers in China based on their awareness of price, health, and environmental factors.H3, H4, and H5 may therefore be supported by some, but not all, consumers in our sample, so it is critical that our empirical analysis adequately captures preference heterogeneity.

Experimental Overview
We designed and administered a web-based survey to a paid online participant pool.All research activities were reviewed by Cornell University Institutional Review Board for Human Participants, and were approved and granted exemption under protocol number 2011009947.
The goal of the study was to elicit consumer preferences for milk products that use smart labels to provide a more accurate date label predicted from up-to-date data from temperature sensors throughout the milk's supply chain.Based on the intended design, each milk jug would hypothetically have a smart label and an ecolabel.Shoppers would then need to scan the QR code with a smartphone to access the embedded date, which would be updated in real time (Figure 1).

Survey Design
We designed the survey instrument and the CBC study using Qualtrics (CoreXM and ProductXM;Qualtrics, 2021).The full survey instrument is provided in Supplemental File S1 (https: / / hdl .handle.net/1813/ 112859).The survey was divided into 5 main sections: (1) milk purchase behavior; (2) milk CBC study; (3) online shopping habits and QR code use; (4) perceptions about discarding milk, food waste, labeling, and social currency; and (5) demographic information.We limited the CBC study to 4 choice tasks and sequenced it early in the survey to ensure data quality and minimize survey fatigue.The perceptions were assessed through a series of 5-point Likert questions based on the work of Richter and Bokelmann (2018).The 5-point Likert scale asked respondents the extent to which they agreed with each statement from strongly disagree (1) to strongly agree (5).
To test H5, the CBC section of the survey was immediately preceded by a between-subject information treatment about the environmental impact of food waste.Respondents were randomly assigned to control or treatment, and the treated group was presented with an information passage containing statistics about food loss and waste (Figure 2).Respondents in the control group did not see this information passage.
For the CBC study, the Qualtrics software designs the choice sets based on the fractional factorial method (i.e., a fraction of the full factorial).Choice sets are profiles (i.e., products, in this case, a gallon of dairy milk) based on the attributes and levels determined for the experiment (Qualtrics, 2021).The design algorithm uses a randomized, balanced design to determine choice sets for each respondent.Based on 5 attributes, each with 2 to 6 levels, 192 product combinations were creat-   2. 55, 3.28, 4.39, 5.31, 6.28, 7.20 1 Reference level in the data analysis. 2 Variable coded as numeric to determine the daily effect.
ed.Based on this design, 500 respondents were needed to ensure a minimum of 1,000 exposures per attribute level (Johnson and Orme, 2003).To prevent imbalance and dependencies between attribute combinations and to ensure precise estimation of preferences, we did not use prohibitions in the conjoint design (Chrzan and Orme, 2000).
The product attributes analyzed in the CBC study were as follows (Table 1): milk type (organic or conventional); label type (static or smart QR code); additional information label (ecolabel or no label); date printed on the container or digitally displayed by a smart label (tomorrow, 7 d from today, 18 d from today, or 30 d from today); and price ($2.55,$3.28,$4.39,$5.31,$6.28,and $7.20 in USD).Milk type (organic) is an important product attribute due to its market relevance as a differentiator in consumer demand (Hasselbach and Roosen, 2015;Scozzafava et al., 2020).For instance, 24% of consumers said organic labels mattered a lot when shopping for dairy products (Statista, 2019a), and organic milk commands a significant price premium over conventional milk.Moreover, consumers choose organic milk because they perceive that it addresses environmental, food safety, and animal welfare concerns (Hughner et al., 2007), which is particularly relevant in the context of a food waste information treatment and an ecolabel attribute.We determined price levels based on price data collected for one gallon conventional and organic fluid milk from 7 retailers in Ithaca, New York, during June 2020 (Table 2).We developed an initial menu of prices from this empirical distribution, which we augmented with average national prices for one gallon of conventional and organic milk reported by USDA (2020) to better reflect the range of milk prices across the United States and avoid biasing the distribution (Table 3).
To introduce the CBC study, respondents were asked to imagine they were in their favorite grocery store, looking to purchase one gallon of dairy milk.They were also asked to assume any attributes not listed were identical across all product alternatives (e.g., type of container, package size, fat content, heat treatment).Four CBC tasks were presented to each respondent.Each CBC task presented respondents with a discrete choice between 3 one-gallon milk products characterized by 5 attributes (Table 1).Each attribute was clearly explained before the start of the CBC tasks (Figure 3).In each CBC task, respondents were also given a fourth opt-out option if they did not wish to purchase any of the 3 products presented.Figure 4 presents an example of a CBC task a respondent might encounter.

Survey Implementation
We distributed the survey through Prolific, an online platform providing access to a paid survey participant pool of US consumers (Prolific, 2021).We used demographic screeners in Prolific to target participants that met 3 requirements: (1) someone in their household had to consume dairy milk; (2) they had to live in the United States; and (3) they had to be at least 18 years old.We piloted the survey on January 12, 2021, and implemented some technical improvements based on those results before launching the final survey on February 8, 2021.Each participant received $2.00 for a completed response, and we expected the survey to We collected data for many types of milk products, but we only used prices for conventional and organic dairy milk to generate the menu of prices for this study.Average price for a gallon of conventional milk (USDA, 2020). 3 Average price for a gallon of organic milk (USDA, 2020).
take at most 12 min to complete, for a targeted effective hourly compensation rate of $10.00.The actual average response time was 8.8 min, which, based on our own internal testing, was an adequate duration to fully process the information in the survey.We collected 500 responses in total, which yielded 498 usable observations after removing 2 responses with missing data.

Empirical Framework
A CBC model provides an economic framework to determine consumer preferences and calculate WTP.It is particularly valuable when considering products with multiple attributes.We analyze the CBC data using a mixed logit model, which captures unobserved het- erogeneity by allowing preference parameters to differ for each respondent.The gmnl package (Sarrias and Daziano, 2017) in R (R Software version 4.2.1) was used to estimate the mixed logit models.
We model individual choice using a random utility model (Louviere et al., 2000).When individual i is faced with alternative j in task t, the utility (U) of that choice can be represented as In this representation, β 0 is the price parameter and β k is an I × K matrix of individual-specific parameters for individuals i = 1…I and product attributes k = 1…K, both of which reflect changes in utility associated with a change in a given attribute level.Price ijt represents the price presented to individual i for product alternative j in task t, and X ijt k represents the kth nonprice attribute of product alternative j in task t.Finally, ε ijt is a general random component of utility, and θ K is a matrix of individual-and attribute-specific random utility parameters that capture unobserved preference heterogeneity (Kalkbrenner et al., 2017).Thus, if individual i chooses alternative j in task t, we assume that U ijt is greater than or equal to U imt for all m ≠ j, m∈ J.
Let Y it be a random variable that indicates the choice made in task t.If we further assume that θ i k is meanzero normally distributed with variance σ k 2 and ε ijt is independently and identically distributed extreme value type 1, then the probability that individual i chooses alternative j in task t, can be estimated using a mixed logit model (Kalkbrenner et al., 2017;Greene, 2018) as where F(•) is the cumulative standard normal distribution.We estimate a mixed logit specification based on Equation [2] to directly test each of the 5 hypotheses.
The model includes main effects for all 5 attributes with price as a fixed parameter (Hole, 2007) and all other parameters distributed randomly (for H1, H2, and H3), an interaction effect between date label type and additional information label attributes (for H4), interaction effects between each nonprice attribute and the information treatment group (for H5a), and a 3-way interaction between date label type, additional information label, and treatment group (for H5b).
In a subsequent exploratory analysis, we created interaction terms between each product attribute and individual demographic, behavioral, and perceptional variables to characterize preference heterogeneity based on observable factors.We used the main model as a baseline specification to which each group of interaction terms was separately added to explore whether individual characteristics moderated the value of product attributes when choosing to buy milk.

Willingness to Pay
The WTP for a level within an attribute (e.g., organic, ecolabel) is defined as the change in price that keeps the customers' utility constant.Based on random utility theory, we calculated average WTP values for each product attribute k using the estimated coefficients from the mixed logit model (Greene, 2018) as follows: Standard errors for the WTP estimates were computed using the delta method (Greene, 2018).

Summary Statistics
A total of 500 respondents completed the survey, of which there were 498 usable observations.We summarized the demographic characteristics for the total sample and for the control and treatment groups in Table 4. Overall, most respondents are the primary shopper in their household (78%), less than 35 years old (62%), never married (56%), identify as white (76%), live in the southern region of the United States (41%), and hold either a bachelor's degree (37%) or some college education (27%).The average household size is 2.89 people, and only a third of the respondents have children.Household income and gender are more evenly distributed.
We also summarized the data relating to purchase behaviors collected in the first part of the survey in Table 5, and average perceptions about a list of statements regarding labeling, food waste, discard intentions, and social currency in Table 6.Most shoppers in our sample purchase milk in large supermarkets or big-box stores (67%), drink whole and reduced-fat milk (77%), buy gallon and half-gallon size formats (86%), and read food labels always or most of the time (52%).Over one-third of respondents never consume nondairy milk (36%) and have never scanned a QR code within food products (36%).Lastly, about half of respondents are not familiar at all with food waste related to dairy milk (45%).
Our sample captures a broad representation of our market of interest-US fluid milk consumers; but it may not be representative of the entire US population.When generalizing our results, we therefore return to these individual demographic, behavioral, and perceptional characteristics and explore how they moderate consumer milk purchasing decisions.

Choice-Based Conjoint Study
Table 7 summarizes the results from the CBC study.The reference level for all analyses was conventional milk, static date label, and no additional information label.Date printed on the container and price were coded as continuous variables.Each coefficient estimate represents the average marginal utility of its respective attribute level in the CBC study.The estimated standard deviations reflect heterogeneity in the point estimates captured by the mixed logit model; thus, if the standard deviation for a particular attribute is statistically significant, preferences for that attribute vary to a certain degree.As expected, the coefficient estimate for price is negative, which implies downward-sloping demand curves (i.e., customers prefer lower prices).
The mean estimates for all the main effects, except milk type (organic) and additional information label, are statistically significant (P < 0.001), as are the standard deviation estimates (P < 0.001).Although the coefficient for organic milk is not statistically significant (P = 0.15), the large and statistically significant estimate for standard deviation (SD = 1.35;P < 0.001) indicates substantial heterogeneity among consumers.

Endara et al.: WILLINGNESS TO PAY FOR FLUID MILK SHELF LIFE
The coefficient estimate for date on label is positive, which implies that greater shelf life increases the likelihood of buying a particular milk product under the experimental conditions.Surprisingly, the smart label coefficient estimate is negative, which means that a smart label decreases the likelihood of buying a particular milk product under experimental conditions.In other words, shoppers prefer longer over shorter shelf life, and a static label over a smart label.
The point estimate for the interaction effect between date label type and additional information label is positive but not statistically significant (P = 0.32).For the information treatment, we find positive interaction effects (P < 0.01) with the additional information label.
The treatment positively moderates the effect of the additional ecolabel, implying a greater preference for the ecolabel when information on food waste is provided.Lastly, the coefficient estimate for the 3-way interaction between the smart label, ecolabel, and the information treatment is negative (P < 0.05).The positive effect of the ecolabel coupled with the treatment was attenuated by the presence of the smart label, consumers' disfavor of QR code use.
The precise standard deviation estimates for milk type, date label type, and additional information label indicate substantial preference heterogeneity among customers.Individual characteristics (e.g., age, income, behaviors) may explain some of that heterogeneity.Al- though the standard deviation for the date on the label is also statistically significant (P < 0.001), the small magnitude of the coefficient suggests that heterogeneity may not be economically meaningful.

Willingness to Pay
The WTP captures the predicted market value and relative importance of each product attribute level to respondents in our sample.We calculated attributelevel average WTP estimates, with standard errors calculated using the delta method (Table 7).The levels of precision for the WTP estimates match those for the coefficient estimates from the associated mixed logit model.On average, the price premium for each additional day of shelf life is valued at $0.08.On the other hand, consumers are willing to pay $0.86 less for a smart label over a static label.
Consistent with the coefficient estimates, the WTP estimate for the interaction between the smart label and ecolabel is positive but not statistically significant (P = 0.32).The WTP estimate for an ecolabel coupled with the information treatment is $1.12 (P < 0.001), which implies that exposure to the information treatment increases consumers' WTP for the ecolabel.However, the WTP estimate for the 3-way interaction between the smart label, ecolabel, and information treatment is −$0.85 (P < 0.05), which implies that  the presence of a smart label effectively reduces that (increase in) WTP for the ecolabel.

Preference Heterogeneity
Our exploratory analysis of preference heterogeneity suggests that many observable individual characteristics of respondents moderate the value of product attributes when purchasing milk (Table 8).For example, respondents that usually shop at warehouse club stores or specialty stores or drink lactose-free milk are more likely to choose (i.e., positively value) organic milk.Conversely, respondents that are lower-income, elderly, regular dairy milk drinkers, or do not shop online frequently are all less likely to choose (i.e., negatively value) organic milk.With respect to label types, respondents that frequently order online or show slight interest in using QR codes are more likely to choose a smart label.Furthermore, respondents that regularly drink nondairy alternatives, shop online, systematically read food labels, or use QR codes are more likely to choose an ecolabel.On the other hand, respondents that are middle-aged or older, male, or consume milk intermittently are less likely to choose an ecolabel.Lastly, respondents that drink lactose-free milk often are more likely to choose milk with longer shelf life.

DISCUSSION
This paper examines consumer preferences for HTST fluid milk shelf life and smart date labels and tests how information about the environmental impact of fluid milk food waste affects consumers' acceptance and WTP.
First, we hypothesized that consumers are willing to pay more for HTST milk with longer shelf life (H1).The results of the mixed logit models indicate that consumers prefer milk with a longer date on the label, consistent with existing literature regarding the importance of shelf life as a value driver for consumers and retailers (Theotokis et al., 2012;Wu et al., 2017;Roe et al., 2018;ReFED, 2021).We also found that lactose-free milk consumers more strongly prefer longer shelf life, perhaps because they are accustomed to the longer shelf life afforded by UP milk.Our results are similar to Tsiros and Heilman (2005), who studied the relationship between consumers' WTP and shelf life for many perishable products, including dairy milk, and found that WTP linearly decreases as milk approaches its expiration date.They estimate that consumers are willing to pay about $0.24 for an additional day of shelf life in milk, whereas our estimate is slightly lower at $0.08/d.On the other hand, our results contrast that of Schroeter et al. (2016), who evaluated consumers' WTP for extended shelf life by comparing none, short, and long options, and found no significant differences among the 3 options despite consumers ranking shelf life as the most important attribute when selecting milk.In terms of consumer preferences, longer shelf life may represent a convenience factor as it decreases the number of trips consumers need to make to the store and helps reduce food waste (Yu and Jaenicke, 2021), partially explaining our positive estimates for consumer WTP.Our results therefore support H1.
Second, we hypothesized that consumers are willing to pay more for HTST milk with precise spoilage information provided by a smart label (H2).We used the date label type attribute to test consumer preferences for a smart label that collects up-to-date data throughout the supply chain to predict spoilage more accurately.We find that consumers disprefer a smart label compared with a static label and would therefore need to receive an incentive valued at $0.86 to choose the smart label.This result sharply contrasts the findings in Yin et al. (2020), who found that WTP increases for an organic and traceable product.However, their setting was very different; they analyzed shrimp in Asia, where a food safety outbreak related to this product occurred, which may have increased people's acceptability toward traceability (Yin et al., 2020).One possible explanation for this result may be related to milk consumers' motivations to choose the smart label.Previous studies find that most QR code applications focus on providing information related to safety (Wu et al., 2020), sustainability (Atkinson, 2013), and loyalty, sometimes coupled with gamification (Okazaki et al., 2013).In our study, the smart label provides up-todate shelf life information, which consumers may not readily identify as a sustainable attribute.This process also adds an additional step to purchasing a low-priced staple product such as milk, so shoppers may need an incentive to offset the hassle cost of use their smartphones while buying dairy milk (Hansen et al., 2021).For example, offering discounts based on the remaining shelf life may encourage shoppers to use the smart label (Lau et al., 2022).Our results therefore do not support H2.
Third, we hypothesized that consumers are willing to pay more for HTST milk with a food waste-related ecolabel (H3).We tested this by including an ecolabel as a product attribute.The WTP estimate for the ecolabel was positive but not significant (P = 0.26).However, the WTP for the ecolabel in the treatment group was $1.12 (P < 0.001), which implies that the information treatment (i.e., the information passage with statistics about food loss and waste; see Figure 2) moderates the WTP for the ecolabel.These results imply that preferences among respondents to decrease food waste are only motivated by context-specific information about the environmental impact of food waste (i.e., H5a).Thus, our results do not support H3.
Fourth, we hypothesized that consumers are willing to pay more for a smart label when it is coupled to a food waste related ecolabel (H4).We tested this by including an interaction effect between the smart label and ecolabel.Although the likelihood to consume milk in the presence of a smart label and an ecolabel is positive, the interaction effect is not significant (P = 0.32).Our results therefore do not support H4.
Lastly, we hypothesized that providing consumers with product-specific information about food waste would increase WTP for HTST milk with an ecolabel (H5a) and with a smart label coupled to an ecolabel (H5b).To test consumers' WTP in the presence of an ecolabel and the information treatment, we included an interaction between the treatment group and the ecolabel.As stated before, that interaction was positive and statistically significant (P < 0.001).This result can be partly explained by recent trends in purpose-driven consumption (i.e., services and products aligned with consumers ' lifestyles and values;Haller et al., 2020).Moreover, purpose-driven consumers are willing to adjust their buying behaviors to diminish their environmental impact and promote sustainable practices (Haller et al., 2020).
We implemented a 3-way interaction to determine the effect of information provision on food waste on consumer preferences for a smart label coupled to an ecolabel.The WTP for estimate for this interaction effect was $0.85 less than the control group (P < 0.05).Though seemingly unexpected, this result implies that for consumers that received the information treatment, their positive WTP for ecolabel (the results for H5a) was attenuated by the presence of the smart label (QR code).Our results therefore support H5a but do not support H5b, suggesting that milk with an ecolabel can be sold at a higher price, but that price premium will have to be reduced if both an ecolabel and a smart label (QR code) are present in a given milk container.
As seen in the present study, consumers favor HTST milk with longer shelf life.However, our study indicates that shoppers prefer not to use a QR code (i.e., smart label).Several reasons may explain this behavior.First, the benefits of using the more precise date label may be outweighed by the additional effort required to obtain that information.For instance, scanning the milk container to get an accurate date adds an additional step to consumers' shopping routine.Moreover, it requires that consumers use a smartphone to access the embedded date or a specific application to read the QR code.Furthermore, consumers' likelihood to purchase sustainable products increases when ecolabels are coupled with targeted information (i.e., analogies; White et al., 2019).
Our study found that the WTP for organic milk compared with conventional milk was not statistically significant in either the control or treatment groups.Our results differ from those of previous literature, which found premiums for organic milk (Wolf et al., 2011;Adesina and Zinnah, 2015;Gayle et al., 2022).However, our findings may be partially explained by consumer heterogeneity (Hasselbach and Roosen, 2015;Li and Kallas 2021), perception of price (Kim et al., 2018;Wu et al., 2020), andtrust (Nuttavuthisit andThøgersen, 2017;Curvelo et al., 2019).Although younger consumers have a positive attitude toward organic claims, studies have shown that older consumers are far more likely to purchase organic products (Hasselbach and Roosen, 2015).In fact, of those consumers willing to pay more for organic, the vast majority may be predominantly female, older, and married with at least one child living at home (Li and Kallas, 2021).Our sample largely consisted of single, young shoppers liv-ing with roommates, a sample that may not represent typical organic shoppers.Moreover, Kim et al. (2018) did not find significant price premiums for organic milk among price-conscious consumers in the United States and concluded that roughly a third of consumers would choose organic instead of conventional milk if prices were the same (Kim et al., 2018).Lastly, it could also be the case that other attributes such as the ecolabel are perceived by consumers as substitutes for the organic attribute.
Our findings suggest that consumers' value perceptions of extended shelf life and an ecolabel are positive, increasing their likelihood of buying HTST fluid milk.That said, although shoppers value the extended shelf life, their preference for the smart label is not favorable.Our findings suggest that retailers should find alternative means to enhance the communication of greater shelf life and may need to ease consumers into the use of smart labels (e.g., using intelligent shelves that provide information without use of a smartphone, gamifying the experience through rewards, implementing loyalty programs).
The previous results have several managerial and policy implications.First, a positive preference toward extended shelf life and the ecolabel implies that retailers can charge premiums for HTST milk with a longer shelf life (i.e., $0.08 per additional day of shelf life).Moreover, to the extent that these retail premiums are reflected in wholesale prices, milk processors would have a profit incentive to invest in technology to extend shelf life.Then, as milk nears its date label, retailers could discount these products while framing it in the context of food waste reduction to increase consumers' perception of value.Given that ecolabels positively affect consumers WTP, retailers could use them to partially offset the discount for milk nearing its date label since customers' attitudes would be more positive if framed around food waste reduction.

Study Limitations
There are several limitations of our study that warrant consideration.First, given the ubiquity of QR codes in restaurants, hotels, and television and print ad campaigns since the onset of the COVID-19 pandemic, it is likely that consumer preferences for QR code labels are greater now than when these data were collected and tested.Second, we selected fluid milk as a focal product because of its impact on food waste and its relevance for retailers; however, choosing a higher-margin product may elicit stronger consumer preferences.Similarly, given that outbreaks related to pasteurized milk are not common in the United States, choosing a product with a higher food safety risk may also provide stronger incentives for consumers to use smart label technology.
Another limitation of our study stems from our use of a hypothetical instrument to elicit consumer WTP.Nonincentivized elicitation mechanisms may lead to inflated WTP estimates for attributes that provide a social benefit (e.g., food waste reduction), as respondents may choose more socially beneficial products due to a social desirability bias (Grimm 2010).In addition, our results may not fully reflect retail purchase behavior due to the online context of the experiment.An online setting fails to capture all the nuances of a brick-and-mortar retail shopping environment, and an online-recruited survey sample may not be fully representative of the relevant population of retail fluid milk consumers.Furthermore, to facilitate choice in our experimental design we explained the purpose of a smart label to participants (Figure 3), information not typically provided to consumers in a retail environment.The external validity of our results is therefore conditional on the assumption that retail customers understand smart labels, which could be achieved through an in-store display.

Future Research
Our findings also suggest several fruitful avenues for future research.First, our exploratory analysis of consumer heterogeneity motivates future studies that leverage more sophisticated methods of cluster analysis to fully characterize relevant customer segments for HTST milk.Also, to take advantage of consumer WTP for greater shelf life, retailers may be able to implement a shelf life based dynamic pricing schedule to charge premiums for HTST milk with a longer shelf life.Furthermore, retailers could use ecolabelling to frame lower prices in the context of food waste for milk nearing its sell-by date, thereby increasing consumer value perception.Given that consumers are willing to pay more for an ecolabel, this may help retailers design a revenueneutral or revenue-enhancing solution that improves inventory management while reducing food waste.Finally, more research is needed in retail store settings using real transactions to build upon existing consumer acceptance studies (Lau et al., 2022) and estimate WTP of the target segment of fluid milk shoppers.

CONCLUSIONS
This paper examines consumer preferences for HTST fluid milk shelf life and smart date labels and tests how information about the environmental impact of fluid milk food waste affects consumers' acceptance and WTP.Our findings suggest that milk consumers value greater shelf life and an ecolabel and are willing to pay more for these attributes, but they do not prefer smart QR code labels and require a discount to use this technology.

Figure 1 .
Figure 1.Hypothetical milk container design with smart label and ecolabel.

Figure 2 .
Figure 2. Between-subject information treatment.The treated group was presented with this information passage containing statistics about food loss and waste.USDA, 2010.
Figure 3. Information graphic presented to survey respondents prior to the choice-based conjoint study detailing product attributes and levels.
Figure 4. Example of a choice task presented to respondents in the choice-based conjoint study.Each respondent faced 4 such choice tasks in the survey.
Endara et al.: WILLINGNESS TO PAY FOR FLUID MILK SHELF LIFE

Table 1 .
Milk product attributes and levels used in the choice-based conjoint study

Table 2 .
Endara et al.: WILLINGNESS TO PAY FOR FLUID MILK SHELF LIFE Milk pricing data

Table 3 .
Menu of prices to implement in the choice-based conjoint study 1Minimum value from sample prices. 2

Table 4 .
Endara et al.: WILLINGNESS TO PAY FOR FLUID MILK SHELF LIFE Demographic characteristics 1 1Values are presented as n (%).2 Reference level for further analyses.Endara et al.: WILLINGNESS TO PAY FOR FLUID MILK SHELF LIFE

Table 5 .
Behavioral characteristics 1 Endara et al.:WILLINGNESS TO PAY FOR FLUID MILK SHELF LIFE Data presented as n (%).The responses for some variables may not add up to 498, due to the voluntary nature of survey question responses.
1 2 Reference level for further analyses.3QR code = quick response code.

Table 7 .
Endara et al.: WILLINGNESS TO PAY FOR FLUID MILK SHELF LIFE Choice-based conjoint model and willingness to purchase (WTP) Data presented as coefficient (SE) [95% CI, calculated using the Delta method].Willingness to pay (WTP) captures the predicted market value and relative importance of each attribute.
1 Data presented as coefficient (SE) The coefficients represent the utility for each attribute level. 2