Mandatory nutrition attributes labeling and consumer demand: a structural approach analysis of the US soft drink market

Abstract

While soft drinks stand out as a major cause of obesity and overweight worldwide, the USA is the country most concerned with this problem. To reverse the situation, experts have agreed that labels and taxes represent promising policy tools. Focusing on the soda market, this paper investigates how consumer demand for nutrient contents responds to the US revised nutrition facts label policy implemented in 2020. Relying on retail scanner data, the study finds after estimation of a discrete choice logit demand model that the change in nutrition labels caused consumers to modify their purchasing behavior. However, they did not systematically switch from unhealthy beverages to healthy ones. The analysis finds evidence that the label revision policy was mostly impactful in decreasing consumers’ preference for soda with superhigh calories and high sugar content. Surprisingly, the joint effect of label and tax did not decrease the sampled consumers’ preference for unhealthy nutrient contents. They still preferred buying soft drinks superhigh in calorie content inside US cities with a tax on sugar-sweetened beverages (SSB), despite the revision of the facts label. Finally, the new nutrition label has improved the average surplus of consumers and total welfare. However, the gain is lower within the cities implementing the sugar tax.

Introduction

In the USA, a prominent update in the regulation governing food product labeling came into effect on January 1, 2020. Issued by the U.S. Food and Drug Administration (FDA), the new regulation mandates packaged foods and beverages to carry a revised version of the nutrition facts label (NFL). The updated label should primarily facilitate consumers' process of making informed choices by reflecting relevant information pertaining to the connection between diet and obesity or heart chronic illnesses (FDA 2022). However, obesity and overweight are crucial health concerns faced by all of humanity at present. According to the World Health Organization (WHO), the number of obese people worldwide exceeds 1 billion, including 650 million adults, 340 million teenagers and 39 million children (WHO 2022). It is predicted by this institution that by 2025, an additional 167 million adults and youths will fall under the category of less healthy individuals due to overweight or obesity.

Furthermore, the USA is one of the countries most concerned with the problem of a culminating level of obesity among adults in 2016, where 28.6% of adults were obese, corresponding to an increase of 42% compared to the prevalence of 20.2% in 2000 (WHO 2023). Soft drinks stand out as one serious root of the problem under consideration. Accordingly, several authors (Basu et al. 2013; Tahmassebi and Banihani 2020) indicate that soda consumption significantly exposes any person concerned to overweight and obesity, in addition to increasing the risk of type 2 diabetes, regardless of the country. For Grabow et al. (2020), in contributing to the increase in chronic diseases, unhealthy food choices produce negative reverse effects on consumers. The nutritional quality of carbonated drinks adds an additional layer of concern to the problem. Tahmassebi and Banihani (2020) highlight that soft drinks are of little nutritional benefit, deficient in micronutrients, vitamins, and minerals, and can thus provide energy only. Compared to other beverages, such as bottled water, milk, or coffee, carbonated soft drinks constitute the most popular beverage in the USA (FDA 2018). The above arguments exemplify how relevant the concern is in the USA.

Addressing the issue of obesity and overweight necessitates the design and application of multiple strategies led by the WHO globally (WHO 2023), similar to other specific institutions acting at the country level. Their efforts involve an assortment of mechanisms from fiscal regulations to marketing policies (Barahona et al. 2023; Araya et al. 2022). In a concrete way, the policy tools used to put the mechanisms into action entail enforcing thresholds on sugar or fat content in food and beverage products, imposing taxes on sugar-sweetened beverages, and requiring manufacturers to reveal nutritional information through labels. While the first policy tool can be translated into warning or front-of-package labels, the NFL implements the latter to the benefit of consumers.

On the other side, although the influence of nutrition labels on consumers’ choices and preferences has been precedingly studied in the economics literature, a gap persists at several levels. First, while previous studies have stressed the links between labels and healthy food choices (Kumar and Kapoor 2017; Van der Merwe et al. 2022; Cabrera et al. 2023), the US soft drink market has not been especially considered. Second, preceding papers have relied on hypothetical choice experiments and examined front-of-package labels (Findling et al. 2018); traffic light nutrition labels (Cabrera et al. 2023); and the mix of nutrition facts panel, nutrient claim, and health claim (Barreiro‐Hurle et al. 2010); but they have not been able to focus on nutrition facts panel and account for interactions of major nutrient contents. Third, a previous study has opted for a randomized controlled trial (Mhurchu et al. 2018) to analyze the implications of label viewing behaviors for food purchases, especially traffic light labels and health star rating labels usage, while taking facts panel labels as control. Yet, beyond this, the direct relationship between product nutrient contents and purchasing preferences has not been investigated in depth for soft drinks and the US market has not been considered. Fourth, a constant limitation pointed out by previous papers (Barreiro‐Hurle et al. 2010; Findling et al. 2018; Mhurchu et al. 2018; Villas-Boas et al. 2020) is the lack of studies based on real-world consumer purchases.

In this framework, the research goal of this manuscript is to empirically explore, relying on a discrete choice logit model, how and in what fashion US consumer demand for the nutrient attributes of soft drinks reacts to the NFL update in 2020. To this end, the study asks three specific questions. First, to what extent was consumers’ purchasing behavior responsive to the nutrition label change? Second, is there a dissimilarity in the trend and the relative importance of effects concerning cities where a beverage tax is already in effect? Finally, how was the policy of label update translated in terms of impacts on consumer surplus and total welfare? To the best of our knowledge, no earlier studies have answered these questions. Accordingly, this paper mainly contributes to the literature on consumers’ revealed preference and demand for labels presenting nutrition attributes. The contribution made is twofold: unlike preceding papers, this work, with the support of data from real market transactions, is the first to involve the idea of nutrient interactions in the analysis of the US revised NFL. Through its second novel idea, this manuscript adds to the literature about the joint effect of labels and taxes on consumer preference for soft drinks. Otherwise stated, the sample analyzed allows the analysis to uniquely shed light on how label and SSB taxes can jointly affect the demand for soda, consumer surplus, and total welfare. The remainder of the document is structured as follows: "Background and preceding literature" section explains the background for understanding the US NFL and evaluates the available literature. "Methodology and data" section describes the methodology adopted in addition to the data examined. "Empirical results and discussions" section reports and discusses the empirical results, while the last section presents the conclusion.

Background and preceding literature

US NFL background

The Nutrition Labeling and Education Act of 1990 (NLEA) defines the legal frame regulating, in the USA, the labeling of packaged food and beverage products. It coerces all packaged food and drinks to display a nutrition facts panel label. By permitting buyers not only to access but also to process the nutritional information they need before making a purchase, the NLEA thus represents a stimulus for them in developing healthier dietary practices. However, the challenge of understanding a nutrition facts label (Barreiro‐Hurle et al. 2010; Ufer et al. 2022), from the perspective of shoppers, led manufacturers to start putting on front-of-package (FOP) labels, a simplified version of the panel label. While the FOP label can display other claims related to safety, sustainability, or geographical origin, some retailers also use shelf labels.

Therefore, in 2016, to render it less challenging for consumers to make informed food and beverage choices that are consistent with a healthy diet, a first major revision was made to the NLEA. This important update fell under the guidance of the FDA, the authority in charge of enforcing the act of 1990. It mandates that firms signal calorie content in larger and bold format, include added sugar in grams and, as a percent daily value, and remove calories from fat information since research shows that the type of fat consumed is more important than the amount (FDA 2022). Figure 1 depicts in detail the mandatory modifications imposed by the NFL revision of 2016.

Fig. 1

Source: U.S. Food and Drug Administration (the new nutrition facts label)

Key modifications of the NFL update policy.

Depending on the size of their sale revenues, manufacturers were asked to update their nutrition label with distinct deadlines (FDA 2022). The new policy of label regulation took $10 million as a reference. Consequently, it requires firms whose annual revenue equals or exceeds this limit to adopt the updated NFL by January 1, 2020. The study uses these two specific budget and time requirements to empirically explore, using a structural analysis approach, how consumer beverage demand reacts to the NFL update in the USA, as no previous paper elaborates on this.

Relevant literature

A wide body of literature involves food labels. For this reason, the review has been oriented toward the literature stream that underlines the effects of nutrition labels on food choice, consumer demand, and welfare. To be clear, a nutrition label can include not only the detailed facts panel label on the back side of a food product but also the simple nutrient quality claims on the FOP label. What this study calls nutrition label corresponds to the updated NFL. Different authors argue for the usefulness of nutrition labels. Galati et al. (2019) studied Italian natural wine consumers to understand their interest in label information, in particular, what information on the label influences their choice. The findings of these authors, based on a hypothetical choice experiment, demonstrate that the ingredient contents included on the wine label positively affected consumers’ choice decisions. On the other hand, Salvatore et al. (2022) employed a principal component analysis and delved into the factors leading the purchasing behaviors for functional food of Italian and Georgia consumers. They found that information from nutritional labels, among other factors, determine food purchases. Their findings also inform that the inadequate knowledge about the chosen food products same as the impossibility to buy the appropriate products constrain consumers to unhealthy eating habits. However, emerging digital technologies represent an alternative to bring transparency to consumers about the nutritional contents of the food they purchase. For example, Silvestri et al. (2023) investigated the case of the wine industry in Italy and their work established that blockchain technology contributes in delivering trust-based information to consumers through the QR codes included on the labels.

Furthermore, Ufer et al. (2022) point out that food labels signal to consumers the presence of valuable traits. Barahona et al. (2023) recall that the purpose of the information provided on a food label is to improve consumer welfare and diet quality. However, Araya et al. (2022) stress that nutritional labels require consumers to read, interpret, and understand the information displayed. Nutrient labels can help consumers improve diet quality or switch from unhealthy products to healthy ones. In accordance with this, Bollinger et al. (2011) demonstrate that mandatory calorie posting for restaurants causes average calorie consumption to fall by 6% in New York City. For Zhang et al. (2017), individuals who frequently use nutrition labels consume 92.79 mg less sodium on a daily basis. The present analysis does not ignore how the revised NFL policy influences the purchase of calorie-dense beverages with sodium included.

More specifically, label effects on consumers' choice of healthy sugary beverages have been explored in past literature. Accordingly, Roberto et al. (2016) explored how health warning labels influence parents’ choice of SSB in the USA. They reported that SSB health warning labels are associated with a decreasing purchase of such beverages for their children, in this country. Similarly, in Australia, Billich et al. (2018) conducted an online choice experiment with young adults and concluded that displaying graphic warnings on FOP labels reduces the intended choice of SSB. Moreover, through a systematic review of the influence that label formats can have on consumers’ comprehension of displayed sugar information, Scapin et al. (2021) mentioned that, compared to only numerical information, the formats that deliver an interpretation of the level of sugar contents, for example, high in sugar, were more useful for improving consumer understanding and promoting reduced sugar food choices. In the same way, Acton and Hammond (2018) noticed that high sugar label is associated with a decreasing likelihood of sugary beverage selection. But, except for FOP and warning labels, the specific contribution of the US revised NFL to reduced consumption of unhealthy soft drinks remains unexplored. Besides labels, former studies have indicated price as an additional determinant of soft drinks acquisition. More precisely, past research (Temple et al. 2016; Acton and Hammond 2018) established that higher price diminishes SSB purchases. Thus, this study will include price as an explanatory variable.

Various methods have been used in the literature to explore the linkages between nutrition facts labels and consumer demand for distinct products. Several papers (Graham and Roberto 2016; Fang et al. 2019; Neuhofer et al. 2020, 2023; Kim et al. 2021; Marchini et al. 2021) use choice experiment approaches to examine how the revised NFL label impacts food demand in the USA. It follows that compared to the previous label format, the information displayed on the updated NFL roughly requires the same visual attention from consumers (Graham and Roberto 2016; Neuhofer et al. 2020). Neuhofer et al. (2023) state that at least 75% of consumers access added sugar information on the new NFL. For Kim et al. (2021), this group of consumers is less likely to buy high added sugar products than the group of information avoiders while demonstrating healthier purchasing behaviors regardless of the product category. Regarding total sugar information on the revised label, the results from Neuhofer et al. (2023) establish that it appears more salient and gains clear attention from consumers. Adopting a different methodological approach, the current article addresses both added sugar and total sugar in the analysis. Regarding noticeable improvements, both Fang et al. (2019) and Khandpur et al. (2020) report that compared to the former version, the new NFL is less ambiguous regarding the provision of information on calories, total sugar, added sugar and serving size. However, Khandpur et al. (2020) argue that sources of added sugar remain unstated. Despite these improvements, in most situations, more health-conscious consumers and those who rely on labels for nutrient information mainly use the NFL (Fang et al. 2019; Marchini et al. 2021). Likewise, as discussed in recent papers (Fang et al. 2019; Neuhofer et al. 2023), although the new NFL induces changes in food choice behavior, its effects on purchase intentions remain mixed. This study considers all the experimental findings presented above, based on observational data, as a starting point for the present paper and empirically extend them to the US soft drink market.

Relying on structural estimation methods, other authors use scanner data to investigate the ties between food labels and consumer demand. This approach aligns more with the present work. A review of articles on the impacts of mandatory food labeling policy on consumer behavior in Chile indicated that warning labels decrease the purchase likelihood and overall demand for labeled beverages (Taillie et al. 2020); for the cereal category but not for cookies or chocolates (Araya et al. 2022). On this topic, Barahona et al. (2023) prove that the warning food labels of products whose sugar or caloric content exceeds some well-defined thresholds increase consumer welfare by 1.6% of total expenditure in the same country. In contrast, the present work includes an extended number of nutrient contents in addition to their interaction effects. According to Zhang and Gallardo (2022), households in the USA have a positive preference for convenience foods with a higher concentration of sugar, fat, sodium, cholesterol, and fiber and a lower concentration of calories.

The last category of papers the study reviewed combines experimental and structural estimation approaches to evaluate the causal impact of nutrition labels on demand for differentiated products. Using experimental labels for microwave popcorn, the authors (Kiesel and Villas-Boas 2013; Villas-Boas et al. 2020) find that conditional on which and how many claims are presented, consumers process nutrition information differently and prefer products labeled with a unique claim overall. Furthermore, Villas-Boas et al. (2020) emphasize that in the presence of experimental nutrition labels, consumers reexamine their purchases in more complex ways, inducing a demand rotation rather than an identical increase or shift in demand. The novel idea of this work consists of using real market labels, in contrast to the experimental approach adopted in the previous study.

Methodology and data

In this study, the main objective is to examine the complex relationships between nutritional labels and soft drinks purchases, shedding light on how US consumer demand for the nutrient contents of soft drinks has reacted to the NFL update in 2020. Also, the research involves three questions, each contributing on the basis of a discrete choice modeling with retail scanner data to a comprehensive understanding of the main goal under investigation. Hence, to address the first research question concerned by understating how responsive to the NFL revision was consumer demand for soda nutrient attributes, the analysis steps on two types of regression. The first regression consists of a non-causal inference regression analysis to examine the responsiveness of the purchases of eligible products to the revised NFL. In contrast, the other regression is a causal inference regression to evaluate how soda nutrients demand reacts to the label update. For the second research question relative to the induced effects of the NFL revision within cities where a beverage tax is already in application, a causal inference regression is performed allowing the research to elucidate such a query. For addressing the third research question regarding how the label update policy translated in terms of effects on consumer surplus and total welfare, the study opts for a post-estimation counterfactual simulation analysis with an emphasis on two distinct scenarios: no NFL update and no added sugar on the updated NFL.

Analytical framework

The random utility framework (McFadden 1974; McFadden and Train 2000) is the cornerstone of this paper. One crucial assumption of random utility theory follows from Lancaster (1966), who posits that utility does not directly depend on the products purchased but rather derives from desirable attributes possessed by the products bought. Built on this assumption, the random utility framework underlines that the utility derived by consumers from purchasing a product entails two main terms: a deterministic and a random component. For instance, the deterministic term refers to the product intrinsic characteristics, while the random term substitutes the unobserved factors. Regarding the subject of this paper, the well-known concern called the curse of dimensionality prevents the study from claiming that products determine utility in a direct relationship. Instead, the analysis assumes that demand for soft drinks is a function of both label attributes and price as a deterministic component and that the remaining undetectable factors can be assigned to the random component. In light of this framework, the present study carries out a discrete choice modeling approach for aggregate market data, as illustrated in the following sections.

Data

The retail scanner data obtained through NielsenIQ Datasets at the Kilts Center for Marketing Data serve as the primary data source for this article. The sample selected and analyzed spans one year with July 2019 and June 2020 as endpoints. Together, the three largest channels, supermarkets, drug stores, and mass merchandisers, represent 93% of the total revenue for the year 2019. Thus, the study excludes from the selected sample stores not belonging to these three channels. Directly collected at the point of sale, the dataset results from a systematic recording of all the transactions that occur in each participating store. Organized by universal product code (UPC), store, and week, the dataset includes unit sales in addition to a measure of average price for every product, which is obtained by dividing the value of revenue over unit sales. Furthermore, in addition to store characteristics, such as retailer chain and designated market area (DMA) or city, the dataset provides other product characteristics, such as brand and package size.

Of all the modules included in the scanner data, the analysis restricts its attention to carbonated soft drinks for various reasons. At least three reasons in addition to those already presented in the introduction section can explain the decision to focus the present work on carbonated beverages. First, soft drinks are a well-defined category. Each product in this group falls under either the carbonated or low-calorie subcategory. Consumers have no doubt that regular soft drinks, in the former category, are less healthy than diet beverages (Bonnet and Requillart 2013), the latter subcategory. Second, soft drinks offer considerable labeling variation due to the diversity of nutrient contents involved in terms of calories, sugar, added sugar, and sodium. Last, soft drinks appear in the top range of the products with a high frequency of purchase, generating the largest revenues for retailers. For example, in accordance with Dopper et al. (2023), when ranked by average yearly sales between 2006 and 2016, carbonated soft drinks rank second out of 200 categories in terms of revenue in the USA and are widely present in the NielsenIQ scanner data, while low-calorie soft drinks rank sixth.

Notably, the scanner dataset the study analyzed lacks product nutritional information and annual revenue of manufacturers. Overcoming such a challenge requires us to follow the standard practices adopted in recent studies to handle a similar concern. Therefore, this study supplements the NielsenIQ dataset with nutritional information (Villas-Boas et al. 2020; Backus et al. 2021; Barahona et al. 2023) and revenue of product manufacturers (Dopper et al. 2023). Concretely, focusing on 2019, the analysis explores the internet to obtain the annual revenue of each firm and manually search for the nutritional information of each product.

The model

Identification

To examine the impacts of the US revised NFL on consumer demand for soft drinks, the identification approach adopted by the study starts by defining products at the brand level, where the definition of brand accounts for flavor (Barahona et al. 2023; Dopper et al. 2023). Therefore, irrespective of their box size, all UPCs possessing the same brand and product name in the dataset were assigned the same product ID. In the retail scanner dataset, neither the packaging size nor the corresponding unit are identical for all UPCs. For this reason, the analysis weights each UPC quantity by unit, converting each weekly quantity of packages sold in ounces. The average price is adjusted accordingly. Subsequently, the manuscript aggregates the thousands of unique UPCs to the brand level. However, there are more than 563 brands.

Table 1 presents an overview of the sample used in this paper. To reduce the computational cost, the study includes only the top 150 brands of carbonated soft drinks by revenue for 2019. The 150 brands selected represent 98% of the total revenue. The present work ensures the overlap of these top brands with regard to 2020. Following this step, selection has been made to keep only the top 50% of the 208 DMAs in the dataset by annual revenue in 2019 annual revenue. Because the prices applied within chains are similar across neighboring stores (DellaVigna and Gentzkow 2019), the analysis consolidates the store size data to the level of the retailer chain. The study also aggregates the weekly data to the quarter level. Next, the study defines the market for soft drinks and refer to each combination of city, chain and period (Backus et al. 2021; Dopper et al. 2023) as a market. According to Table 1, the mean quarterly per ounce price for all the quantity of soft drinks sold is 0.099 US dollars. When evaluated by the distinct markets (DMA-Chain-Quarter), the mean per ounce price is almost similar to the previous value and equals 0.104 US dollars. At the market level, the minimum number of brands sold is 7, the maximum is 128, and the average is 62. At the DMA level, the number of chains ranges from 5 to 28 with an average of 14, whereas the number of brands ranges from 104 to 143 with an average of 128.

Table 1 Aggregate sample overview

A description of the variables involved in this study is presented in Table 2. To indicate the time effect, the analysis creates a dummy variable named $Post$, which takes the value one if the year of an observation is 2020 and zero for 2019. In addition, the study generates a policy dummy variable called $\text{newNFL}$, which takes the value one if the revised NFL is required for a product as of January 1, 2020, and zero otherwise. With this policy variable, all the products required to update their NFL are assigned to a treatment group, and the remaining products are assigned to a control group.

Table 2 Variable description and volume share

Ultimately, the final aggregated sample contains 150 carbonated soft drink products, the inside goods, and one alternative category of low-calorie soft drink products, the outside option. The latter offers consumers an alternative to the main category. Stated in other words, it can act as a substitute for a product in the carbonated soft drink category. For instance, inside goods refers to the group of the main products analyzed, whereas the outside option corresponds to the alternative situation where consumers did not purchase regular soft drinks but bought low-calorie soft drinks instead. Table 3 describes in detail how the treatment and control products differ in price and quantity in the sample analyzed.

Table 3 Summary of average quantity and price over treated and control products for the updated NFL

Empirical model of structural consumer demand

Following Berry (1994), the study specifies a discrete choice logit demand model in accordance with the analytical framework described earlier. The structural model adopted linearly incorporates labels and prices as product attributes (Villas-Boas et al. 2020), in the consumer indirect utility. Regarding the assumptions, the present work posits that consumers are risk neutral (Barahona et al. 2023), and nutrient labels and prices drive their choices. For this work, the products available for purchase in DMA $d$, chain $r$, and quarter $t$ correspond to $j=0, 1, \dots ,{ J}_{drt}$, including the alternative option $j=0$. Therefore, the indirect utility for consumer $i$ from purchasing product $j$ is:

$$\begin{array}{*{20}c} {U_{ijdrt} = \alpha_{j} + \alpha_{t} - \beta {\text{Price}}_{jdrt} + \gamma {\text{Nutrient}}_{jdrt} + {\upxi }_{jdrt} + \varepsilon_{ijdrt} } \\ \end{array}$$

(1)

where ${\text{Price}}_{jdrt}$ denotes the amount spent on a specific product. $\beta$ captures the consumer marginal utility as a result of paying this price. ${\text{Nutrient}}_{j\text{drt}}$ indicates the nutrient attributes of product $j$. In this form, up to eight nutrient variables are included (see Table 2 again if needed). Stated in other words, it contains three single nutrients and five nutrient interactions. $\gamma$ corresponds to the marginal utility that consumer $i$ receives from the nutrients as specific product attributes signaled through the NFL. ${\upxi }_{jdrt}$ captures any unobserved product heterogeneities, for instance, the quarterly variations in marketing factors that both consumers and firms observe but researchers do not. ${\alpha }_{j}$ and ${\alpha }_{t}$ refer to product and quarter fixed effects, respectively. Finally, ${\varepsilon }_{ijdrt}$, the random component, consumer-specific and unobserved, represents the error term.

In Eq. (1), ${{\delta }_{jdrt}=\alpha }_{j}+{\alpha }_{t}-\beta {\text{Price}}_{jdrt}+{\gamma Nutrient}_{jdrt}+{\upxi }_{jdrt}$ determines the mean utility. The analysis normalizes it to zero for the alternative good, ${\delta }_{0drt}= 0$, such as ${U}_{i0drt}={\varepsilon }_{i0drt}.$ Because consumers are utility maximizers, they purchase the product of all those available that best maximizes their indirect utility, ${U}_{ijdrt}>{U}_{ikdrt} \forall k\ne j$; and the choice probability of purchasing product $j$, ${S}_{jdrt}={\int }_{({\varepsilon }_{ijdrt})}({U}_{ijdrt})dF(\epsilon )$ yields the market share. Integrating this over the error term and assuming that ${\varepsilon }_{ijdrt}$ are independent, identically distributed, and follow a type 1 extreme value distribution (Gumbel distribution), a closed-form expression for the market share of product $j$ can be formulated as presented in Eq. (2):

$$\begin{array}{*{20}c} {S_{jdrt} = \frac{{{\text{exp}}\left( {\delta_{jdrt} } \right)}}{{1 + \mathop \sum \nolimits_{k = 1}^{K} {\text{exp}}\left( {\delta_{kdrt} } \right)}}} \\ \end{array}$$

(2)

From Eq. (2), it can be established that the market share of the alternative option is:

$$\begin{array}{*{20}c} {S_{0drt} = \frac{1}{{1 + \mathop \sum \nolimits_{k = 1}^{K} {\text{exp}}\left( {\delta_{kdrt} } \right)}}} \\ \end{array}$$

(3)

The so-called inversion (Berry et al. 1995) of Eq. (3) gives $\text{ln}\left({S}_{jdrt}\right)- \text{ln}\left({S}_{0drt}\right)={\delta }_{jdrt}.$ such that the difference between the log of observed market share for a product and the log of the observed market share of the alternative option is equivalent to the mean utility. Finally, Eq. (4) presents the full extent of the homogeneous logit model estimated in this paper.

$$\begin{array}{*{20}c} {\ln \left( {S_{jdrt} } \right) - \ln \left( {S_{0drt} } \right) = \alpha_{j} + \alpha_{t} - \beta {\text{Price}}_{jdrt} + \gamma {\text{Nutrient}}_{jdrt} + {\upxi }_{jdrt} } \\ \end{array}$$

(4)

Next, the ratio of total product $j$ sold in a market, ${Quantity}_{jdrt}$, to the potential market size, ${M}_{dr}$, provides the observed market share ${S}_{jdrt}$ (Villas-Boas et al. 2020; Dopper et al. 2023). However, the quantity of all products sold over a city-retailer-quarter implies market size ${M}_{drt}$. Therefore, within each retailer-city, this paper assesses the potential market size as the highest market size.

Variable of interest, instrument, and estimation procedures

The study estimates two specifications for the variable of interest ${\text{Nutrient}}_{jdrt}$ included in Eq. (4). The first specification explores the responsiveness of the eligible products to the revised NFL. This implies that ${\text{Nutrient}}_{jdrt}= {\text{NFLattribute}}_{jdrt}*Post$. Thus, only the treatment products, $\text{newNFL}=1,$ as defined previously serves this purpose. Therefore, for a given product, the nutrient variable takes the value one during the postpolicy period and zero otherwise.

However, in addition to the treatment products, the second specification involves the control products, $\text{newNFL}= 0$, then allowing the study to estimate a causal effect of the label policy under consideration. In this case, the variable nutrient receives the value one exclusively in the postpolicy period if the updated NFL is required for a product. Consequently, ${\text{N}utrient}_{jdrt}= {\text{NFLattribute}}_{jdrt}*\text{newNFL}*\text{ Post}$. The latter specification mutates Eq. (4) into a structural demand model embedded in a triple difference method (Villas-Boas et al. 2020; Chetty et al. 2009). It follows that the model in Eq. (4) has been estimated separately by distinct variables: newNFL = 1 under the first specification; newNFL = 1 and newNFL = 0, SSB tax = 0, and SSB tax = 1 under the second specification; following Taylor and Villas-Boas (2016). They adopted a discrete choice linear logit model and separately estimated it for several variables. Yet, inferring a causal impact requires the treatment and control groups to have similar trends (Codjia 2022) in the absence of the policy in such a setting.

Various approaches can be used to demonstrate distinct degrees of compliance with this crucial assumption, but this study relies on two of them. Figure 2, presenting the first verification method, reports the trend in market share with regard to the core nutrients. Regarding the second verification procedure, the reduced-form regression estimates presented in Table 4 provide further evidence. This verification procedure requires the study to create a pseudo treatment period corresponding to the last quarter of 2019. Hence, the observed market share is the dependent variable used for this reduced-form regression. The analysis estimates it using the panel ordinary least squares method. In each of the three situations considered, all point estimates are statistically non-significant, evidencing that before the US revised NFL policy, treatment and control products are alike in the trends of their market share.

Fig. 2

Source: Own computation by authors from NielsenIQ retail scanner data

Trend of prepolicy market share for main nutrients.

Table 4 Reduced-form regression estimates for prepolicy period

Finally, accounting for the concern of price endogeneity, which is inherent to most structural demand models, the analysis finds it necessary to include the Hausman price instrument (Hausman 1996; Nevo 2001) as a control variable for estimating the discrete choice model described in Eq. (4). For instance, the Hausman instrument refers to the average price of the same product in other markets. But its usage requires some caution, as the analysis must positively confirm that it is not a weak instrument in the estimated models. To this end, the study relies on the first-stage F test for weak instruments. The test outcome is reported along with the regression results in the subsequent section. In all cases, the first-stage F statistic is large enough to reject the null hypothesis of a weak instrument. Furthermore, the independent variables successfully pass the test for the detection of critical collinearity issues. But it is worth mentioning that HS has not been evaluated alone as a single nutrient variable in the model because it has been automatically reduced during the estimation process (Conlon and Gortmaker 2020).

Thus, regardless of the specifications described above, the analysis estimates Eq. (4) using the two-stage generalized method of moments (GMM). The study does this with the package PyBLP through Python 3 (Conlon and Gortmaker 2020). The optimization algorithm used to estimate the model was set to BFGS, and the tolerance level for its convergence criterion to 10^e−6.

Empirical results and discussions

Responsiveness of the purchases of eligible products to the revised NFL

Column 1 of Table 5 reports the point estimates for measuring how purchases of eligible soft drinks reacted to the US NFL update of January 1, 2020. The results indicate positive and significant mean marginal utilities for calories, sugar or added sugar. This implies that after the update, purchasing carbonated soft drinks with a revised NFL, either superhigh in calories, high in total sugar or containing added sugar increases the utility derived, on average, for the sampled consumers. However, it is unlikely that only the soft drinks with single a nutrient contents are sold on the market, and it is necessary to look at the case of the products which contains a combination of nutrients.

Table 5 Structural logit demand regression estimates

Regarding the nutrient interactions, the results show positive and significant mean marginal utilities for two of them: sugar–sodium in addition to calories–sugar–added sugar. The reverse is true for three combinations: calories–sugar, sugar–added sugar, and calorie–sugar–sodium. It can be deduced that post the label update, when they lead a purchase of soft drinks with an updated NFL, the attribute interactions mentioned in the first case increase the mean utility consumers derive, while the interactions cited in the reverse case decrease the mean utility. In accordance with the parameter estimates, the calories–sugar interaction exhibits the highest magnitude. This implies that relative to the eligible products, the NFL update policy was mostly beneficial in reducing the consumer preference for soft drinks that were both superhigh in calories and high in total sugar content.

As expected, price presents a negative coefficient, meaning that consumers dislike higher prices, but the effect is not significant. Even the aforementioned results, exclusively from a perspective of the eligible products, provide initial evidence that consumers were not unresponsive to the NFL change policy, yet it is difficult to draw a conclusion regarding causality. By deepening the early results, the next section thus strengthens the supporting evidences.

Demand response to nutrient label update

The parameter estimates reported in Column 2 of Table 5 present, in overall, how revising the US NFL impacted consumer preference for nutrition attributes with respect to the soft drinks market. Of the single nutrition attributes included in the model estimated, only calories exhibit a positive and statistically significant average marginal utility. Despite the NFL revision, consumers still prefer soft drinks with superhigh calorie content. The mean marginal utility of the remaining two single nutrients is positive but not statistically significant. As nutrient contents and separately considered, sugar and added sugar have no significant impact on the mean utility consumers derive from purchasing carbonated soft drinks under the NFL update policy, in accordance with the estimated results.

Regarding the nutrition combinations, only calories–sugar and calories–sugar–sodium show a significantly negative mean marginal utility. The implication is that, in overall, the label update policy has been beneficial in reducing consumers’ preference for soft drinks superhigh in calories with high sugar content or simultaneously superhigh in calories, high in sugar, and high in sodium. This finding, even more specific in accounting for nutrient interactions, is consistent with Zhang et al. (2017), who argue based on an analysis of ready-to-heat meals products that nutrition label use is associated with marginally lower consumption of high-sodium foods in the USA.

In contrast, two interaction variables in the estimated model present positive and significant average marginal utility: sugar–sodium and calories–sugar–added sugar. This means that the sampled consumers still prefer soda that is either high in total sugar and sodium or superhigh in calories, total sugar, and added sugar following the NFL revision. Again, the calories–sugar interaction displays the highest magnitude among all the parameter estimates. The label revision policy was mostly important in lowering consumers’ preference toward soft drinks superhigh in calories and with high sugar content. Consistent with Van der Merwe et al. (2022), nutritional information displayed on food labels can lead consumers to make healthy food choices. Also, this result aligns with Mhurchu et al. (2018) whose analysis, in opposite to this work was based on a random control trial, noticed a significant relationship between label use and purchase of healthy products. Without accounting for the SSB tax applied in some US cities, however, the analysis may remain less compelling, and particularly the third research question unaddressed.

Implications of revising NFL policy for SSB tax: label and tax joint effect

SSB taxes have been applied since 2015 in some US metropolitan areas, including Seattle, Washington; Boulder, Colorado; the District of Columbia; Philadelphia, Pennsylvania; and four cities of California: San Francisco, Berkeley, Oakland, and Albany. Hence, the study isolates these eight cities implementing the SSB tax and estimates the model by the variable SSB tax = 1 and the alternate case of SSB tax = 0. Column 4 of Table 5 thus presents the estimates of evaluating demand response to nutrient label update, within DMAs impacted by the SSB tax.

The estimation outcome shows for calories a mean marginal utility exerting a significantly positive impact on average utility. The study infers that even though the NFL was updated starting in 2020, consumers still prefer buying soft drinks superhigh in calorie content within US cities implementing the SSB tax. According to the results, none of the remaining nutrients, alone or in any interaction, had a significant effect on mean utility. The analysis deduces that the effects of the US revised nutrient label are somehow ambiguous in terms of shifting consumers from unhealthy beverages to healthy ones inside the cities implementing the SSB tax. Similar to this finding, Cawley et al. (2019), in studying the Philadelphia beverage tax, find that the tax overall had no detectable impacts on beverage consumption by adults because residents increased acquisitions of taxed beverages outside of the city.

Consistent with the study expectation, the outcome exhibits significant but negative marginal utility of price, implying that higher prices lessen the resulting mean utility that consumers derive. Such a result is consistent with findings from the studies of Cawley et al. (2019) and Barker et al. (2022), who noticeably stress that measured in cents per ounce of beverage, the SSB tax significantly decreases the purchase of taxed drinks in stores located within concerned cities. Price displays the highest magnitude of marginal utility for the current case under consideration, but one plausible explanation can be found in the fact that firms generally pass through SSB taxes to consumers.

Column 3 of Table 5 reports the estimates of the demand response to the nutrient label revision for DMAs not affected by the SSB tax. However, the results here are quite similar to those of the overall model, as presented in the earlier section. The results on the two kinds of cities covered provide some evidence that, compared to tax, label policy can produce more tangible effects in regard to the promotion of healthy purchasing behavior among consumers. For Bernheim and Taubinsky (2018), whose findings are similar to this result, policy interventions implemented by the means of informational labels can be more efficient than taxes since their effects are better targeted.

Counterfactual analysis: consumer surplus and welfare

One advantage of adopting a structural approach analysis is the possibility of completing, post-estimation, rigorous counterfactual simulations. With attention to the US soft drinks market, the current paper did not miss leveraging such a possibility for the purpose of evaluating the short-term impacts of the updated NFL on consumer surplus and welfare since 2020. To start the counterfactual analysis, the manuscript assumes that price remains unchanged. Two scenarios are investigated: first, no change to the US NFL and, second, no disclosure of added sugar on the updated NFL. Following Small and Rosen (1981), the mean expected consumer surplus is equivalent to Eq. (5).

$$\begin{array}{*{20}c} {{\text{CS}}_{jdrt} = \frac{1}{\left| \beta \right|}ln\mathop \sum \limits_{j} {\text{exp}}\left( {\alpha_{j} + \alpha_{t} - \beta {\text{Price}}_{jdrt} + \gamma {\text{Nutrient}}_{jdrt} } \right) + {\text{Contante}}} \\ \end{array}$$

(5)

The estimated value defines a population-normalized consumer surplus in a specific market. With the baseline model where the label has been updated with the disclosure of added sugar, the analysis computes consumer surplus, firm profit, and the associated total welfare, which measures profit and surplus combined. Under a given scenario, differentiating the simulated value from the baseline result permits the analysis to obtain the resulting variation or change. Table 6 summarizes the estimates.

Table 6 Counterfactual simulation

Regardless of the sample used, the estimates show a negative value for variation in firm profit. The manuscript infers that firms do not obtain net gains from complying with the policy requiring them to update the NFL for the soft drink market. This also applies to disclosing added sugar information on the updated label. In contrast, estimated variations in consumer surplus present a positive value in each of the distinct situations investigated. The study thus provides evidence that consumers obtain a surplus from the policy of new nutrition labels, the soda market considered. However, the gain is less for consumers in cities where the SSB tax is applied compared to other cities where it is not.

Regarding total welfare, the findings indicate a trend similar to that of consumer surplus. The study deduces that adopting a revised version of NFL in the USA from 2020 was beneficial to total welfare. A recent study finds a similar result: Barahona et al. (2023) used a random coefficient logit model, a model slightly distinct from the one used in this study and observed that consumer welfare increases by 1.6% of total expenditure in Chile following the application of warning food labels.

Discussions

This work describes and quantitatively analyzes the effects the applications in 2020 of a revised NFL has induced on demand for soft drinks beverages. In particular, the modifications observed in the nutritional characteristics of purchased drinks, as well as the implications for consumer surplus and total welfare have been studied. Thus, it is convenient to further discuss the differences and similarities between the present research and other studies that have delved into food labels and their relationships with consumer demand and preferences for nutrient contents. For this purpose, given that research results are always contingent on the methodologies employed, not only the discrepancies and the convergences in terms of findings need to be discussed, but the ones about the designs also need to be examined.

Therefore, regarding the research design, this work is different from the past studies on the US revised NLF; because previous studies relied on an experimental design, where sophisticated eye-tracking technology has been employed (Graham and Roberto 2016; Neuhofer et al. 2020, 2023), and were prominently interested in elucidating the links between label viewing and the healthiness of the products consumers choose to purchase. Even though most of these studies had considered multiple products in their analysis, they have estimated a reduced-form model where causality is missed and have not extended their work to the potential effects of the new label on welfare and the surplus of consumers. In contrast, the goal of this research is exclusively oriented toward unveiling the shifts in the nutritional characteristics of purchased soft drinks, following the NFL update. Yet, past work has underlined that the new NFL has been viewed by more than half, specifically 58%, of the participants involved in an experiment the authors performed (Graham and Roberto 2016). Likewise, concerning added sugar, one of the additional details displayed on the updated NFL, past works mentioned 75% for experiment participants who read it before making their choice (Kim et al. 2021; Neuhofer et al. 2023), thus a largely higher value. But, under the design adopted for the present research, whether the consumers read the NFL before every single purchase they make, or not, has a relatively low importance. Sated in other words, what matters in the analysis carried out throughout this article is the nutrient contents associated with the products consumers have been purchasing after the label policy has been revised. For such a reason, the study has only involved purchases made from real market transactions and accounted for a wide range of retail chains across the four US census regions.

Despite these differences, the study has some features of convergence with several noticeable research. Accordingly, involving a set of more than 200 brands of carbonated soft drinks products commercialized within the US retail markets, and unlike the previous studies on NFL, this analysis depends on a structural logit model (Berry 1994; Taylor and Villas-Boas 2016; Villas-Boas et al. 2020) to access the types of shifts occurred in consumer preferences for soda nutrient contents, as well as the implications for consumers surplus and total welfare. Defined as the combination of DMA-Chain-Quarter, a total of 5782 unique markets has been studied whereas the real market purchase transactions involved in the analysis have been recorded between the last two quarters of 2019 and the first two ones of 2020. In addition to the empirical estimation of the causal impacts, while assessing the revised label policy effects, the study has separately considered the situation of the metropolitan regions where the SSB tax is already in application. However, the design of this research is slightly similar to the one of other recent studies which have covered the effect of food labels on consumer demand (Villas-Boas et al. 2020; Araya et al. 2022; Zhang and Gallardo 2022; Barahona et al. 2023).

From the perspective of the findings, overall, this study is coherent with prior papers. Accordingly, the outcome of this work demonstrates that information displayed on product nutritional labels, among other factors, determines consumers' purchase behaviors. This aligns well with prior research, as different authors (Roberto et al. 2016; Billich et al. 2018; Galati et al. 2019; Salvatore et al. 2022; Zhang and Gallardo 2022) have reported a similar result. In particular, the findings of this study show that the adoption of a new NFL has induced changes in consumer demand for the nutrient contents associated with the products they buy. Also, this is consistent with the results of recent studies (Fang et al. 2019; Araya et al. 2022; Barahona et al. 2023; Neuhofer et al. 2023), whereas distinct methodologies have been adopted. Moreover, for what concerns the implications of food labels for consumer valuation and welfare, the results underline that the nutritional information provided on a label can improve them. This also lines up with a recent work (Barahona et al. 2023); where it has been reported that the application of a nutritional warning label has increased consumer welfare by 1.6% of total expenditure, even though, compared with this study, some variations can be noticed in the methodology the authors employed. Moreover, consistent with preceding works (Villas-Boas et al. 2020; Araya et al. 2022; Zhang and Gallardo 2022; Barahona et al. 2023), the results indicate a negative sign for estimated price coefficients, meaning that higher soda prices are not attractive for consumers, regardless of information provided on labels. This implies that, in addition to nutritional labels, price also determines soft drinks purchases. For instance, as prices increase, consumers are less likely to purchase sugary beverages (Temple et al. 2016; Acton and Hammond 2018).

Despite these coherences with former studies, as mentioned above, some divergences in results have been noticed. Nevertheless, the findings show a positive sign for the single nutrients included in the model while some of their interactions exhibit a negative sign, emphasizing that the revised NFL has been more efficient in lowering consumer preference for soft drinks which simultaneously combine two or three negative nutrient attributes, for instance, calorie–sugar, sugar–added sugar, and calorie–sugar–sodium. Therefore, it can be said that such a result represents a major added value from the present research as the existing studies missed considering the effects of nutrient interactions on consumer preferences. An additional added value from this study is related to the joint effects of label and SSB tax that this work has allowed to unveil. Thus, for the situation of the cities where the SSB tax is in application, the findings show that the updated NFL did not permit a reduction of consumer preferences for specific unhealthy nutrient contents. While some past studies have considered the SSB tax separately (Cawley et al. 2019; Barker et al. 2022); others have extended their analysis to tax just through a counterfactual simulation (Araya et al. 2022; Barahona et al. 2023); and another simply failed to engage the SSB tax effects in its analysis (Zhang and Gallardo 2022). On the other hand, because causality has been involved in this analysis, the interpretation of the results has necessitated a lot of caution, regardless of the validity tests performed. For example, the analysis of the responsiveness of the purchases of eligible products to the revised NFL cannot be interpreted from a perspective of causal effect, as a control has not been involved.

Finally, the results indicate a negative but statistically non-significant price coefficient for the situation where the responsiveness of the eligible products is examined, the same for the case of the cities where the SSB tax is not applied. However, this non-significance of price effect does not imply that consumers are indifferent to price. Rather, for each of these two separate situations, the study stresses that better than price, label nutrient contents determine consumers' preference for soft drinks (Roberto et al. 2016; Billich et al. 2018; Scapin et al. 2021). For instance, all the markets analyzed considered, per ounce soft drinks price ranges from 0.024 to 0.281 US dollars, while its average value is relatively low and equals 0.104 US dollars. However, these non-significant prices should be taken as an exception because for both the overall model and the case of the cities applying the SSB tax, as the study highlights, in addition to labeling nutrient contents, price has a statistically significant effect on soda purchase. Therefore, as an exception, the non-significant prices could have perhaps resulted from the absence from the estimated model of other consumer choice determinants like label color, product disposition, and consumer heterogeneity.

Conclusion

This paper discusses how and in what fashion US consumer demand for nutrient contents in soft drinks reacts to the NFL revision in 2020. This work uniquely highlights the complexity of consumer preferences for combinations of nutritional characteristics with attention to carbonated soft drinks. The analysis finds evidence that consumers have modified their purchasing behavior in response to the change in nutrition labels. However, they did not systematically switch from unhealthy beverages to healthy ones. For instance, the single nutrient variables contained in the estimated model depict unhealthy attributes. So, their interactions refer to a combination of negative attributes. The results from this study indicate a positive sign for all the single nutrients in the estimated model, while some of their interactions exhibit a negative sign. Subsequently, the consumers are tolerating products with a single negative attribute whereas they become less tolerant and do not prefer some of the products which combine many negative attributes at the same time. Furthermore, the joint effect of label and tax was surprisingly not in favor of reducing the preference for unhealthy nutrient contents. Although the NFL was updated starting in 2020, consumers still prefer buying soft drinks superhigh in calorie content within US cities implementing the SSB tax. Finally, the new nutrition label enhanced the average surplus of consumers and total welfare, but the gain is less for consumers within cities with the SSB tax.

Beyond this, as implications, the article provides policymakers in charge of designing tools for reducing obesity and overweight with reliable insights to direct their future decision-making. The work also informs firm executives, retailers, and marketing specialists about valuable data-driven points which can be useful for growing their business size and profit. Ultimately, researchers can find avenues for future research directions whereas the general public can become more educated and more aware of nutrition label usefulness in making informed choices consistent with a healthy diet.

Finally, this study has three limitations. First, the analysis does not include other products, such as cookies, popcorn, and ready-to-eat cereals, which could further expand its scope and allow a wide comparison of the NFL effect across products. Like sugary beverages, these products have a considerable amount of varying unhealthy nutrient content as well. For example, ready-to-eat cereals and cookies are high in unhealthy nutrient contents like sugar, added sugar, and calories while popcorn products are high in saturated fats and trans-fat in addition to calories. Therefore, the scope of the study could have been widened further with the inclusion of these products. Second, by relying on a homogeneous logit model, the work does not explicitly account for heterogeneity in consumer preference. In addition to the product characteristics the work involves, other consumer-related factors could have influenced the purchasing behaviors observed. Namely, income, gender, age, marital status, and household size are other potential consumer-related characteristics that have not been involved by the model employed, as they are not available in the datasets analyzed. Third, other factors such as labels colors, product positioning in the stores, and taste can intervene in the choices made by consumers in addition to label information. Undoubtedly, certain colors can be more eye-appealing than others. Then, it is possible that beyond the label's nutritional information, label colors could have also influenced product acquisition. In the same way, certain products could have been better positioned than others in the stores, thus increasing the likelihood of purchase for the products well located. The same reasoning applies to taste because people have differentiated preferences for product tastes.

However, these limitations do not undermine the quality of the study. Moreover, because the methodology behind the manuscript successfully passes the required validity tests, the reported results and conclusions drawn on the basis of the findings are valid and do not suffer from any reliability issues. It follows that avenues for future research encompass two perspectives. To further access the efficacity of the label policy across multiple products, future research needs to investigate the updated NLF effects for additional foods like ready-to-eat cereals, cookies, and popcorn. The selection of these products might also permit evaluating the label policy impact for saturated fats and trans-fat which have not been considered in this work. To engage the factors connected to consumer heterogeneity, the second perspective for future research consists of adopting an alternative estimation model. For instance, employing a random coefficient logit represents one way to incorporate efficiently heterogeneities in consumer characteristics. Finally, a dynamic demand modeling approach can be used to properly handle the potential issues related to infrequent product purchases, changing consumers behaviors, same as the additional factors cited above.

Appendix

See Tables 

Table 7 Region and state of all 104 DMAs included the data sample

7 and

Table 8 Variance inflation factor (VIF) test for multicollinearity

8.