Consumers preferences on nutritional attributes of dairy‐alternative beverages: hedonic pricing models

Abstract Dairy products, especially milk play a crucial role in assuring dietary quality for U.S. households. However, due to taste, nutrition, health and environmental concerns, households increasingly prefer to consume dairy alternative beverages instead of conventional milk in the U.S. This work is motivated by the need to take into consideration of intrinsic characteristics and differences of such characteristics when analyzing the changes of consumers' purchasing behavior of and willingness to pay for dairy alternative beverages and conventional milk products. After aggregating and organizing the purchase data of Nielsen Homescan and first‐hand nutrition data, this study estimates both linear and semi‐log hedonic pricing models. The results show that consumers exert the highest weights and assign highest evaluation on such qualitative characteristic as nutritional attributes which include calories, protein, fat, vitamin A and vitamin D in which protein is the most valued attribute and other characteristics such as package size, multi pack and brand. The hedonic pricing order and value of these qualitative characteristic are indicative of consumers' purchasing behavior and thus provide essential information for manufacturers to better differentiated their products and develop products catering to consumer's preferred attributes.


| INTRODUC TI ON
In the past decade, dairy alternative beverages have gained its market position as a robust competitor for conventional milk in the United States. Consumers have gradually turned away from conventional milk, leading the push towards plant-based milk products because a growing number of consumers beginning to believe that plant-based foods are healthier and more environmentally friendly than animal-based foods. As indicated by Singhal, Baker, and Baker (2017), the increasing sales trend of non-dairy beverages in westernized counties is due to consideration that foods labeled as natural are perceived to be the most healthy and appropriate nutritional choice ponents. The central question addressed in this article pertains to whether qualitative attributes especially macronutrients and micronutrients have significant effects on consumers' willingness to pay for dairy alternative beverages, what are consumers' subjective evaluations on these attributes and to what extent these attributes influence prices of the products.
As for nutritional contents in the products play an important role in making consumption choices, hedonic pricing model estimation can provide information about consumer's cognition and preference on each qualitative attribute of agricultural products and their willingness to pay for each attribute. Consumers' willingness to pay for purchasing the products (essentially the combination of different qualitative characteristics) and the satisfaction they received from consumption are greatly related to the companies' marketing strategy and selling behavior. Based on the assumption that consumers' utility is gained through consuming the intrinsic properties of a particular good rather than the simple quality and that intrinsic characteristics are combined to constitute the product's market price, this paper contributes to provide manufacturers information about what and how qualitative advancement and differentiation can be made to produce and market better dairy alternative beverages to cater to consumers' preferences. This is of great help to enhance market competitiveness and to expand market share of dairy alternative beverage companies. Also, the results of the study would have important implications for the targeting of nutrition education programs.

| LITER ATURE RE VIE W
Some existing literature about the nutritional components of dairy alternative beverages are mainly review articles. For example, Vanga and Raghavan (2018) outlined the differences of nutritional contents among various dairy alternative milks (including almond milk, soy milk, rice milk) and cow's milk and through comparison noted that nutritionally soy milk is the best alternative for replacing cow's milk in human diet. Sethi, Tyagi, and Anurag (2016) introduced the functional components of diary alternative beverages and their health benefit of different products appeared in the market and the technological interventions that should be made to improve the quality and acceptability of plant-based milk alternatives. Mäkinen et al. (2016) gave an overview on the technology of production, nutritional properties, consumer acceptance and environmental impacts of dairy alternative beverages. Verduci et al. (2019) reviewed the different compositions in terms of macronutrients and micronutrients of milk from different mammalian species, including special milk formulas indicated for cow's milk allergy, and of dairy alternative beverages.
The empirical research about consumer preference on dairy alternative products embraces Laassal and Kallas (2019) who applied revealed preference discrete choice experiment to analyze consumers' preferences toward dairy-alternative products in Catalonia using Home-Scan data of 343 households and the results showed that price was the major driving factor, followed by the original non-dairy beverage flavor attribute. With the elevated demand on plant-based alternative milk beverages in U.S., Dharmasena and Capps (2014) estimated the demand for soy milk, white milk, and flavored milk.
In addition, Copeland and Dharmasena (2016) analyzed demand for dairy alternative beverages and the effect of increased demand for those products on dairy farmers' welfare.
The concept of food hedonic pricing is first introduced by Waugh (1928) for analyzing the prices of vegetables. He argued that prices of vegetables are closely related to the sizes, lengths, ingredients and other characteristics. Rosen (1974) provided mathematical proof for hedonic pricing model and showed that the intrinsic value of products can be calculated based on econometric methods, thereby analyzing the demand for the bundle of characteristics of certain products. In the year 1966, hedonic pricing models received a great progress. Lancaster (1966) proposed that product attributes (or characteristics) with which the good possessed give rise to utility and not just the quantity of the consumed good. Epple (1987) argued that in the empirical investigation of hedonic models, one issue of interest is to determine how the price of a unit of the commodity varies with the set of characteristics or attributes it possesses.
Hedonic approach has been applied in many research areas to measure consumer's willingness to pay for the products. For example, Ghali (2020) applied structural equation modeling to explore the influence of organic food perceived values (utilitarian vs. hedonic) on consumer willingness to buy and willingness to pay for organic oil in a developing country and found that both utilitarian and hedonic values have significant influence on consumer willingness to buy and to pay for organic olive oil. Nepal, Rai, Khadayat, and Somanathan (2020) used hedonic pricing model to analyze the characteristics that affect consumer purchasing decisions on house units in Nepal based on sub-sample of nationally representative household survey data from urban areas as well as primary data collected from one of the metropolitan cities. Bonanno (2016) used a hedonic price model and 2 years of weekly sales data of yogurts in eight Metropolitan U.S. markets to assess the market value of several health and nonhealth-related attributes of yogurt, accounting also for their differences across markets. Even though hedonic pricing method has been widely applied in the area of agricultural commodities and other differentiated products, little work has been done to examine the link between the quality attributes and price differentials to explore the pricing mechanism of dairy alternative beverages and conventional milk products. Furthermore, few studies organize and pool the purchase data from Nielsen Homescan in a way that it could not only capture enough qualitative information about the purchased products as well as time effects but also merge with the nutritional data just available from the products' nutrition facts label. Therefore, given the lack of research on dairy alternative beverage market and application of hedonic pricing model to analyze consumers' preference and pricing mechanism of milk alternative beverages, we attempt to (a) develop linear and semi-log hedonic pricing models for almond milk, soy milk, rice milk and four types of conventional milk (1% fat, 2% fat, fat-free milk, and whole milk); (b) conduct statistical analysis on all the qualitative characteristics fitted in hedonic pricing models; (c) examine the effect of different characteristics on prices and summarize consumers' preference toward these characteristics.
The organization of the rest of this article is as follows. Section 3 focuses on introducing the methodology applied in this work. We estimate hedonic pricing models, where prices are defined as a function of products' qualitative characteristic. Section 4 focuses on discussing how the data is acquired and organized for this work.
Section 5 shows the estimated results of hedonic pricing models.
Section 6 offers concluding remarks, research limitations and some interesting future research topics.

| HEDONI C PRI CING MODEL DE VELOPMENT
Hedonic pricing models assume that the consumer maximizes utility by selecting products that maximize the sum of the utilities derived from each attribute (Rosen, 1974). Therefore, the price of each beverage in this study can be explained by the set of attributes of the product. X = (X 1 ,X 2, …,Xj) represents the qualitative characteristic combination. Qualitative characteristic information has close relationship with prices and hedonic pricing model thus is shown as: where is the error vector, and P is the observed price. If the relationship between prices and attributes is assumed to be linear, price of a good i can be derived as the sum of the attribute values (Ladd & Suvannunt, 1976). Thus, the total value of each attribute is equal to the quantity of the attribute multiplied by the implicit price of that attribute (Gulseven & Wohlgenant, 2015). The linear and semi-log hedonic pricing models are constructed as follows: where A ij is the amount of nutritional attribute j contained in product i. X ik is other factors that might affect prices. P i is the monthly average price recorded in Nielsen database by different Universal Product Codes (UPC) that have been purchased from the year 2004 to 2015.
These nutritional attributes include calories, fat, fiber, protein, calcium, vitamin A, etc. If the price attribute relationship is assumed to be in a semi-log form (Nimon & Beghin, 1999), then instead of price, the log-price of the product is defined regarding attributes as is shown in Equation (3). Similarly, as shown in Equation (2), P i is the monthly average prices of a beverage from year 2004 to 2015. The implicit prices are the coefficients to be estimated which are represented by β j and D k . In the linear hedonic pricing model, the implicit prices or shadow prices can be shown as: Marginal effect of semi-log hedonic pricing model is derived as follows. First, solve for P i from Equation (3):

packs
Note: 1. multi its value represents number of units in multipack; 2. multi packs (i.e. "multi">1) is total units for a product; 3. size1_amount is package size (numeric size of the product).
a The base category of package size dummies and multi-pack dummies for almond milk, soy milk, rice milk, 2% mill, 1% milk, whole milk and far-free milk; The base category of multi-pack dummies for the seven products is D multi1 . b For package_size dummy variables that are in ranges, for example size1_amount < 8 oz., they are created because there are many package sizes that are not integers and for different types of beverages, the values vary a lot. In order to make the package sizes comparable from one beverage to another, we created some package size dummies that are in ranges. c 1 is single serve.  (9) and (10)  of each quality attribute available in the product.

| DATA
As shown in Tables 2, the  Three types of dairy alternative beverages (almond milk, soy milk and rice milk) and the four most common types of milk products (whole milk, 1% milk, 2% milk and fat free milk (or skim milk) are included in this work and monthly average price variable is acquired as follows. Monthly average price is the "unit price paid" as shown at the bottom in Figure 1. First, we obtain each product's information from products files; and then we merge the information with trips files to acquire the dataset, which include variables of quantities sold, total price paid by consumers, coupon value, deal_flag_uc, multi_pack, product's package size and size unit. 1 As Figure 1 shows, the unit price paid (per unit cost) is calculated by first dividing the final price paid by the quantity variable. Final_price_paid is calculated by subtracting the value of variables "coupon value" from the value of "total_price_paid". Then, we average the unit prices paid in each month in each year to get the monthly average price per oz. and multiply by 8 to get monthly average price per 8oz. (unit monthly average price).
Obtaining data on nutritional information of dairy milk alternative beverages is one of the biggest concerns in estimating the  Besides the price variable and nutritional variables, we attempt to find out other factors that might exert impact on the prices of those products. The first group of such variables includes package size and multi-pack. 2 We also consider that the available deal/coupon when consumers purchase the products should affect prices.
Therefore, we select two variables which are deal_flag_uc and coupon_value. "deal_flag_uc" is a dummy variable which indicates if the panelist received a deal. Also if the panelist used a coupon, they enter the amount discounted. If coupon_value and deal_flag_uc are both zero, there is no deal on the purchase. In addition, in order to take into consideration of the time effects on the prices, we add yearly dummies. Since the format of original data file in which households' purchase information is recorded by their trip date, it is very common that the purchase may happen many times each month or no purchase activities within a month. That's one reason why we aggregate the data into UPC level. Another imperative variable considered to affect price is brand. Therefore, a brand dummy is added which equals 1 if it is a store brand and equals zero if it is a national brand.

| EMPIRI C AL RE SULTS AND D ISCUSS I ON
Applying the model developed in Section 3 and using the data file we constructed in Section 4, we acquired estimates for all the variables considered for each product. The results of linear and semi-log hedonic regressions are shown in Tables 3 and 4 respectively. In this section, we discuss the empirical results derived from hedonic pricing models in detail and compare and contrast our estimations with those observed in the extant literature.   In terms of the effect of fat content, Gulseven and Wohlgenant (2015) found that lipid fat contributes 0.861 to prices of milk products considered in their study. Comparatively, in linear hedonic pricing models, fat content contributes negatively to unit price (monthly average price per 8 oz.) in our study with estimated coefficient  (2015), Americans were advised to get the most nutrition out of their calories and to make smart, nutrient-dense choices from every food group. Taubes (2008) argued that there are good and bad calories; the key to good health is the kind of calories we take in, not the number. As shown in Table 3, vitamin A contributes positively to prices of soy milk and almond milk, with coefficient being 0.0029 and 0.073 respectively, but it contributes negatively to rice milk. It is interesting to find that vitamin D, however, has the mirror effect on these three beverage types. These results can be possibly explained by consumers different attitudes toward these two vitamins. By analogy with Bonanno (2016)'s study which shows that fiber, a health-related attribute in food products, is perceive unfavorably by yogurt consumers if the yogurt is enriched and fortified with fiber, even though vitamin A and vitamin D are considered as a beneficial nutrient, consumers might have negative attitude toward them when they are artificially enriched in dairy alternative beverages. Also, Willett (2013) indicates that when it comes to vitamins and minerals, the notion of "the more, the better" is incorrect since nutrients can be harmful when taken in amounts above what is considered beneficial and multivitamin is one of them. The positive and significant effect of protein and calcium on prices manifest consumers' favorable acceptance of these nutritional attributes . Comparing different weights of nutritional   variable from Tables 3 and 4, we can witness that among all the nutritional variables considered, protein has highest weight meaning it is regarded by consumers as the most preferred qualitative characteristics for soy milk and almond milk and calorie is least valued by consumers.
The dummy variable "Brands" has negative sign and is significant as expected, indicating that prices of private label products are lower than that of national brand products. Packaging size and shape are also significant factors in designing the package and a decision-making instrument (Ksenia, 2013

| CON CLUS I ON S , LIMITATI ON S AND FUTURE RE S E ARCH
In terms of rate of growth, dairy alternative beverage market in the United States has surpassed the growth of conventional milk market in recent years. The ongoing competition between dairy alternative beverages and conventional milk is expected to intensify over the next several years as consumers become more comfortable with milk alternative beverages and criticism of dairy foods continues to grow. This work focuses on analyzing the relationship between the qualitative characteristics embedded in the differentiated dairy alternative beverages and their market prices.
We take product characteristics approach, specifically hedonic pricing model to study consumers' preference and their willing- This analysis does however show limitations. Due to data limitations, we can only use pooled UPC level information to estimate the hedonic pricing models. Because milk alternative beverages are starting to gain ground in the recent years, adequate purchase observations were not available at the beginning of the time period pertaining to this study. In addition, we need variations on the nutritional attributes, but household level data cannot guarantee enough variability. Therefore, we consider pooled data which can not only capture variability of the nutritional attributes but also enable us to expand the time period to be considered. Data limitations have also constrained our selection of related dairy alternative beverages from which we can only include soy milk, almond milk and rice milk. Besides, the information about nutritional data is very scarce and limited for dairy alternative beverages in the Nielsen Homescan database and also in the USDA nutritional database. Therefore, bulk of the nutritional data was collected from product labels. The estimated results of demand could possibly be more definitive and convincing if data after the year 2010 were used and more nutritional information about diary alternative beverages is available. Another limitation is that the analysis performed cannot provide insights on the impacts of demographic variation across markets.
Studying consumer behavior cannot leave without consumer demand analysis. Traditionally, consumer demand is analyzed using demand system such as AIDS, Rotterdam and some modifications to these two models. The basic assumption for these conventional demand models is that consumers' utility is obtained from the quantity of goods they consumed which is also the assumption of hedonic pricing model as aforementioned. However, the conventional demand system estimation is complicated in that the number of parameters we need to estimate is large. Therefore, some innovative method to estimate demand which is based on hedonic pricing model estimation and then reparameterizing the estimators is developed such as Distance Matrix method and Hedonic Metric approach.
The latter is based on hedonic pricing model estimates to estimate demand of milk products. This approach overcomes the shortages of Distance Matrix method but also greatly reduced the number of parameters to be estimated. It is expected that our future work will develop this method and apply it into the analysis of demand on dairy alternative beverages to explore their expenditure, own-price and cross-price elasticities.

ACK N OWLED G EM ENT
Thanks are due to Dr. Dharmasena for assistance with the accomplishment of this paper.

CO N FLI C T O F I NTE R E S T
We declare that we do not have any conflict of interest.

E TH I C A L R E V I E W
This study does not involve any human or animal testing.

I N FO R M E D CO N S E NT
Written informed consent was obtained from all study participants.

E N D N OTE S
1 size1_units is the unit of measure. For example, "size1_amount" might be "16.0", and " size1_units" might be "OZ."; coupon is total discount for amount due to coupon; deal_flag_uc is presence of a deal (1 = deal, 0 = no deal).
2 These variable names used in this study are referenced to the original variable names in Neilson HomeScan dataset.

R E FE R E N C E S
Aggarwal, A., Rehm, C. D., Monsivais, P., & Drewnowski, A. (2016). Importance of taste, nutrition, cost and convenience in relation to diet quality: Evidence of nutrition resilience among US adults using