Impact of Point-of-Sale-Systems (POS) in Traditional and Fast Food Restaurants

Impact of Point-of-Sale-Systems (POS) in Traditional and Fast Food Restaurants

Grocery retailers are joining the fray against obesity by offering a wide range of health and wellness programs at the point of sale. However, the success of such programs in promoting healthy choices remains an open question. The authors examine the effectiveness of a growing health and wellness initiative: a simplified nutrition scoring system. They present a conceptual framework that predicts the effect of such a scoring system on shoppers’ food decisions and their sensitivity to price and promotion, as well as the moderating influence of category-level factors. Using a large-scale quasi experiment and panel data across eight product categories for more than 535,000 members of a grocery chain’s frequent shopper program, the authors demonstrate that the point- of-sale nutrition scoring systemhelped consumersmake healthier food choices, such that they switched to higher-scoring products in the postrollout period. The results also reveal that shoppers became less price sensitive and more promotion sensitive following the introduction of the food scoring system. The authors discuss implications for research and practice.

Keywords: point-of-sale nutritional information, healthy eating, food purchases, obesity, in-store decision making

Healthy Choice: The Effect of Simplified Point-of-Sale Nutritional Information on Consumer Food Choice Behavior

TheNutritional Labeling andEducationAct (NLEA)was one of policy makers’ most important efforts to promote healthy food choices among Americans. Unfortunately, it had little effect in curbing the obesity epidemic; more than 20 years after the act was implemented, Americans are still struggling with their expanding waistlines. Indeed, the United States is the most obese country in the developed world (Organisation for Eco- nomic Co-operation and Development 2012), with obesity rates reaching epidemic proportions (Flegal et al. 2012). Moreover, with obesity-related health care costs projected to exceed $344 billion by 2018 (Thorpe 2009), obesity is considered one of the

costliest health conditions of the twenty-first century. Indeed, obesity is now officially classified as a disease by the American Medical Association (Pollack 2013).

Given the NLEA’s failure to improve Americans’ eating habits, public policymakers, consumers, andmarketers alike are still seeking effective ways to combat obesity. For example, the recent public policy initiatives for solving the problemof obesity include the “Let’s Move” campaign, launched by First Lady Michelle Obama; the first-ever White House Task Force on Childhood Obesity (Letsmove.gov); and proposed legislation to impose taxes on all sugary drinks in California, Vermont, and Texas (Sankin 2013).

Consumers themselves have also begun to take actions to improve their eating habits. Nutrition has increasingly become a driving factor in grocery shopping decisions. Maintaining or losing weight was the top health concern reported by consumers in 2010, and 66% of grocery shoppers indicated that they were trying to improve their health by purchasing healthier foods (Catalina Marketing 2010). Relatedly, approximately 70% of shoppers have reported an interest in nutritional information and a will- ingness to pay more for healthier organic products (Food Marketing Institute [FMI] 2012c).

*Hristina Dzhogleva Nikolova is the Coughlin Sesquicentennial Assistant Professor ofMarketing, Carroll School ofManagement, BostonCollege (e-mail: hristina.nikolova@bc.edu). J. Jeffrey Inman is Associate Dean for Research and Faculty and the AlbertWesley Frey Professor ofMarketing, JosephM.Katz Graduate School of Business, University of Pittsburgh (e-mail: jinman@katz. pitt.edu). The authors thank EllieWilson, MS, RDN, CDN, for her instrumental contribution to this research. They also thank Catalina Marketing and the Kilts Center at the University of Chicago for providing the purchase data as well as NuVal LLC for providing the nutrition scoring information. The authors especially thank the JMR review team for their constructive comments and guidance. Rik Pieters served as associate editor for this article.

© 2015, American Marketing Association Journal of Marketing Research ISSN: 0022-2437 (print) Vol. LII (December 2015), 817–835

1547-7193 (electronic) DOI: 10.1509/jmr.13.0270817

In response, grocery stores have changed their product offerings and in-store environments. An increasing number of food retailers have implemented a variety of health and wellness initiatives to help shoppers improve their eating habits. For example, 98% of grocery chains in the United States provide health and wellness information on their websites, 79% offer store tours to shoppers to help them select healthier foods, 59% have registered dietitians on site, and 28% offer weight management classes (FMI 2012b).

In-store nutrition scoring and labeling systems that communicate the nutritional value of foods in a simplified manner are key components of food retailers’ efforts to help shoppers make healthier food choices. Figure 1 shows examples of the three most popular simplified nutrition scoring systems at the point of sale (POS): Guiding Stars (available in approximately 1,500 U.S. stores), NuVal (introduced in 1,650 U.S. stores), and Traffic Light Label (used in several European countries). However, as more food retailers consider licensing one of these scoring systems, there is little hard evidence as to whether POS nutrition scoring systems are effective in promoting healthier consumer choices. This is the objective of our research.

In this article, we examine the effectiveness of a sim- plified POS nutrition scoring system. Such POS nutrition scoring systems differ from the NLEA-mandated labels in that they do not list the amounts of all nutrients but rather summarize the nutritional information and communicate it to consumers in a simple, easy-to-process format (see Figure 1). According to Nielsen (2012), 59% of grocery shoppers admit that they experience difficulty in un- derstanding the nutritional facts on product packaging. Given that nutrition scoring systems reduce the complexity of nutrition information, it is important to examine whether they are indeed successful in helping consumers make healthier food decisions. Furthermore, we examine the effect of POS nutrition scores on shoppers’ sensitivity to two key marketing-mix elements—price and promotion—as well as the moderating roles of category characteristics. Given that the decision of whether to implement a POS nutrition scoring system to improve shoppers’ food choices does not occur in a vacuum and must be coordinated with other

strategic goals and efforts, it is important for grocery retailers to understand the potential impacts of POS nu- trition scores on their strategic initiatives for pricing and promotion.

Our research makes the following four key contributions. First, we demonstrate that the introduction of a simple POS nutrition scoring system helps consumers make healthier food choices. We find that as a result of the implementation of a food scoring system, consumers switch to healthier alternatives and thus improve the nutrition content of their purchases by 21.8% on average across the eight product categories. Second, we show that the extent of improve- ment in the nutrition content of shoppers’ purchases varies as a function of the healthiness of the product category such that it is stronger in healthier product categories. Third, we demonstrate that the introduction of a POS nutrition scoring system alters the effectiveness of two key marketing-mix elements, price and promotion. Specifically, shoppers be- come less price sensitive and more promotion sensitive as a consequence of the introduction of the POS nutrition scoring system. Fourth, we also provide evidence for the moderating influence of category healthiness and the variation in nutrition scores across the category assortment on the changes in price and promotion sensitivity. The attenuation in price sensitivity is greater in categories that are healthier and have lower within-category variability in nutrition scores, while the increase in sensitivity to pro- motions is greater in product categories that are less healthy and have greater nutrition score heterogeneity.

The remainder of the article is organized as follows. We first briefly review prior research on the effect of nutrition information on purchase behavior. We then develop our predictions regarding the impact of a POS nutrition scoring system on shoppers’ food choices and their sensitivity to price and promotion, as well as the moderating influence of product category characteristics on those effects. We test these predictions in a large-scale quasi-experimental field study across eight product categories, using purchase data for over 535,000 members of a frequent shopper program (FSP) at a U.S. grocery chain. After presenting the sta- tistical methodology and reporting the results, we highlight our theoretical contributions and discuss the implications of our findings for practitioners and consumers.

Figure 1 EXAMPLES OF THE MOST POPULAR SIMPLIFIED POS NUTRITION SCORING SYSTEMS

Guiding Stars NuVal Traffic Light

Each portion contains

of your guideline daily amount

Calories Fat Sugars SaltSaturates

HIGH HIGH MED LOW

32% 28% 7% 3%

Kcal

272

22.1g 5.5g 6.1g 0.2g

818 JOURNAL OF MARKETING RESEARCH, DECEMBER 2015

THEORETICAL DEVELOPMENT

Nutrition Information and Consumers’ Food Choices

The NLEA was signed into federal law in 1990 and mandated all food manufacturers to disclose the nutritional content of food products in a Nutrition Facts panel on their packaging. The act was considered one of the most im- portant policy initiatives designed to help consumers make healthier food choices and combat the obesity epidemic through the provision of standardized and consistent nu- tritional information at the POS. Unsurprisingly, the implementation of the NLEA has attracted the attention of the academic community as well (e.g., Balasubramanian and Cole 2002; Keller et al. 1997; Mathios 2000; Moorman 1996; Variyam and Cawley 2006). Unfortunately, con- sistent with the rising obesity levels in the United States over the last two decades, the prevailing consensus among researchers is that the NLEA was at best only partially successful in improving consumers’ food decisions and diets (e.g., Balasubramanian and Cole 2002; Moorman 1996; Variyam and Cawley 2006).

One potential reason why the NLEA was ineffective is that the nutritional labels are somewhat difficult and time- consuming to understand. As mentioned previously, 59% of consumers have trouble comprehending the information on the Nutrition Facts panel (Nielsen 2012). Therefore, consumers prefer to use simple heuristics to gather in- formation about the nutrition content of the products they consider purchasing. That is, they rely more on the easy-to- process information presented in the form of descriptor terms (e.g., low fat, high fiber) or health claims (e.g., “Oatmeal helps reduce cholesterol!”) and ignore the comprehensive Nutrition Facts panel (Balasubramanian and Cole 2002; Roe, Levy, and Derby 1999). Researchers have studied these types of simplified nutrition information extensively (e.g., Andrews, Netemeyer, and Burton 1998; Garretson and Burton 2000; Levy, Fein, and Schucker 1996) and have demonstrated that, in general, they have favorable effects on consumers’ food choices (Berning and Sprott 2011; Kozup, Creyer, and Burton 2003).

However, researchers have also warned that the use of verbal descriptions and health claims might be dangerous because such claims do not integrate all of the relevant information (Berning, Chouinard, and McCluskey 2008; Wansink and Chandon 2006). For example, “low-fat” nu- trition claims are based solely on the grams of fat contained in the food and do not consider the number of calories. Wansink and Chandon (2006) find that, on average, consumers are misled by such “low-fat” claims to increase the amount of snack foods they consume by approximately 50%.

The nutrition scoring systems we examine in our research are another tool for reducing the complexity of nutrition information for consumers. The similarity between the nutrition scores and the front-of-package health claims and verbal descriptions is that they all communicate the product’s nutrition content in a simple and easy-to-digest format. Unlike the health claims and verbal descriptors, the food scores are based on a more comprehensive combi- nation of nutrients and thus represent a single summary indicator of the food’s overall healthiness. However, there is a paucity of research examining the effectiveness of these nutrition scoring systems.

Two exceptions are the works of Feunekes et al. (2008) and Sutherland, Kaley, and Fischer (2010). Feunekes et al. study a series of six front-of-pack nutrition labeling formats that vary in complexity, evaluating them on their consumer friendliness (e.g., comprehension, liking) and their effec- tiveness in changing consumers’ purchase intentions for unhealthy products. Their findings suggest that the six labeling formats did not differ in terms of friendliness; more importantly, the authors also show that the front-of-pack labels were successful in increasing (decreasing) purchase intentions for healthy (unhealthy) products.

Relatedly, Sutherland, Kaley, and Fischer (2010) examine the effect of a POS nutrition scoring system (Guiding Stars) on aggregated food and beverage sales. They demonstrate that the POS nutrition scoring system implementation slightly increased the sales share of products with a star rating and decreased the sales share of no-star products. Moreover, the purchase of more cereals with star ratings and fewer no-star products was associated with a decrease in sugars and an increase in dietary fiber in purchased products (Sutherland, Kaley, and Fischer 2010). We build on this work in four key ways. First, we use FSP data to examine the change in nu- trition content at the household level (vs. the aggregated sales data used by Sutherland, Kaley, and Fischer 2010). Second, we examine the effect of the POS nutrition scoring system on shoppers’ sensitivity to price and promotion. Third, we assess the moderating influence of category characteristics on nu- trition content improvement in shoppers’ purchases and on the changes in shoppers’ price and promotion sensitivity. Finally, in contrast to Sutherland, Kaley, and Fischer, who combined products with one, two, and three stars into a single group of “products with star rating” and evaluated the changes in sales of those products versus no-star products, we examine the effect of nutrition scores in continuous form. We present our hypotheses in the following subsections (for a summary, see Table 1).

Impact of POS Nutrition Scoring System on Nutrition Content of Shoppers’ Purchases

Research has suggested that for nutritional information to be incorporated into consumers’ decisions, it must be not only available at the POS but also easily “processable” (Bettman 1975; Russo, Krieser, and Miyashita 1975). Russo et al. (1986) argue that consumers face three types of costs in incorporating nutritional information into their food decisions: collection costs (time and effort expended to acquire the information), computation costs (effort in combining the gathered information into an overall eval- uation), and comprehension costs (effort necessary to un- derstand the nutritional information). Levy and Fein (1998) highlight that shoppers primarily struggle with the com- putation and comprehension costs and suggest that dietary guidance will be more useful if it does not involve quan- titative tasks.

The NLEA-mandated Nutrition Facts panel largely eliminated consumers’ collection costs because it made all the nutrition information available at the POS, but it ar- guably had less impact on their computation and com- prehension costs. The consumer still has to pick up the product; examine and comprehend the nutritional label, which reports information on many types of nutrients (e.g., total fat, sodium); and combine all the information into one

Healthy Choice 819

overall evaluation of how (un)healthy the product is. Furthermore, comparing the nutrient density of different products using the Nutrition Facts panel is a difficult and time-consuming task (Berning et al. 2008).

Unlike the nutrition labels, the POS nutrition scoring systems reduce all three types of costs. They are readily available at the shelf and thus reduce the collection costs. The scores are a summary evaluation of the products’ healthiness, thereby largely eliminating the computation costs. Moreover, the scores are easy to understand and almost effortless to use in comparing foods, greatly re- ducing the comprehension costs as well. Russo et al. (1986) suggest that reducing all three types of costs leads to greater reliance on the nutritional content attribute in making food choices. Furthermore, the display of food nutrition scores at the POS makes nutritional content more perceptually sa- lient, which should also increase its use by consumers in making purchase decisions (Bettman, Payne, and Staelin 1986; Raghubir and Krishna 1996, 1999). Thus, we expect that the implementation of POS nutrition scoring systems will lead consumers to switch to healthier alternatives within the category.1 Formally, we propose the following:

H1: Introducing a simplified POS nutrition scoring system leads shoppers to switch to healthier products within the category.

Furthermore, we predict that the strength of this effect will be moderated by the product category healthiness, such that it will be stronger in healthier categories. This prediction derives from research showing that consumers are more likely to use nu- trition information in healthier categories (Brucks, Mitchell, and Staelin 1984). Balasubramanian and Cole (2002) also demonstrate that consumers may largely ignore nutrition in- formation for what they call “fun foods” (e.g., ice cream) that primarily satisfy hedonistic needs, to prevent the negative emotions (guilt) thatmay arise fromconsidering howunhealthy the food is (Ehrich and Irwin 2005; Wansink and Chandon 2006); in contrast, consumers are more willing to base their decisions on the nutrition content in healthier categories that meet their health-related goals. Thus, we predict the following:

H2: The effect predicted in H1 is stronger in healthier product categories.

We also argue that the effect of the POS nutrition scoring system on the nutrition content of shoppers’ purchases will be stronger in product categories with greater variability in the products’ nutrition scores. This could be due to various factors. First, greater within-category heterogeneity in the products’ nutrition scores leads to increased diagnosticity and weight of the nutrition content attribute relative to other attributes considered in the purchase decision (Burson, Larrick, and Lynch 2009; Pieters, Wedel, and Zhang 2007). In contrast, greater homogeneity in the products’ nutrition content might prompt consumers to pay less attention to nutritional information because it is no longer an attribute that can help them discriminate between alternatives (Balasubramanian and Cole 2002). Second, in categories with greater within-category heterogeneity in the nutrition scores, consumers have a wider range of alternatives to switch to and thus greater opportunity to find something healthier that satisfies their preferences (Lancaster 1990). Third, shoppers’ utility from switching to healthier alter- natives might be greater when such a switch is made in a choice set containing both relatively healthy and unhealthy alternatives. Dhar and Wertenbroch (2012) show that consumers derive self-signaling utility from the choice context. Specifically, they report that selecting a vice provides more utility when the choice is made from a homogeneous set versus a mixed set, whereas choosing a virtue provides more utility if it is chosen from a mixed versus a homogeneous set. Therefore, selecting a healthier product from a category that contains relatively healthy and unhealthy options provides shoppers with greater self- signaling utility than if the choice is made from a more homogeneous category. Thus, we propose the following:

H3: The effect predicted in H1 is stronger in product categories with more within-category variability in the products’ nutrition scores across the assortment.

Impact of POS Nutrition Scoring System on Shoppers’ Price Sensitivity

Previous research has argued that price sensitivity re- quires an awareness of the distribution of prices, which in turn necessitates a significant amount of time and cognitive effort (Hoch, Kim, Montgomery, and Rossi 1995). There- fore, factors that decrease the amount of effort expended on price search and comparison should in turn reduce con- sumers’ price sensitivity. When a nutrition scoring system is

Table 1 SUMMARY OF HYPOTHESES

Impact of

Impact on

Healthiness of Shoppers’ Purchases Price Sensitivity Promotion Sensitivity

Nutrition scoring system implementation (main effect) + Supported

Supported +

Supported Category healthinessa (interaction of category

healthiness × Nutrition scoring system implementation) +

Supported +

Supported −

Supported Variability of nutrition scores in the categorya (interaction

of category variability × Nutrition scoring system implementation)

+ Not supported

+ Not supported

+ Supported

aThe hypothesis concerns the moderating effect of category variables of the effect of the nutrition scoring system on the outcomes of interest.

1We note that our research examines only purchase shifts that occurwithin the category (e.g., a shopper who used to purchase Brand A frozen pizza switching to a healthier Brand B frozen pizza after the implementation of a POS nutrition scoring system) because it is reasonable to expect that shoppers first look for healthier alternatives within the category before switching to a different category. However, further research should also explore cross- category purchase shifts that may result from the implementation of a POS nutrition scoring system.

820 JOURNAL OF MARKETING RESEARCH, DECEMBER 2015

implemented, it introduces an additional salient piece of information (i.e., another attribute) that consumers need to process while making food decisions: the nutrition scores posted on the shelf tags. Before the implementation of the POS nutrition scoring system, shoppers might not have paid much attention to the nutrition content attribute and might have largely ignored the Nutrition Facts panel, but after the introduction of the scoring system they are less likely to ignore the nutrition content attribute because of its salience.

Because such salient stimuli tend to attract attention and influence decisions (Berlyne 1970), after the imple- mentation of a POS nutrition scoring system, the food scores are likely to draw shoppers’ attention. The products’ nutrition content would become another salient attribute in the in-store decision-making process that requires addi- tional cognitive effort (Shugan 1980). As a result, the examination of the newly available nutrition scores on the shelf tags would increase the cognitive effort involved in comparing the purchase alternatives (i.e., the attributes that shoppers use to compare the products will now include the nutrition score in addition to price, promotions, brand name, and other characteristics that shoppers may use to make a choice). Shoppers have a limited amount of time and cognitive resources to expend on the grocery shopping task and tend to simplify decision making by reducing the number of attributes they consider (Johnson et al. 2012). Thus, we propose that focusing on the nutrition content attribute will divert cognitive resources from the price attribute (Krishna and Raghubir 1997) and thus reduce consumers’ sensitivity to price. Formally,

H4: Introducing a simplified POS nutrition scoring system decreases shoppers’ price sensitivity.

We expect that this effect will be stronger in healthier product categories. As we have discussed, in such cate- gories consumers usually look for products that help them achieve their health-related goals and therefore are even more likely to base their decisions on the nutrition content attribute (Balasubramanian and Cole 2002; Brucks et al. 1984). Consequently, ceteris paribus, we anticipate that shoppers’ sensitivity to price will be further reduced in healthier product categories because in such categories they will be more motivated to expend more effort in examining the nutrition scores of the available alternatives and will place less emphasis on price as a factor in their decision- making process. Formally,

H5: The effect predicted in H4 is stronger in healthier product categories.

We also predict that the reduction in price sensitivity will be greater in categories with more within-category vari- ability in the products’ nutrition scores. In categories with greater variability in healthiness across the available op- tions, shoppers have to examine a wider range of scores, necessitating greater cognitive effort. This in turn di- minishes the amount of cognitive resources available for examination of the price attribute and should further de- crease shoppers’ price sensitivity. Thus, we predict the following:

H6: The effect predicted in H4 is stronger in product categories with more within-category variability in the products’ nutrition scores across the assortment.

Impact of POS Nutrition Scoring System on Shoppers’ Promotion Sensitivity

As noted previously, the introduction of the nutrition scores increases the complexity of the grocery shopping task because it introduces an additional salient attribute on which the various alternatives must be compared (Shugan 1980). Even though the nutrition scores are simple them- selves, they still introduce an additional attribute that shoppers need to process, thus making the decision task more involved. When faced with increasingly complex decision environments, consumers often simplify their decision making by using heuristics or choice tactics (Eagly and Chaiken 1993; Payne 1976; Payne, Bettman, and Johnson 1992). These choice tactics minimize cognitive effort and enable shoppers to solve complex purchase decision problems in a less effortful manner (Hoyer 1984). Examples of such tactics include buying the brand that works the best (performance-based tactic) and buying the brand that is on sale (promotion-based tactic) (Hoyer 1984).

We argue that the more complex decision environment that results from the introduction of nutrition scores on shelf tags should lead to an increased reliance on heuristics that simplify the decision process. Because promotions are usually prominently displayed and highlighted in stores (e.g., end-of-aisle displays, bright colors and large fonts on the shelf tags) and thus do not require a lot of cognitive effort to process (Inman, McAlister, and Hoyer 1990), shoppers should be more likely to use the “what’s on promotion” attribute as a simple heuristic to minimize cognitive effort (Zhang 2006). As a result, they should exhibit greater promotion sensitivity after the introduction of the nutrition scores. Thus,

H7: Introducing a simplified POS nutrition scoring system increases shoppers’ sensitivity to promotion.

Although H4 and H7 might seem somewhat contradic- tory, we note that they converge around the same con- ceptual ideas. That is, the POS nutrition scoring system lowers the emphasis put on price and increases shoppers’ reliance on promotions because both effects decrease the demand for cognitive resources, which are now allocated to the examination of the nutrition scores. Price sensitivity requires the examination and comparison of prices, necessitating a sig- nificant amount of cognitive resources. Therefore, price sensi- tivity should be reduced because, in the postimplementation period, shoppers begin to focus more on the nutrition scores, which diverts attention and resources from the price attri- bute. In contrast, promotions do not require much cognitive resources to process because they are prominently high- lighted in the stores and are very easy to spot; consequently, shoppers should become more promotion sensitive in the postimplementation period because they can use promotions as resource-saving decision heuristics.

We also argue that there will be a lesser increase in sensitivity to promotions in healthier product categories. When consumers shop in healthier categories that are in- strumental to the accomplishment of their health-related goals, they should pay more attention to the nutrition content attribute and thus be less likely to use promotions as choice tactics to simplify the decision-making process or to rationalize the purchase of unhealthy products. In contrast,

Healthy Choice 821

in unhealthy product categories shoppers may use pro- motions as justifications to purchase unhealthy products, thus increasing their promotion sensitivity even further (Mishra and Mishra 2011). We hypothesize the following:

H8: The effect predicted in H7 is weaker in healthier product categories.

Finally, we predict that the increase in promotion sensitivity will be greater in categories with more within- category variability in nutrition scores across the assort- ment. In such categories, the need for simplifying choice tactics will be greater because more cognitive effort will be needed to examine the wider range of nutrition scores. Thus, we propose the following

H9: The effect predicted in H7 is stronger in product categories with more within-category variability in the products’ nutrition scores across the assortment.

We test our hypotheses in a large quasiexperimental field study. Next, we describe our data set andmeasures, present our methodological approach and analysis, and discuss the results.

EMPIRICAL TEST

Description of Data Set

In 2008, a large grocery chain began to implement the NuVal Nutritional Scoring System in over 100 stores. The NuVal system computes a summary nutrition score for each stockkeeping unit on the basis of the food nutrient content as well as each nutrient’s association with different health conditions (e.g., heart disease, diabetes). The NuVal scores range from 1 to 100, such that higher scores signify healthier and more nutritious foods. Currently, the NuVal database contains more than 120,000 scored products (NuVal.com). Concurrent with several large national retailers, the NuVal scores were introduced in the chain’s stores across different product categories at various points of time during 2008–2009 and were prominently displayed to shoppers on the shelf pricing tags.

The grocery chain provided us the dates when the NuVal scores were introduced in each of eight product categories: frozen pizza, tomato products, soup, salad dressing, yogurt, spaghetti sauce, granola bars, and ice cream. The nutrition scores were implemented at different points of time during 2009 (see Table 2). However, the chain’s information re- garding the release dates of the nutrition scores in the eight categories varied in terms of specificity, such that the

introduction dates for six of the categories were more specific and indicated the release month (e.g., April 2009), whereas the dates for the remaining two categories (yogurt and granola bars) indicated only the release quarter (e.g., first quarter of 2009).

We selected these eight product categories on the basis of discussionwith the grocery chain and data availability. First, these eight product categories represent a good mix of cate- gories that vary in their healthiness as well as in the variability of nutrition scores across the product assortment in each category. Second, although we wanted to include categories with high household penetration rates, such as cereal, the grocery chain’s records regarding the exact implementation date of nutrition scores for these products were too vague (e.g., the recorded rollout date for cereal was 2008). Because we wanted to take advantage of the natural quasiexperiment arising from the implementation of the POS scoring system, the lack of specific introduction dates for cereal and other product categories such as salty snacks and cookies prevented us from using them in our analysis.

Our data set, which was provided to us by Catalina Marketing, consists of the weekly purchases of over 535,000 shoppers in each of the eight product categories (with all personally identifying information removed). The shoppers are members of the grocery chain’s FSP. The data set captures each shopper’s weekly purchases during the six months before the NuVal Nutritional Scoring System implementation, as well as their purchases during the six months following its implementation, in each of the eight product categories. We defined the six-month pre- and postimplementation periods using the rollout month of the NuVal scores in each category. In the yogurt and granola bars product categories, in which information regarding the rollout date was specified only at the quarter level, we defined the pre- and post-NuVal periods as the six months before and the six months after the quarter in which the scores were implemented. The fact that the NuVal scores were implemented at different times across the eight product categories helps mitigate concerns that might arise from the use of a quasi-experimental design that other changes oc- curring in the postimplementation period might account for the observed effects. Furthermore, we report results of a control group analysis to rule out an alternative explanation that our findings are due to spurious correlation.

The FSP data set (with personally identifiable infor- mation removed) contains a total of more than 29 million

Table 2 SAMPLE COMPOSITION

Scores Release Date

Number of Purchases

Category Average NuVal Score

SD of NuVal Scores of all UPCs in the Category

Average Price per Ounce

% of Purchases with Promotion

% D in Shoppers’ Volume-Weighted NuVal Scores

Frozen pizza August 2009 1,913,134 10.34 4.84 $.26 73.3% 10.9% Tomato products April 2009 2,807,504 43.79 15.24 $.08 77.1% 6.4% Soup July 2009 7,962,022 22.55 12.99 $.16 60.0% 68.2% Salad dressing June 2009 2,441,076 4.06 3.05 $.18 49.9% 12.4% Yogurt Q1 2009 6,140,670 44.20 24.02 $.14 51.9% 16.0% Spaghetti sauce April 2009 3,011,690 39.39 12.40 $.10 71.4% 6.3% Granola bars Q1 2009 520,183 22.43 7.96 $.46 39.4% 24.5% Ice cream June 2009 3,723,251 18.79 13.93 $.12 78.7% 29.5%

822 JOURNAL OF MARKETING RESEARCH, DECEMBER 2015

purchases across the eight categories. The data set includes key purchase variables: the week in which the purchase was made, the shopper who made the purchase, the total number of units bought, the total dollar amount the shopper paid, the price of the product, whether the product was on promotion at the time of purchase, the size of the product (in ounces), and the nutrition score of the product. We note that in each of the eight product categories, there were products that did not have nutrition scores because they had not been scored and were not in the NuVal database. Because the focus of our project is to examine the effect of POS nutrition scores on consumer choice behavior, we used only the data for purchases of products with NuVal scores (approxi- mately 81% of total sales). Table 2 summarizes the sample composition.

Descriptive Analysis

We first conducted preliminary analysis to investigate our main research question—that is, whether the POS nutrition scoring system helped shoppers make healthier food choices. To examine the change in the nutrition content of shoppers’ purchases, we constructed a volume- weighted average of the nutrition score of each shopper’s purchases in the six-month pre- and postrollout periods in each of the eight categories. Thus, in each category for each shopper, we had two observations: the shopper’s volume- weighted average nutrition score of his or her purchases in the category in the pre-NuVal period and the volume- weighted average nutrition score of the shopper’s pur- chases in the category in the post-NuVal period.2

We then calculated the average percentage change in each shopper’s volume-weighted nutrition score in the postrollout period relative to the prerollout period in each category (the last column in Table 2 displays the average percentage change across all shoppers in each category). Shoppers improved the nutrition content of their purchases by 21.8% on average across the eight different categories. Canned soup (68.2%), ice cream (29.5%), and granola bars (24.5%) were the three categories in which shoppers exhibited the greatest increase in their purchases’ nutrition content. In contrast, in the spaghetti sauce (6.3%), tomatoes (6.4%), and frozen pizza (10.9%) categories, shoppers exhibited the smallest, but still significant, improvement in the healthiness of their purchases. Thus, these descriptive results provide initial support for H1 and show that the POS nutrition scoring system was indeed successful in pro- moting healthier food choices. Next, we present our more robust and comprehensive modeling approach with which we test all hypotheses.

Modeling Approach

We estimate two regression models to test our pre- dictions. In both models, we consider each weekly purchase of a Universal Product Code (UPC) by each individual shopper as a separate observation. Table 3 summarizes all measures used in the models to operationalize the con- structs. Table 4 presents the descriptive statistics of all variables. We did not include category indicator variables in the analysis because category differences are largely

captured by category healthiness and category NuVal score variability.3

Model 1 tests H1–H3 regarding the impact of the POS nutrition scoring system on the healthiness of shoppers’ weekly purchases and the moderating influence of category healthiness and within-category variability in nutrition scores. Therefore, the dependent variable in the first model is the nutrition score of the UPC i in category j purchased by shopper k in week t. The specification of the model is as follows:

(1)

NUTRISCOREijkt = b0 + b1 × � CATHEALTHYj

+ b2 × � CATVARj

� + b3 ×

� RELEASEQj

+ b4 × � SEASONINDEXjt

+ b5 × � PRICEOZijkt

� + b6 ×

� PROMOijkt

+ b7 × ðPOSNUTRIPROMOtÞ + b8 ×

� PERIODijkt

� + b9 ×

� PERIODijkt

× � CATHEALTHYj

� + b10 ×

� PERIODijkt

× � CATVARj

� + eijkt,

where

NUTRISCOREijkt = the NuVal score of UPC i in category j purchased by shopper k in week t,

PERIODijkt = an indicator variable that equals −1 if the purchase of UPC i in category j by shopper k in week t occurred in the pre- NuVal period and 1 if the purchase oc- curred in the post-NuVal period,

PRICEOZijkt = the price per ounce of UPC i in category j purchased by shopper k in week t (mean- centered),4

PROMOijkt = an indicator variable that equals −1 if UPC i in category j purchased by shopper k in week t was not on pro- motion and 1 if the UPC was on promotion,

CATHEALTHYj = the average NuVal score of all UPCs in category j (mean-centered),

CATVARj = the standard deviation of the NuVal scores of all UPCs in category j (mean- centered),

RELEASEQj = a trend variable whose value corresponds to the 2009 quarter in which the NuVal system was implemented in category j (Q1 2009 = 1, Q2 2009 = 2, Q3 2009 = 3, Q4 2009 = 4),

SEASONINDEXjt = the seasonality index of category j in week t (mean-centered), and

POSNUTRIPROMOt = an indicator variable that equals 1 if the NuVal scoring system was promoted in the grocery chain’s weekly circular during the week of purchase and −1 otherwise.

2We included in our analysis only shoppers who purchased in the category in both the pre- and postperiods.

3A multivariate analysis of variance in which the seven category dummy variables predicted the category healthiness and variability variables reveals that the seven dummy variables are strong predictors of category healthiness and variability (Wilks’ lambda = 0, p < .0001).

4We used the price per ounce (i.e., the price per equivalized unit) to be consistent with our purchase volume measure used as the dependent variable in the second model described subsequently.

Healthy Choice 823

The effect of PERIOD in Model 1 enables us to test H1, which predicts the change in the healthiness of shoppers’ purchases resulting from the implementation of the NuVal system. The interaction terms—PERIOD × CATHEALTHY and PERIOD × CATVAR—test H2 and H3, respectively. Furthermore, we included a trend control variable for the time (i.e., quarter) when the NuVal scores were introduced in each category (RELEASEQ) to account for the fact that the nutrition scores were implemented at different points in time, which could have affected shop- pers’ reactions to them. Furthermore, we controlled for the timing of the grocery chain’s promotion and awareness campaigns of the NuVal scoring system by including an in- dicator variable in the model that indicates whether the NuVal scores were promoted in the chain’s weekly circular during the week of purchase (POSNUTRIPROMO). We also used the weekly seasonality index for each category (SEASONINDEX) as a covariate to control for seasonal differences in the demand for products. Finally, we also controlled for the price per ounce of the purchased UPC (PRICEOZ), as well as whether the product was on promotion (PROMO).

We specified a second model to test the remaining six hypotheses regarding the changes in shoppers’ price and promotion sensitivities resulting from the implementation of the NuVal nutrition scoring system (H4–H9). The de- pendent variable in Model 2 is the total number of equivalized units5 of a UPC purchased by each individual

shopper per week. The specification of the model is as follows: (2)

EQVUNITijkt = b0 + b1 × � CATHEALTHYj

+ b2 × � CATVARj

� + b3 ×

� RELEASEQj

+ b4 × � SEASONINDEXjt

+ b5 × ðPOSNUTRIPROMOtÞ + b6 ×

� PERIODijkt

� + b7 ×

� PRICEOZijkt

+ b8 × � PROMOijkt

+ b9 × � PERIODijkt

� × � CATHEALTHYj

+ b10 × � PERIODijkt

� × � CATVARj

+ b11 × � PERIODijkt

� × � PRICEOZijkt

+ b12 × � PRICEOZijkt

� × � CATHEALTHYj

+ b13 × � PRICEOZijkt

� × � CATVARj

+ b14 × � PERIODijkt

� × � PRICEOZijkt

× � CATHEALTHYj

� + b15 ×

� PERIODijkt

× � PRICEOZijkt

� × � CATVARj

+ b16 × � PERIODijkt

� × � PROMOijkt

+ b17 × � PROMOijkt

� × � CATHEALTHYj

+ b18 × � PROMOijkt

� × � CATVARj

+ b19 × � PERIODijkt

� × � PROMOijkt

× � CATHEALTHYj

� + b20 ×

� PERIODijkt

× � PROMOijkt

� × � CATVARj

� + eijkt,where

EQVUNITijkt = the total number of equivalized units of UPC i in category j purchased by shopper k in week t,

PRICEOZijkt = the price per ounce of UPC i in category j pur- chased by shopper k in week t (mean-centered),

TABLE 3 DETAILED DESCRIPTIONS OF MEASURES FOR THE VARIABLES OF INTEREST

Variable Description

Dependent Variables

Nutrition content of shoppers’ purchases Indicates the NuVal score of each shopper’s purchase of a UPC per week. Larger numbers indicate more nutritious purchases (NUTRISCORE; DV in Model 1).

Equivalized units purchased by each shopper per week

Indicates the total number of equivalized units (units × ounces per unit) of a UPC purchased by each shopper per week (EQVUNIT; DV in Model 2).

Independent Variables

Period Indicates the period in which the purchase is made; coded as −1 = pre-NuVal introduction period and 1 = post-NuVal introduction period (PERIOD).

Price per ounce Indicates the price per ounce of each UPC purchased per week (PRICEOZ).

Promotion Indicates whether the purchased product was on promotion; coded as – 1 = product not on promotion and 1 = product on promotion (PROMO).

Product category healthiness Indicates the healthiness of each product category; measured by the averageNuVal score of all UPCs in each category (CATHEALTHY).

Variability of nutrition scores in the category Indicates the variability of the nutrition scores in each product category; measured by the standard deviation of the NuVal scores of all UPCs in each category (CATVAR).

Seasonality index Indicates the weekly seasonality index of the product category; obtained from Nielsen (SEASONINDEX).

Release quarter A trend variable that corresponds to the 2009 quarter in which the nutrition scoring system was implemented in each category; ranges from 1 to 4 (RELEASEQ).

NuVal promotion and awareness campaigns

Indicates whether the NuVal scores were promoted in the grocery chain’s weekly circular during the week of purchase; coded as 1 if the NuVal scoring system was featured in the weekly circular and −1 otherwise (POSNUTRIPROMO).

5We determined the total number of equivalized units by multiplying the number of units purchased by the product size in ounces. Equivalized units are a better measure of volume sales because they account for differences in package size.

824 JOURNAL OF MARKETING RESEARCH, DECEMBER 2015

PROMOijkt = an indicator variable that equals 1 if UPC i in category j purchased by shopper k in week t was on promotion and −1 otherwise,

PERIODijkt = an indicator variable that equals −1 if the purchase of UPC i in category j by shopper k in week t occurred in the pre-NuVal period and 1 if the purchase occurred in the post-NuVal period,

CATHEALTHYj = the average NuVal score of all UPCs in category j (mean-centered),

CATVARj = the standard deviation of the NuVal scores across the UPCs in category j (mean- centered),

RELEASEQj = a trend variable whose value corresponds to the 2009 quarter in which the NuVal sys- tem was implemented in category j (Q1 2009 = 1, Q2 2009 = 2, Q3 2009 = 3, Q4 2009 = 4),

SEASONINDEXjt = the seasonality index of category j in week t (mean-centered), and

POSNUTRIPROMOt = an indicator variable that equals 1 if the NuVal scoring system was promoted in the grocery chain’s weekly circular during the week of purchase and −1 otherwise.

In Model 2, the two-way interactions of PERIOD × PRICEOZ and PERIOD × PROMO test the changes of price and pro- motion sensitivities in the post-NuVal period (i.e., H4 and H7, respectively). The four three-way interaction terms (PERIOD × PRICEOZ × CATHEALTHY, PERIOD × PRICEOZ × CATVAR, PERIOD × PROMO × CATHEALTHY, and PERIOD× PROMO×CATVAR) test the hypotheses regarding the moderating influence of the two category characteristics on the shifts in price and promotion sensitivities in the post- implementation period. As in Model 1, we controlled for the release quarter of the NuVal scores in each category, the timing of theNuVal promotion and awareness campaigns in the grocery chain’s circular, and theweekly seasonality in category demand.

Results

We estimated the two regression models simultaneously as a system of seemingly unrelated regressions (SURs) to account for the correlations of the error terms of the two equations (Srivastava and Giles 1987; Zellner 1962). Table 5 summarizes the results of the SUR model. The system-weighted R-square was .35, indicating a good overall model fit. The intercorrelation was −.04, suggesting that the purchased products’ nutrition scores (Model 1) and purchase volume (Model 2) are not very highly correlated.

Model 1 (nutrition score of purchased product). In Model 1, the main effects of category healthiness and variability in the nutrition scores are both positive and significant, in- dicating that the NuVal scores of shoppers’ purchases were higher in healthier product categories (b1 = .93, p < .0001) and in categories with greater within-category variability (b2 = .12, p < .0001). More importantly, our results indicate that the introduction of the NuVal nutrition scoring system increased the nutrition content of shoppers’ purchases (b8 = .39, p < .0001). That is, implementation of the NuVal nutrition scoring system prompted shoppers to switch to products with higher scores and thus make healthier food choices in the six-month postrollout period. Thus, H1 is supported.

In addition, as we predicted in H2, the strength of this effect depends on the product category healthiness, such that the improvement in the nutrition content of shoppers’ purchases was significantly greater in healthier categories (b9 = .03, p < .0001). In other words, shoppers seemed to incorporate the nutrition scores into their food decisions to a larger extent in healthier product categories. Fur- thermore, consistent with H3, the impact of the nutrition scoring system on shoppers’ food choices was also stronger in product categories characterized by greater variability in nutrition scores across the assortment (b10 = .03, p < .0001).6 The effects of the control variables were all sig- nificant (see Table 5).

Model 2 (purchase volume). The results of Model 2 show that the period effect is significant and positive (b6 = .30, p < .0001), indicating that the average number of equivalized units purchased per week increased after the implementation of the NuVal system. In addition, the effect of price on the total number of equivalized units bought per week is negative (b7 = −64.98, p < .0001), such that higher product prices were associated with a lower number of equivalized units pur- chased. Furthermore, the effect of promotion is positive (b8 = 3.02, p < .0001), meaning that shoppers bought more equivalized units during a promotion.

More importantly, as we predicted in H4, the two-way interaction7 of period × price per ounce is positive and

Table 4 DESCRIPTIVE STATISTICS

Variable Mean SD Min Maxa

NuVal score of purchased UPC 27.89 20.50 1.00 100.00 Total number of equivalized units purchased 27.16 26.53 .90 13,440.00 Price per ounce .15 .13 .004 2.77 Promotion .26 .97 −1.00 1.00 Category healthiness 28.19 13.59 4.06 44.20 Category variability 14.16 6.23 3.05 24.02 Seasonality index 1.06 .23 .48 1.77 Release quarter 2.11 .75 1.00 3.00 NuVal promotion and awareness campaign −.41 .91 −1.00 1.00

aThe results are substantively unchanged if purchases only under the size of 500 equivalized units are included in the analysis.

6However, in the omnibus analysis we report subsequently, this effect is not significantly different from the control group of panelists who purchased from grocery chains without the NuVal system.

7We included all lower-level interactions in the model and present the results in Table 5; however, in the interest of space, we focus our discussion on the parameters germane to our hypotheses.

Healthy Choice 825

significant (b11 = 6.87, p < .0001), indicating that shoppers’ price sensitivity decreased after the nutrition scoring system was implemented. Whereas a unit in- crease in price per ounce led to approximately a 72-ounce decrease in the weekly quantity purchased in the pre- NuVal period (−64.98 – 6.87), a unit increase in price per ounce in the post-NuVal period led to approxi- mately a 58-ounce decrease in quantity (−64.98 + 6.87), reflecting a 19.4% decrease in price sensitivity. In line with H5, the three-way interaction of period × price per ounce × category healthiness was also positive and significant (b14 = .66, p < .0001), indicating that the decrease in price sensitivity was greater in healthier product categories. However, in contrast to our pre- diction in H6, there was less decrease in price sensitivity in product categories characterized by more within- category variability in healthiness across the available products (b15 = −.59, p < .0001). We speculate that this might be due to the fact that in such categories, shoppers have a wider range of alternatives to switch to, giving them

more decision latitude to select healthy alternatives while still taking price into consideration.

The promotion × period interaction is also significant and positive (b16 = .81, p < .0001), indicating that shoppers’ sensitivity to promotion increased in the post- NuVal period relative to the pre-NuVal period, thus providing support for H7. In the prerollout period, the availability of a promotion led to an additional 2.21 ounces purchased (3.02 – .81); in contrast, after the implementation of the NuVal system, the availability of a promotion in- creased the quantity purchased by 3.83 ounces (3.02 + .81), reflecting a 73.3% increase in promotion sensitivity. This effect was weaker in healthier product categories (b19 = −.07, p < .0001), in support of H8. Finally, the results also provide support for H9; the three-way in- teraction of period × promotion × within-category var- iability in healthiness across the available products is positive and significant (b20 = .05, p < .0001), thus re- vealing that the increase in shoppers’ promotion sen- sitivity in the postimplementation period was greater in

Table 5 SUR MODEL RESULTS (EXPERIMENTAL STORES)

Regression Coefficient (B)

Hypothesis Supported (Experimental

Store Analysis) Hypothesis Supported (Omnibus Analysis)

Model 1: Nutrition Score

Intercept (b0) 30.68*** Category healthiness (b1) .93*** Category variability (b2) .12*** Release quarter (b3) −1.38*** Seasonality index (b4) .39*** Price per ounce (b5) −7.69*** Promotion (b6) .44*** NuVal promotion campaign (b7) .12*** Period (b8) .39*** H1 supported H1 supported Period × Category healthiness (b9) .03*** H2 supported H2 supported Period × Category variability (b10) .03*** H3 supported H3 not supported

Model 2: Equivalized Units

Intercept (b0) 21.35*** Category healthiness (b1) .09*** Category variability (b2) −.63*** Release quarter (b3) 2.43*** Seasonality index (b4) −7.04*** NuVal promotion campaign (b5) .32*** Period (b6) .30*** Price per ounce (b7) −64.98*** Promotion (b8) 3.02*** Period × Category healthiness (b9) .04*** Period × Category variability (b10) −.17*** Period × Price per ounce (b11) 6.87*** H4 supported H4 supported Price per ounce × Category healthiness (b12) −.62*** Price per ounce × Category variability (b13) 1.90*** Period × Price per ounce × Category healthiness (b14) .66*** H5 supported H5 supported Period × Price per ounce × Category variability (b15) −.59*** H6 not supported H6 not supported Period × Promotion (b16) .81*** H7 supported H7 supported Promotion × Category healthiness (b17) −.07*** Promotion × Category variability (b18) .07*** Period × Promotion × Category healthiness (b19) −.07*** H8 supported H8 supported Period × Promotion × Category variability (b20) .05*** H9 supported H9 supported System-weighted R-square .35

***p < .0001.

826 JOURNAL OF MARKETING RESEARCH, DECEMBER 2015

categories with more variability in nutrition scores across the assortment.8

Robustness Checks and Follow-Up Analyses

Control group analysis. Because all stores in the grocery chain implemented the NuVal program, we are unable to use the chain’s loyalty data to create a control condition in which shoppers were not exposed to the NuVal scores at the POS. Therefore, we sought purchase data for a group of shoppers in geographically contiguous regions during the same time as in our main analysis. Fortunately, we were able to obtain Nielsen wand panel data for the desired time frame and categories. We restricted our analysis to panelists who reside in a region that is geographically contiguous to the focal grocery chain’s stores and who purchased a given category at a grocery store in both the “pre-NuVal” and “post-NuVal” periods. This resulted in 81,977 purchase occasions for 8,720 households.

We merged the data from the control stores and the experimental stores (i.e., those in which the NuVal scoring system was implemented) and conducted an omnibus analysis to compare the effects of the nutrition scoring system implementation in the control and treatment groups (similarly to Ailawadi et al. 2010). To do so, we included a TREATMENT indicator variable (coded as −1 if the pur- chase occurred in the control stores and as 1 if the purchase occurred in the experimental stores) and all appropriate higher-order interactions. The complete model specifica- tions appear in Appendix A. Briefly, in Model 1 the two-way interaction of TREATMENT × PERIOD and the three-way interactions of TREATMENT × PERIOD × CATHEALTHY and TREATMENT × PERIOD × CATVAR indicate whether the differences in the effects predicted in H1, H2, and H3, respectively, are significant between the control and treatment groups. Similarly, in Model 2 the three-way interactions of TREATMENT × PERIOD × PRICEOZ and TREATMENT × PERIOD × PROMO examine differences between the control and experimental stores in the hypothesized effects in H4 and H7, while the four-way interactions of TREATMENT × PERIOD × PRICEOZ × CATHEALTHY, TREATMENT × PERIOD × PRICEOZ × CATVAR, TREATMENT × PERIOD × PROMO × CATHEALTHY, and TREATMENT × PERIOD × PROMO × CATVAR test differences in the effects predicted in H5, H6, H8, and H9, respectively.

Table 6 presents a summary of the results of this analysis. The results reveal that for eight of the nine hypotheses, there are significant differences in the hy- pothesized effects between the experimental and control stores (the rightmost column of Table 5 notes the eight hypotheses that are supported by the omnibus analysis). The only hypothesis that is not supported in the omnibus analysis is H3. Therefore, we suggest in the “General Discussion” section that further research is needed to determine whether the impact of POS nutrition scoring systems on the nutrition content of shoppers’ purchases is moderated by the heterogeneity of the nutrition scores in the category.

Finally, note that although we find that the price sen- sitivity attenuation is weaker in product categories with greater within-category variability in the nutrition scores (which is the opposite of what we predicted in H6), this effect is significantly different between the experimental and control stores. This suggests that the moderating in- fluence of category variability on the decrease in shoppers’ price sensitivity is not driven by spurious correlation with an unmeasured variable. In summary, this omnibus analysis further underscores the robustness of our findings in the experimental stores.

Trends in effects of the POS nutrition scoring system implementation.As a robustness check, we assessed whether our findings are robust to different lengths of the postrollout window. In addition, we were interested in examining whether the effects of the POSnutrition scoring systemon the healthiness of shoppers’ purchases, as well as their price and promotion sensitivity, tend to change over time.9 For this purpose, we divided the six-month postimplementation period into two three-month post-NuVal periods. We retained the six-month pre-NuVal period to obtain a more stable baseline estimate of shoppers’ regular shopping behaviors before the NuVal system was implemented.

The two three-month postimplementation periods were represented by two dummy variables, with the six-month preimplementation period used as the reference category (i.e., POSTPERIOD1 = 1 if the purchase was made in the first three months after the NuVal scores introduction and 0 otherwise, and POSTPERIOD2 = 1 if the purchase was made in the second three months after the POS nutrition scoring system introduction and 0 otherwise). We then re-estimated the SUR model, substituting the two POST- PERIOD dummy variables for the PERIOD dummy vari- able. We also interacted the two new dummy variables with all other variables of interest. Appendix B provides the exact specification of the models, and Table 7 presents the results of these sensitivity analyses. The system-weighted R-square was .35, thus showing a good overall model fit. A summary of the trends in the effects of the POS nutrition scoring system between the two three-month post-NuVal periods appears in Table 8.

The results suggest that our findings are robust; we find support for most of our hypotheses in both postrollout periods. Specifically, eight of our nine hypotheses are supported in Months 1–3 following the NuVal rollout and six of the nine hypotheses are supported inMonths 4–6 after the NuVal system was implemented in the stores. Fur- thermore, the effects tend to weaken over time. That is, the coefficient for the second three-month period after NuVal was introduced is smaller than the coefficient for the first three-month postrollout period in seven out of nine cases. This is probably due to shoppers gaining knowledge of the nutrition scores over time and becoming accustomed to their display in the stores. Finally, it is also worthwhile to note that of the four parameters that reverse signs between the first and second post-NuVal rollout periods, three in- volve the moderating effect of category variability in the nutrition scores across the assortment, which could again be due to shoppers becoming more fluent with the scores over

8We also analyzed the data using hierarchical linear modeling, and the results are substantively unchanged. These results are available from the authors on request. 9We thank the associate editor for this suggestion.

Healthy Choice 827

time and becoming more adept at incorporating them into their purchase decisions. Specifically, category score vari- ability tends to be initially associated with greater sensitivity to price (which requires greater cognitive resources) and more reliance on promotion (b20 = −3.19 and b28 = .48 for the Months 1–3 interaction of category variability with price and promotion, respectively), but these effects reverse in Months 4–6 after shoppers have had an opportunity to use the scores over multiple trips (b21 = 1.58 and b29 = −.17 for the Months 4–6 interaction of category variability with price and pro- motion, respectively).

Assortment change as an alternative explanation. Change in the product assortment in the examined categories is an alternative explanation of the observed improvement in the nutrition content of shoppers’ purchases in the post-NuVal period. Specifically, the introduction of healthier products in the postimplementation period could in itself lead to healthier purchases even without the introduction of the nutrition scoring system.10 To test this possibility, we conducted an additional ancillary analysis. We divided the prerollout period into two three-month pre-NuVal periods and identified any new UPCs introduced in the second three-month prerollout period. We define newly introduced products as UPCs purchased in the second three-month pre- NuVal period that did not have any sales in the first three- month pre-NuVal period. Given that we have purchase data for approximately 535,000 shoppers, we believe that at least one person would have bought the products if they had indeed been available. Similarly, we divided the postrollout period into two three-month post-NuVal periods and identified the newUPCs introduced in the second three-month postrollout period.We identified 45 newUPCs in the prerollout period and 41 in the postrollout period.

To control for category healthiness (i.e., to calculate the relative healthiness of the newUPCs), we then subtracted each new UPC’s NuVal score from the category average. The average deviation from the category average across the 45 new UPCs introduced in the prerollout period was −6.83, compared with −2.58 across the 41 UPCs introduced in the postrollout period. This difference is not statistically sig- nificant (t(84) = 1.37, n.s.). Furthermore, the average volume purchased for new UPCs in the prerollout period (55,300 equivalent units) is more than the average volume purchased in the postrollout period (37,200 equivalent units), though this difference is not statistically significant (t(84) = 1.08, n.s.). These results indicate that change in the assortment in the examined product categories cannot account for the observed improvement in the nutrition content of shoppers’ purchases.

GENERAL DISCUSSION

Grocery retailers have become increasingly interested in implementing health and wellness initiatives at the POS to help their shoppers improve their diets and overall health. In the present research, we assessed the effectiveness of one of the most prevalent initiatives implemented at a wide range of grocery stores across the country, namely, the use of a simplified nutrition scoring system at the POS. The nutrition scoring system combines all the nutritional in- formation into a single summary indicator of the relative

Table 6 SUR MODEL RESULTS (COMBINED ANALYSIS OF

EXPERIMENTAL AND CONTROL STORES)

Regression Coefficient (B)

Model 1: Nutrition Score

Intercept (b0) 28.95*** Release quarter (b1) −1.35*** Seasonality index (b2) .52*** Price per ounce (b3) −7.71*** Promotion (b4) .44*** Treatment (b5) 1.62*** Category healthiness (b6) .85*** Treatment × Category healthiness (b7) .08*** Category variability (b8) .26*** Treatment × Category variability (b9) −.14*** Period (b10) .18*** Treatment × Period (b11) .17*** Period × Category healthiness (b12) .02*** Treatment × Period × Category healthiness (b13) .01** Period × Category variability (b14) .04*** Treatment × Period × Category variability (b15) −.004

Model 2: Equivalized Units

Intercept (b0) 25.83*** Release quarter (b1) 2.50*** Seasonality index (b2) −6.70*** Treatment (b3) −4.75*** Category healthiness (b4) −.01 Treatment × Category healthiness (b5) .10*** Category variability (b6) −.26*** Treatment × Category variability (b7) −.37*** Period (b8) .39*** Treatment × Period (b9) −.19*** Price per ounce (b10) −62.17*** Treatment × Price per ounce (b11) −2.84*** Promotion (b12) 3.18*** Treatment × Promotion (b13) −.15** Period × Category healthiness (b14) −.02* Treatment × Period × Category healthiness (b15) .06*** Period × Category variability (b16) −.02 Treatment × Period × Category variability (b17) −.15*** Period × Price per ounce (b18) 2.57*** Treatment × Period × Price per ounce (b19) 4.31*** Price per ounce × Category healthiness (b20) −2.47*** Treatment × Price per ounce × Category healthiness (b21) 1.85*** Price per ounce × Category variability (b22) 5.72*** Treatment × Price per ounce × Category variability (b23) −3.79*** Period × Price per ounce × Category healthiness (b24) .42*** Treatment × Period × Price per ounce × Category

healthiness (b25) .25**

Period × Price per ounce × Category variability (b26) −.35* Treatment × Period × Price per ounce × Category variability (b27) −.32* Period × Promotion (b28) .38*** Treatment × Period × Promotion (b29) .42*** Promotion × Category healthiness (b30) −.08*** Treatment × Promotion × Category healthiness (b31) .007 Promotion × Category variability (b32) .19*** Treatment × Promotion × Category variability (b33) −.12*** Period × Promotion × Category healthiness (b34) −.03*** Treatment × Period × Promotion × Category healthiness (b35) −.03*** Period × Promotion × Category variability (b36) .01 Treatment × Period × Promotion × Category variability (b37) .03* System-weighted R-square .35

*p < .05. **p < .001. ***p < .0001. Notes: Treatment is an indicator variable that equals −1 if the purchase

occurred in the control stores and 1 if the purchase occurred in the experimental stores (i.e., the stores in which the NuVal scoring systemwas implemented).We note that in this combined analysis we do not control for the timing of the stores’ promotion and awareness campaigns of the NuVal scoring system because such campaigns were not used in the control stores. 10We thank an anonymous JMR reviewer for pointing this out.

828 JOURNAL OF MARKETING RESEARCH, DECEMBER 2015

healthiness of a product and communicates it to consumers in a simple and easy-to-process format, thereby reducing the complexity and difficulty in understanding the nutri- tional label. We used a large-scale quasi experiment to assess how the food decisions of 535,000 shoppers of a grocery chain changed after the implementation of a simplified POS nutrition scoring system in eight product categories, as well as how this affected the shoppers’ price and promotion sensitivities.

Our research makes the following key contributions. First, using a large-scale quasi experiment and the pur- chase history data of more than 535,000 shoppers, we

demonstrate that the introduction of a POS nutrition scoring system promoted healthier food choices among the grocery chain’s shoppers. Second, we show that this effect is stronger in healthier product categories. Third, our work reveals spillover effects of the implementation of a POS nutrition scoring system on the effectiveness of two key marketing-mix elements. Specifically, we show that shoppers became less price sensitive and more promotion sensitive in the postrollout period (i.e., price sensitivity decreased by 19.4% and promotion sensitivity increased by 73.3%), presumably because the task of examining the newly available nutrition scores required additional cognitive

Table 7 TREND ANALYSIS RESULTS

Regression Coefficient Hypothesis Supported

Model 1: Nutrition Score

Intercept (b0) 30.61*** Category healthiness (b1) .89*** Category variability (b2) .14*** Release quarter (b3) −1.44*** Seasonality index (b4) .66*** Price per ounce (b5) −7.66*** Promotion (b6) .46*** NuVal promotion campaign (b7) .12*** Postperiod 1 (b8) .84*** H1 supported Postperiod 2 (b9) .18*** H1 supported Postperiod 1 × Category healthiness (b10) .09*** H2 supported Postperiod 2 × Category healthiness (b11) .05*** H2 supported Postperiod 1 × Category variability (b12) .01*** H3 supported Postperiod 2 × Category variability (b13) −.07*** H3 not supported

Model 2: Equivalized Units

Intercept (b0) 20.91*** Category healthiness (b1) .09*** Category variability (b2) −.59*** Release quarter (b3) 2.41*** Seasonality index (b4) −7.43*** NuVal promotion campaign (b5) .31*** Postperiod 1 (b6) .58*** Postperiod 2 (b7) 1.16*** Price per ounce (b8) −70.19*** Promotion (b9) 2.09*** Postperiod 1 × Category healthiness (b10) .08*** Postperiod 2 × Category healthiness (b11) −.04*** Postperiod 1 × Category variability (b12) −.23*** Postperiod 2 × Category variability (b13) .00 Postperiod 1 × Price per ounce (b14) 11.14*** H4 supported Postperiod 2 × Price per ounce (b15) 10.18*** H4 supported Price per ounce × Category healthiness (b16) −.96*** Price per ounce × Category variability (b17) 2.15*** Postperiod 1 × Price per ounce × Category healthiness (b18) 2.73*** H5 supported Postperiod 2 × Price per ounce × Category healthiness (b19) −.83*** H5 not supported Postperiod 1 × Price per ounce × Category variability (b20) −3.19*** H6 not supported Postperiod 2 × Price per ounce × Category variability (b21) 1.58*** H6 supported Postperiod 1 × Promotion (b22) 2.00*** H7 supported Postperiod 2 × Promotion (b23) 1.58*** H7 supported Promotion × Category healthiness (b24) .001 Promotion × Category variability (b25) −.02*** Postperiod 1 × Promotion × Category healthiness (b26) −.26*** H8 supported Postperiod 2 × Promotion × Category healthiness (b27) −.03*** H8 supported Postperiod 1 × Promotion × Category variability (b28) .48*** H9 supported Postperiod 2 × Promotion × Category variability (b29) −.17*** H9 not supported System-weighted R-square .35

***p < .0001.

Healthy Choice 829

effort, which diverted cognitive resources from other tasks (i.e., the examination of the price distribution) and led consumers to rely more on choice tactics to simplify their decision-making process (i.e., purchasing what is on promotion).

Fourth, we demonstrate that the decrease in price sensitivity is moderated by the product category healthiness, such that it is greater in healthier product categories, in which shoppers are more likely to base their food choices on nutrition content. In contradiction to our prediction (H6), we find that the decrease in price sensitivity is weaker in categories with more heteroge- neity in the nutrition scores. This could be due to the fact that those categories offer a wide range of alternatives that shoppers could switch to, thus providing them more decision latitude to take both the nutrition content and price into consideration. Our findings also suggest that the increase in shoppers’ promotion sensitivity is weaker in healthier product categories and those with less within-category variability in the products’ nutrition scores.

Finally, we also conducted more fine-grained analysis of the postrollout period and thus assessed the trends in the POS nutrition scoring system effects over time. This analysis reveals two main findings. First, as a whole, our findings are robust, and most of the hypotheses are sup- ported in both periods. Eight of our nine hypotheses are supported in the first three-month postrollout period, and six of the nine hypotheses are supported in the second three-month postrollout period. Second, the effects tend to weaken over time. That is, the coefficient for the first three-month period after NuVal was introduced in the category is greater than the corresponding coefficient in the second three-month postrollout period in seven out of nine cases. However, the effect remained significant in most cases. In summary, although these more fine-grained analyses of the postrollout period clearly show the ro- bustness of our findings, they also underscore the im- portance of assessing the impact of POS nutrition scoring systems over a longer period of time and highlight this as a fruitful direction for further research. Such an in- vestigation would also prove helpful in examining further the moderating influence of category variability on the change in the nutrition content of shoppers’ purchases

(i.e., H3, for which we did not find conclusive support in this research).

Practical Implications

Our research has important practical implications for consumers, marketers, and public policy makers. Obesity is one of the most serious and prevalent problems con- sumers face today, and many consumers are trying to steer away from high-calorie, high-fat foods and purchase healthier alternatives (FMI 2012a). Our research suggests that one effective way in which consumers can ensure that they purchase nutritious and healthy products is to shop at grocery stores that have implemented a POS nutrition rating system. Our results demonstrate that simplified nutritional scoring systems can help consumers make healthier food choices.

Our research also offers important practical insights to grocery retailers and food manufacturers. First, given that grocery retailers are increasingly implementing a wide range of health and wellness initiatives at the POS, it is essential for them to understand the programs’ effec- tiveness in promoting healthier food choices. Our re- search highlights that a prevalent health and wellness initiative—a POS nutrition scoring system—is effective in improving shoppers’ food decisions. Furthermore, grocery chains should be aware that the improvements in the nutrition content of shoppers’ purchases seem to be stronger in healthier product categories. Therefore, if rolling out the nutrition scores across categories over time (like the grocer in our research), retailers should begin with healthier product categories. In addition, the finding that the effectiveness of the POS nutrition scoring system tended to wane over time suggests that grocery retailers that want to improve the health and wellness of their shoppers might find it beneficial to periodically imple- ment in-store promotion and awareness campaigns re- garding the nutrition scoring system. The novelty of such campaigns might bolster the effectiveness of the nutrition scoring system over time.

Second, we conducted an ancillary analysis (available from the authors) to examine the change in sales between the pre- and postrollout periods. The results demonstrate that the grocery chain experienced an overall increase in the total sales in the eight product categories after the

Table 8 SUMMARY OF THE TRENDS IN THE EFFECTS OF THE POS NUTRITION SCORING SYSTEM IMPLEMENTATION

Impact of

Impact on

Healthiness of Shoppers’ Purchases

Price Sensitivity

Promotion Sensitivity

D in Coefficient (Postperiod 2 vs. Postperiod 1)

Nutrition scoring system implementation (main effect) ↓ .66*** ↓ .96*** ↓ .42*** Category healthinessa (interaction of category

healthiness × nutrition scoring system implementation) ↓ .04*** ↓ 3.56*** ↑ .23***

Variability of nutrition scores in the categorya (interaction of category variability × nutrition scoring system implementation)

↓ .08*** ↑ 4.77*** ↓ .65***

***p < .0001. aThe hypothesis concerns the moderating effect of category variables of the effect of the nutrition scoring system on the outcomes of interest.

830 JOURNAL OF MARKETING RESEARCH, DECEMBER 2015

implementation of the NuVal scoring system. Although this increase could be attributed to factors besides the nutrition scoring system, these results provide some preliminary evidence that the POS nutrition scoring system has a “win-win” benefit for both consumers and grocery stores. That is, grocery retailers might use the food rating system to appeal to the increasingly nutrition-conscious consumer, thereby gaining a com- petitive advantage.

Third, grocery retailers should also note that the use of a nutrition scoring system alters shoppers’ sensitivity to price and promotion. Because our findings reveal that the implementation of a POS nutrition scoring system leads to shoppers becoming less price sensitive, the nutrition scores seem to serve as a justification for the higher prices of healthier options. Furthermore, our re- sults suggest that after the implementation of the POS nutrition scores, marketers should adapt their promotion strategies to focus more on the use of salient promotion signage (rather than price cuts) to increase sales. Finally, grocery retailers should be aware that the changes in shoppers’ price and promotion sensitivity resulting from the use of a POS nutrition scoring system vary as a function of category characteristics, such that the at- tenuation in price sensitivity is even stronger in healthier categories and categories with less heterogeneity in the nutrition scores, while the increase in promotion sensi- tivity is weaker in healthier categories and categories with less variability in the nutrition scores. This means that, for example, nonprice promotions should be utilized more extensively, especially in categories that are un- healthy and include products with more heterogeneous scores. In contrast, price cuts will be even less effective in such categories and should therefore be used more sparingly.

Our research has important implications for food manufacturers as well. The implementation of nutrition scoring systems incentivizes food manufacturers to for- mulate healthier and more nutritious products to stay competitive. The presence of salient and easily un- derstandable nutrition information at the POS exposes unhealthy products that provide little or no nutrition and thus encourages manufacturers to reformulate them or introduce new products that better meet the needs of the increasingly nutrition-conscious shoppers.

Finally, our research should be useful to public policy makers seeking effective ways to combat the obesity ep- idemic. Our findings demonstrate that facilitating the understanding of nutritional information through the provision of salient and easy-to-process nutrition scores at the POS helps consumers make healthier food decisions. Therefore, one successful initiative that public policy makers could undertake to tackle the obesity challenge is to create a simplified overall nutrition scoring system to be used uniformly on all food products. Alternatively, they could provide various incentives to grocery retailers to develop and implement their own food rating systems or license those developed by third parties. Unfortunately, the present changes to the Nutrition Facts panel being pro- posed by the Food and Drug Administration do not include an overall score that could guide shoppers to healthier choices.

Limitations and Further Research

Our research sets the stage for several fruitful future research opportunities. First, we acknowledge that while our research provides robust evidence of the hypothe- sized effects, our data preclude a causal test of the processes driving these effects. For example, although we argue that shoppers become more promotion sensi- tive in the postimplementation period because they use promotions as an effort-minimizing decision heuristic or to justify the purchases of unhealthy foods, there could be other factors driving this effect, and only experi- mental research can test this possibility. Similarly, be- cause healthier products tend to be higher priced, the decrease in price sensitivity may be due to either shoppers simply paying less attention to price or the higher NuVal score serving as a justification for a higher price. Further research is needed to test these alternative processes.

Moreover, our data set contained information about whether the product was on promotion at the time of purchase but did not contain information regarding the specific type of promotion offering that was in place (e.g., bonus pack, price discount, two-for-one deal). Therefore, we could not examine whether consumers’ sensitivity to the different types of promotion offerings changed dif- ferentially as a result of the implementation of a POS nutrition scoring system. Given that prior research has shown that certain promotion offerings provide better justification for purchasing unhealthy food (Mishra and Mishra 2011), it would be insightful to examine whether the use of a food rating system leads to stronger changes in shoppers’ sensitivity to some types of promotions than to others.

Another limitation of the present research is that it is a quasi experiment, and we could not control for all ex- ogenous factors that could influence the results. For example, although we controlled for the presence of NuVal awareness campaigns in the chain’s weekly cir- cular, it could be argued that other communication media (e.g., information on the website, signage in the store) and the relative intensity of these campaigns could have also played a role in the impact of the NuVal scoring system on shoppers’ purchase behaviors. Therefore, future studies should explore the optimal type, timing, and intensity of communication programs about the implemented POS nutrition scoring system that maxi- mizes its effectiveness.

Our research focused on the effectiveness of one specific nutrition scoring system in promoting healthier food choices. However, grocery retailers must choose between several competing food rating systems, such as Guiding Stars and NuVal, that differ in terms of how they are calculated, how they are presented to consumers (using symbols, icons, or numerical scores), and their respective licensing costs. For example, the type of nutrition scoring system used could affect the shifts in price and promotion sensitivities. Specifically, comparing products by number of stars is easier and requires fewer cognitive resources than comparing them on nutrition scores (i.e., numerical values). Thus, the reduction in price sensitivity could be greater for nutrition scoring systems that express the foods’ nutrition

Healthy Choice 831

content using numbers (e.g., NuVal) rather than symbols or other visuals (e.g., Guiding Stars, Traffic Lights). Therefore, further research should assess the differences that might exist among the various nutrition rating systems in terms of their relative effectiveness in facilitating healthier choices.

Future studies should also examine the moderating im- pact of shopper-specific variables such as nutrition and health consciousness, body mass index, and income. In addition, our analysis should be expanded to the house- hold’s entire shopping basket. We focused on the im- provement of nutrition content of individual purchases (i.e., separate UPCs), but it is also important to assess the basket-level nutrition content because this would provide a holistic view of shoppers’ nutritional intake and reveal cross-category changes in shoppers’ purchasing behaviors (i.e., buying higher-scoring items in some categories but lower-scoring items in others).

In conclusion, our research demonstrates that facilitating the understanding of nutritional information through a significant reduction of not only the collection costs but also the computation and comprehension costs indeed helps consumers make healthier food decisions. As such, using a real behavior data set, we provide strong evidence in support of the idea proposed by Bettman (1975) and Russo et al. (1975) that the mere availability of nutritional in- formation at the POS is insufficient; this information must also be easy to process to be incorporated into consumers’ decisions.

APPENDIX A: COMBINED ANALYSIS OF DATA FROM EXPERIMENTAL AND CONTROL STORES:

MODEL SPECIFICATION

Model 1

NUTRISCOREijkt = b0 + b1 × � RELEASEQj

+ b2 × � SEASONINDEXjt

+ b3 × � PRICEOZijkt

+ b4 × � PROMOijkt

+ b5 × � TREATMENTijkt

+ b6 × � CATHEALTHYj

+ b7 × � TREATMENTijkt

× � CATHEALTHYj

+ b8 × � CATVARj

+ b9 × � TREATMENTijkt

× � CATVARj

� + b10 ×

� PERIODijkt

+ b11 × � TREATMENTijkt

× � PERIODijkt

+ b12 × � PERIODijkt

× � CATHEALTHYj

+ b13 × � TREATMENTijkt

× � PERIODijkt

× � CATHEALTHYj

+ b14 × � PERIODijkt

× � CATVARj

� + b15

× � TREATMENTijkt

× � PERIODijkt

� × � CATVARj

� + eijkt:

Model 2

EQVUNITijkt = b0 + b1 × � RELEASEQj

+ b2 × � SEASONINDEXjt

+ b3 × � TREATMENTijkt

+ b4 × � CATHEALTHYj

+ b5 × � TREATMENTijkt

× � CATHEALTHYj

+ b6 × � CATVARj

� + b7 ×

� TREATMENTijkt

× � CATVARj

� + b8 ×

� PERIODijkt

+ b9 × � TREATMENTijkt

� × � PERIODijkt

+ b10 × � PRICEOZijkt

+ b11 × � TREATMENTijkt

× � PRICEOZijkt

� + b12 ×

� PROMOijkt

+ b13 × � TREATMENTijkt

� × � PROMOijkt

+ b14 × � PERIODijkt

� × � CATHEALTHYj

+ b15 × � TREATMENTijkt

� × � PERIODijkt

× � CATHEALTHYj

� + b16 ×

� PERIODijkt

× � CATVARj

� + b17 ×

� TREATMENTijkt

× � PERIODijkt

� × � CATVARj

+ b18 × � PERIODijkt

� × � PRICEOZijkt

+ b19 × � TREATMENTijkt

� × � PERIODijkt

× � PRICEOZijkt

� + b20 ×

� PRICEOZijkt

× � CATHEALTHYj

+ b21 × � TREATMENTijkt

� × � PRICEOZijkt

× � CATHEALTHYj

� + b22 ×

� PRICEOZijkt

× � CATVARj

� + b23 ×

� TREATMENTijkt

× � PRICEOZijkt

� × � CATVARj

+ b24 × � PERIODijkt

� × � PRICEOZijkt

× � CATHEALTHYj

� + b25 ×

� TREATMENTijkt

× � PERIODijkt

� × � PRICEOZijkt

× � CATHEALTHYj

� + b26 ×

� PERIODijkt

× � PRICEOZijkt

� × � CATVARj

+ b27 × � TREATMENTijkt

� × � PERIODijkt

× � PRICEOZijkt

� × � CATVARj

+ b28 × � PERIODijkt

� × � PROMOijkt

+ b29 × � TREATMENTijkt

� × � PERIODijkt

× � PROMOijkt

� + b30 ×

� PROMOijkt

× � CATHEALTHYj

� + b31 ×

� TREATMENTijkt

× � PROMOijkt

� × � CATHEALTHYj

+ b32 × � PROMOijkt

� × � CATVARj

+ b33 × � TREATMENTijkt

� × � PROMOijkt

× � CATVARj

� + b34 ×

� PERIODijkt

× � PROMOijkt

� × � CATHEALTHYj

+ b35 × � TREATMENTijkt

� × � PERIODijkt

× � PROMOijkt

� × � CATHEALTHYj

+ b36 × � PERIODijkt

� × � PROMOijkt

× � CATVARj

� + b37 ×

� TREATMENTijkt

× � PERIODijkt

� × � PROMOijkt

× � CATVARj

� + eijkt,

832 JOURNAL OF MARKETING RESEARCH, DECEMBER 2015

where

NUTRISCOREijkt = the NuVal score of UPC i in category j purchased by shopper k in week t,

EQVUNITijkt = the total number of equivalized units of UPC i in category j purchased by shopper k in week t,

PERIODijkt = an indicator variable that equals −1 if the purchase of UPC i in category j by shopper k in week t occurred in the pre- NuVal period and 1 if the purchase oc- curred in the post-NuVal period,

TREATMENTijkt = an indicator variable that equals −1 if the purchase of UPC i in category j by shopper k in week t occurred in the control stores and 1 if the purchase oc- curred in the treatment stores (i.e., the stores in which the NuVal system was implemented),

PRICEOZijkt = the price per ounce of UPC i in category j purchased by shopper k in week t (mean- centered),

PROMOijkt = an indicator variable that equals −1 if UPC i in category j purchased by shopper k in week t was not on pro- motion and 1 if the UPC was on promotion,

CATHEALTHYj = the average NuVal score of all UPCs in category j (mean-centered),

CATVARj = the standard deviation of the NuVal scores of all UPCs in category j (mean- centered),

RELEASEQj = a trend variable whose value corresponds to the 2009 quarter in which the NuVal system was implemented in category j (Q1 2009 = 1, Q2 2009 = 2, Q3 2009 = 3, Q4 2009 = 4), and

SEASONINDEXjt = the seasonality index of category j in week t (mean-centered).

APPENDIX B: TRENDS IN EFFECTSOF POSNUTRITION SCORING SYSTEM IMPLEMENTATION:

MODEL SPECIFICATION

Model 1

NUTRISCOREijkt = b0 + b1 × � CATHEALTHYj

+ b2 × � CATVARj

� + b3 ×

� RELEASEQj

+ b4 × � SEASONINDEXjt

+ b5 × � PRICEOZijkt

� + b6 ×

� PROMOijkt

+ b7 × ðPOSNUTRIPROMOtÞ + b8 ×

� POSTPERIOD1ijkt

+ b9 × � POSTPERIOD2ijkt

+ b10 × � POSTPERIOD1ijkt

× � CATHEALTHYj

+ b11 × � POSTPERIOD2ijkt

× � CATHEALTHYj

+ b12 × � POSTPERIOD1ijkt

× � CATVARj

� + b13 ×

� POSTPERIOD2ijkt

× � CATVARj

� + eijkt:

Model 2

EQVUNITijkt = b0 + b1 × � CATHEALTHYj

+ b2 × � CATVARj

� + b3 ×

� RELEASEQj

+ b4 × � SEASONINDEXjt

+ b5 × ðPOSNUTRIPROMOtÞ + b6 ×

� POSTPERIOD1ijkt

+ b7 × � POSTPERIOD2ijkt

+ b8 × � PRICEOZijkt

� + b9 ×

� PROMOijkt

+ b10 × � POSTPERIOD1ijkt

× � CATHEALTHYj

+ b11 × � POSTPERIOD2ijkt

× � CATHEALTHYj

+ b12 × � POSTPERIOD1ijkt

× � CATVARj

� + b13 ×

� POSTPERIOD2ijkt

× � CATVARj

� + b14 ×

� POSTPERIOD1ijkt

× � PRICEOZijkt

� + b15 ×

� POSTPERIOD2ijkt

× � PRICEOZijkt

� + b16 ×

� PRICEOZijkt

× � CATHEALTHYj

� + b17 ×

� PRICEOZijkt

× � CATVARj

� + b18 ×

� POSTPERIOD1ijkt

× � PRICEOZijkt

� × � CATHEALTHYj

+ b19 × � POSTPERIOD2ijkt

× � PRICEOZijkt

� × � CATHEALTHYj

+ b20 × � POSTPERIOD1ijkt

� × � PRICEOZijkt

× � CATVARj

� + b21 ×

� POSTPERIOD2ijkt

× � PRICEOZijkt

� × � CATVARj

+ b22 × � POSTPERIOD1ijkt

� × � PROMOijkt

+ b23 × � POSTPERIOD2ijkt

× � PROMOijkt

� + b24 ×

� PROMOijkt

× � CATHEALTHYj

� + b25 ×

� PROMOijkt

× � CATVARj

� + b26 ×

� POSTPERIOD1ijkt

× � PROMOijkt

� × � CATHEALTHYj

+ b27 × � POSTPERIOD2ijkt

� × � PROMOijkt

× � CATHEALTHYj

+ b28 × � POSTPERIOD1ijkt

× � PROMOijkt

� × � CATVARj

+ b29 × � POSTPERIOD2ijkt

× � PROMOijkt

� × � CATVARj

� + eijkt,

where

NUTRISCOREijkt = the NuVal score of UPC i in category j purchased by shopper k in week t,

POSTPERIOD1ijkt = a dummy variable that equals 1 if the purchase of UPC i in category j by shopper k in week t occurred in the first three-month postrollout period and 0 otherwise,

POSTPERIOD2ijkt = a dummy variable that equals 1 if the purchase of UPC i in category j by shopper k in week t occurred in the second three-month postrollout period and 0 otherwise,

CATHEALTHYj = the average NuVal score of all UPCs in category j (mean-centered),

CATVARj = the standard deviation of the NuVal scores of all UPCs in category j (mean- centered),

RELEASEQj = a trend variable whose value corre- sponds to the 2009 quarter in which the NuVal system was implemented in category j (Q1 2009 = 1, Q2 2009 = 2, Q3 2009 = 3, Q4 2009 = 4),

Healthy Choice 833

SEASONINDEXjt = the seasonality index of category j in week t (mean-centered),

EQVUNITijkt = the total number of equivalized units of UPC i in category j purchased by shopper k in week t,

PRICEOZijkt = the price per ounce of UPC i in category j purchased by shopper k in week t (mean- centered),

PROMOijkt = an indicator variable that equals −1 if UPC i in category j purchased by shopper k in week t was not on promotion and 1 if it was on promotion, and

POSNUTRIPROMOt = an indicator variable that equals 1 if a NuVal scoring system was promoted in the grocery chain’s weekly circular during the week and −1 otherwise.

REFERENCES

Ailawadi, Kusum L., JieZhang, Aradhna Krishna, and Michael W.Kruger (2010), “When Wal-Mart Enters: How Incumbent Retailers React and How This Affects Their Sales Outcomes,” Journal of Marketing Research, 47 (August), 577–93.

Andrews, J. Craig, Richard G. Netemeyer, and Scot Burton (1998), “Consumer Generalization of Nutrient Content Claims in Ad- vertising,” Journal of Marketing, 62 (October), 62–75.

Balasubramanian, Siva K. and Catherine Cole (2002), “Consumers’ Search and Use of Nutrition Information: The Challenge and Promise of the Nutrition Labeling and Education Act,” Journal of Marketing, 66 (July), 112–27.

Berlyne, Daniel E. (1970), “Novelty, Complexity, and Hedonic Value,” Perception & Psychophysics, 8 (5), 279–86.

Berning, Joshua P., Hayley H. Chouinard, and Jill J. McCluskey (2008), “Consumer Preferences for Detailed Versus Summary Formats of Nutrition Information on Grocery Store Shelf Labels,” Journal of Agricultural & Food Industrial Organization, 6 (1), 1–20.

——— and David E. Sprott (2011), “Examining the Effectiveness of Nutrition Information in a Simulated Shopping Environment,” Journal of Food Distribution Research, 42 (3), 60–76.

Bettman, James R. (1975), “Issues in Designing Consumer In- formation Environments,” Journal of Consumer Research, 2 (3), 169–77.

———, John W. Payne, and Richard Staelin (1986), “Cognitive Considerations in Designing Effective Labels for Presenting Risk Information,” Journal of Public Policy & Marketing, 5, 1–28.

Brucks, Merrie, Andrew A. Mitchell, and Richard Staelin (1984), “The Effect of Nutritional Information Disclosure in Advertising: An Information Processing Approach,” Journal of Public Policy & Marketing, 3, 1–25.

Burson, Katherine A., Richard P. Larrick, and John G. Lynch Jr. (2009), “Six of One, Half Dozen of the Other: Expanding and Contracting Numerical Dimensions Produces Preference Re- versals,” Psychological Science, 20 (9), 1074–78.

Catalina Marketing (2010), “Helping Shoppers Overcome the Barriers to Choosing Healthful Foods,” (accessed October 5, 2015), [available at http://info.catalinamarketing.com/files/133/ Healthful_Foods_Study.pdf].

Dhar, Ravi and Klaus Wertenbroch (2012), “Self-Signaling and the Costs and Benefits of Temptation in Consumer Choice,” Journal of Marketing Research, 49 (February), 15–25.

Eagly, Alice H. and Shelly Chaiken (1993), The Psychology of Attitudes, Fort Worth, TX: Harcourt Brace Jovanovich.

Ehrich, Kristine R. and Julie R. Irwin (2005), “Willful Ignorance in the Request for Product Attribute Information,” Journal of Marketing Research, 42 (February), 266–77.

Feunekes, Gerda I.J., Ilse A. Gortemaker, Astrid A.Willems, René Lion, and Marcelle van den Kommer (2008), “Front-of- Pack Nutrition Labelling: Testing Effectiveness of Different Nutrition Labeling Formats Front-of-Pack in Four European Countries,” Appetite, 50 (1), 57–70.

Flegal, Katherine M., Margaret D. Carroll, Brian K. Kit, and Cynthia Ogden (2012), “Prevalence of Obesity and Trends in the Distribution of Body Mass Index Among US Adults, 1999–2010,” Journal of the American Medical Association, 307 (5), 491–97.

FMI (2012a), “Prevention Magazine and Food Marketing Institute Reveal 20th Annual ‘Shopping For Health’ Survey Results,” press release, (July 17), (accessed October 5, 2015), [available at http://www.fmi.org/news-room/latest-news/view/2012/07/17/ prevention-magazine-and-food-marketing-institute-reveal-20th- annual-shopping-for-health-survey-results].

——— (2012b), “Survey Report: Retailer Contributions to Health and Wellness,” (accessedOctober 5, 2015), [available at http://www.fmi. org/docs/health-and-wellness/survey-results-report.pdf?sfvrsn=2].

——— (2012c), “U.S. Grocery Shopper Trends 2012: Executive Summary,” (accessed October 5, 2015), [available at http://www. icn-net.com/docs/12086_FMIN_Trends2012_v5.pdf].

Garretson, Judith and Scot Burton (2000), “The Nutrition Facts Panel and Package Claims: The Role of Nutrient Values, Multiple Claims, and Congruent Information on Attitudes, Disease- Related Risks, and Consumer Trust,” Journal of Public Policy & Marketing, 19 (Fall), 213–27.

Hoch, Stephen J., Byung-Do Kim, Alan L. Montgomery, and Peter E. Rossi (1995), “Determinants of Store-Level Price Elasticity,” Journal of Marketing Research, 32 (February), 17–29.

Hoyer, Wayne D. (1984), “An Examination of Consumer Decision Making for a Common Repeat Purchase Product,” Journal of Consumer Research, 11 (3), 822–29.

Inman, J. Jeffrey, Leigh McAlister, and Wayne D. Hoyer (1990), “Promotion Signal: Proxy for a Price Cut?” Journal of Consumer Research, 17 (1), 74–81.

Johnson, Eric, Suzanne B. Shu, Benedict G.C. Dellaert, Craig Fox, Daniel G. Goldstein, Gerald Haubl, et al. (2012), “Beyond Nudges: Tools of a Choice Architecture,” Marketing Letters, 23 (2), 487–504.

Keller, Scott B., Mike Landry, Jeanne Olson, Anne M.Velliquette, Scot Burton, and J. Craig Andrews (1997), “The Effects of Nutrition Package Claims, Nutrition Facts Panels, and Motivation to Process Nutrition Information on Consumer Product Evalua- tions,” Journal of Public Policy & Marketing, 16 (Fall), 256–69.

Kozup, John C., Elizabeth H. Creyer, and Scot Burton (2003), “MakingHealthful FoodChoices: The Influence of Health Claims and Nutrition Information on Consumers’ Evaluations of Pack- aged Food Products and Restaurant Menu Items,” Journal of Marketing, 67 (April), 19–34.

Krishna, Aradhna and Priya Raghubir (1997), “The Effect of Line Configuration on Perceived Numerosity of Dotted Lines,” Memory & Cognition, 25 (4), 492–507.

Lancaster, Kelvin (1990), “The Economics of Product Variety: A Survey,” Marketing Science, 9 (3), 189–206.

Levy, Alan S. and Sara B. Fein (1998), “Consumers’ Ability to Perform Tasks Using Nutrition Labels,” Journal of Nutrition Education, 30 (4), 210–17.

———, Sara B. Fein, and Raymond E. Schucker (1996), “Per- formance Characteristics of Seven Nutrition Label Formats,” Journal of Public Policy & Marketing, 15 (Spring), 1–15.

834 JOURNAL OF MARKETING RESEARCH, DECEMBER 2015

Mathios, Alan D. (2000), “The Impact of Mandatory Disclosure Laws on Product Choices: An Analysis of the Salad Dressing Market,” Journal of Law & Economics, 43 (2), 651–77.

Mishra, Arul and Himanshu Mishra (2011), “The Influence of Price Discount Versus Bonus Pack on the Preference for Virtue and Vice Foods,” Journal of Marketing Research, 48 (February), 196–206.

Moorman, Christine (1996), “A Quasi Experiment to Assess the Consumer and Informational Determinants of Nutrition In- formation Processing Activities: The Case of the Nutrition La- beling and Education Act,” Journal of Public Policy & Marketing, 15 (Spring), 28–44.

Nielsen (2012), “ANielsen Report: Battle of the Bulge andNutrition Labels: Healthy Eating Trends Around the World,” research report, (accessed October 5, 2015), [available at http://silver- group.asia/wp-content/uploads/2012/02/Nielsen-Global-Food- Labeling-Report-Jan2012.pdf].

Organisation for Economic Co-operation and Development (2012), “OECD Obesity Update 2012,” research report, (accessed October 5, 2015), [available at http://www.oecd.org/health/ 49716427.pdf].

Payne, John W. (1976), “Task Complexity and Contingent Pro- cessing in Decision Making: An Information Search and Protocol Analysis,” Organizational Behavior and Human Performance, 16 (2), 366–87.

———, James R. Bettman, and Eric J. Johnson (1992), “Behavioral Decision Research: A Constructive Processing Perspective,” Annual Review of Psychology, 43 (February), 87–131.

Pieters, Rik, Michel Wedel, and Jie Zhang (2007), “Optimal Feature Advertising Design Under Competitive Clutter,” Marketing Science, 53 (11), 1815–28.

Pollack, Andrew (2013), “A.M.A. Recognizes Obesity as a Dis- ease,” The New York Times, (June 18), (accessed October 5, 2015), [available at http://www.nytimes.com/2013/06/19/busi- ness/ama-recognizes-obesity-as-a-disease.html].

Raghubir, Priya and Aradhna Krishna (1996), “As the Crow Flies: Bias in Consumers’Map-Based Distance Judgments,” Journal of Consumer Research, 23 (1), 26–39.

——— and ——— (1999), “Vital Dimensions in Volume Per- ceptions: Can the Eye Fool the Stomach?” Journal of Marketing Research, 36 (August), 313–26.

Roe, Brian, Alan S. Levy, and Brenda M. Derby (1999), “The Impact of Health Claims on Consumer Search and Product Evaluation Outcomes: Results from FDA Experimental Data,” Journal of Public Policy & Marketing, 18 (Spring), 89–105.

Russo, J. Edward, Gene Krieser, and Sally Miyashita (1975), “An Effective Display of Unit Price Information,” Journal of Mar- keting, 39 (May), 11–19.

———, Richard Staelin, Catherine Nolan, Garry Russell, and Barbara Metcalf (1986), “Nutrition Information in the Super- market,” Journal of Consumer Research, 13 (1), 48–70.

Sankin, Aaron (2013), “New California Soda-Tax Bill Under Consideration,”Huffington Post, (April 26), (accessed October 5, 2015), [available at http://www.huffingtonpost.com/2013/04/26/ california-soda-tax_n_3165417.html]

Shugan, Steven M. (1980), “The Cost of Thinking,” Journal of Consumer Research, 7 (2), 99–111.

Srivastava, Virendera K. and David E.A. Giles (1987), Seemingly Unrelated Regression Equations Models: Estimation and In- ferences. New York: Marcel Dekker, Inc.

Sutherland, Lisa A., Lori A. Kaley, and Leslie Fischer (2010), “Guiding Stars: The Effect of a Nutrition Navigation Program on Consumer Purchases at the Supermarket,” American Journal of Clinical Nutrition, 91 (4), 1090–94.

Thorpe, Kenneth (2009), “The Future Costs of Obesity: National and State Estimates of the Impact of Obesity on Direct Health Care Expenses,” research report, United Health Foundation and The American Public Health Association and Partnership for Prevention, (November), (accessed October 5, 2015), [available at http://www. nccor.org/downloads/CostofObesityReport-FINAL.pdf].

Variyam, Jayachandran N. and John Cawley (2006), “Nutrition Labels and Obesity,” NBER Working Paper 11956, National Bureau of Economic Research.

Wansink, Brian and Pierre Chandon (2006), “Can ‘Low-Fat’ Nu- trition Labels Lead to Obesity?” Journal of Marketing Research, 43 (November), 605–17.

Zellner, Arnold (1962), “An Efficient Method of Estimating Seem- ingly Unrelated Regressions and Tests for Aggregation Bias,” Journal of the American Statistical Association, 57 (298), 348–68.

Zhang, Jie (2006), “An Integrated Choice Model Incorporating Al- ternativeMechanisms for Consumers’Reactions to In-Store Display and Feature Advertising,” Marketing Science, 25 (3), 278–90.

Healthy Choice 835

Copyright of Journal of Marketing Research (JMR) is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder’s express written permission. However, users may print, download, or email articles for individual use.


Comments are closed.