A. Experimental design

This study uses Case I (object-case) best-worst scaling (BWS) to understand consumer preferences for various hard cider policies (Finn and Louviere, Reference Finn and Louviere1992; Louviere, Flynn, and Marley, Reference Louviere, Flynn and Marley2015).Footnote ¹ In BWS experiments, respondents are presented with several alternatives and asked to select their most preferred (best) and least preferred (worst) options. BWS has become common in the food and beverage economics literature (e.g., Bazzani et al., Reference Bazzani, Gustavsen, Nayga and Rickertsen2018; Caputo and Lusk, Reference Caputo and Lusk2020; Smith et al., Reference Smith, Lal, Oluoch, Vedwan and Smith2021) because it overcomes traditional shortcomings associated with Likert scale questions (Louviere, Flynn, and Marley, Reference Louviere, Flynn and Marley2015). More specifically, BWS forces respondents to consider tradeoffs between alternatives, reduces respondent cognitive fatigue, and limits scale-use bias (Baumgartner and Steenkamp, Reference Baumgartner and Steenkamp2001; Scarpa et al., Reference Scarpa, Notaro, Louviere and Raffaelli2011; Caputo and Scarpa, Reference Caputo and Scarpa2022).

In this experiment, consumers consider ten hard cider policies. A survey of apple growers (Ostrom et al., Reference Ostrom, Conner, Tambet, Smith, Sirrine, Howard and Miller2022) informed this policy list, as did focus groups with cider producers, key informant interviews with industry stakeholders, and suggestions from the American Cider Association. A U.S. Department of the Treasury report (2022) evaluating recent alcohol consolidation and competition concerns was another important source since it contained views expressed by cider industry stakeholders. Table 1 presents the complete list of policies included in the BWS.

Table 1.

Policies included in the best-worst scaling experiment

The first four policies are tied to the current regulatory structure of alcohol markets and considerations made in the recent Treasury report on competition in the U.S. alcohol market. The report was the product of Executive Order 14036 (Promoting Competition in the American Economy, 2021). It discusses recent trends in the industry, highlights potential areas for regulatory reform, and offers recommendations for boosting opportunities for smaller businesses and entrepreneurs. Informed by this report, we include policies that harmonize regulations across different alcohol sectors (i.e., beer, wine, and spirits), harmonize state-by-state alcohol regulations, ease interstate distribution, and enable retail partnerships. A policy to harmonize regulations across alcohol types is included because there is substantial variation in policies for beer, wine, and distilled spirits. For example, some states only allow beer sales at grocery stores, while others allow beer and wine (and distilled spirits) sales (e.g., Rickard, Costanigro, and Garg, Reference Rickard, Costanigro and Garg2013). Further, alcohol regulations also vary across state lines (Staples, Chambers, and Malone, Reference Staples, Chambers and Malone2022), and this heterogeneity can lead to differences in entrepreneurial and market outcomes in the craft beverage space (Anderson, Meloni, and Swinnen, Reference Anderson, Meloni and Swinnen2018; Malone and Lusk, Reference Malone and Lusk2016). Thus, it is worth considering how consumers feel about creating similar policies across alcohol types and state lines. Easing interstate distribution laws is included because of the growing interest in direct-to-consumer sales through e-commerce markets. Lawmakers and industry groups have debated the policies contributing to competitive barriers in wine e-commerce sales since at least the early 2000s (FTC, 2003), as limiting interstate distribution restricts consumers from accessing out-of-state products. However, there is a renewed interest in policy reform due to advances in e-commerce platforms, evolving direct-to-consumer laws, and growth in small and regional producers. Similarly, a question focused on enabling retail partnerships is included in our list because the American Cider Association notes that tied-house laws disadvantage small producers who may work with single retailers and are currently prohibited from promoting these locations (U.S. Department of the Treasury, 2022). While these policy changes are not direct recommendations from the report, we determined that evaluating these topics within the report warranted further investigation.

Next, four different types of labeling policies are considered, including allergen disclosure, nutrition facts, ingredients, and the use of concentrated juice. Alcohol labeling in the United States is complex, and requirements for one beverage category may not apply to others.Footnote ² Consumers have indicated support for labeling food and non-alcoholic beverages—in a 2019 survey, for example, the U.S. Food and Drug Administration (Lando, Verrill, and Wu, Reference Lando, Verrill and Wu2021) found that 87% of U.S. adults use nutrition labels when making food purchases, 80% report using other label information on the front of packaging, and 14% indicate that they have food allergies.

Sulfites are the most common type of potential allergen for the cider industry (TTB Alcohol Labeling and Formulation Division, 2023), used as preservatives to help wine and cider stop fermentation and slow chemical reactions. A nutrition facts panel would standardize the mandatory labeling information across all ciders, including information such as calories per serving. Labeling ingredients is another potential change and would likely require producers to list each ingredient used in production, including food, color additives, and flavorings (TTB, 2022). Lastly, there is debate in the cider industry on labeling concentrated juice. Concentrated juice refers to the syrup-like liquid created from the heating, extraction, and filtration of the fruit. The process strips much of the flavors and aromas from the input, and its use can require additional ingredients, such as malic acid and caramel color, during the production process to restore the desired cider characteristics. The process provides large producers with a steady supply of inputs year-round. However, smaller orchard-based producers may argue that ciders manufactured with concentrated juice should be labeled as such to assist with product differentiation (Alworth, Reference Alworth2019). Two-thirds of the apple juice supply in the United States includes concentrate imported from China (Gale, Huang, and Gu, Reference Gale, Huang and Gu2010), and imported juice comprises a substantial portion of hard cider produced in the United States (Miles et al., Reference Miles, Alexander, Peck, Galinato, Gottschalk and van Nocker2020).

The last two policies considered in the experimental design relate to broad and regional industry development. We assessed consumers’ relative preferences for allowing cider producers to mark vintage (or harvest date) on the label, as well as for increasing funding efforts for regional cider development. Vintage is well-known in the wine industry (Ashenfelter, Reference Ashenfelter2010; Storchmann, Reference Storchmann2012), serving as a proxy for quality and a potential marketing mechanism for producers. The cider industry has developed an interest in vintage due to the recent growth in imperial ciders, or ciders with 8.0%+ alcohol by volume (ABV) (Shreeves, Reference Shreeves2022). These products are well-suited for aging and could provide producers with a niche marketing opportunity to offer products that fetch a premium price. Our final question assesses support for increasing funding for regional cider development. NielsenIQ (2023) reports that regional and local brands now account for approximately 54% of cider sales in the United States. With a budding interest in local beverage value chains, government support (e.g., grants) could be provided to establish and expand these local markets through infrastructure, R&D, and other forms of support.

As BWS relies on repeated choices and trade-offs between policy sets, respondents were presented with six groups of policy choices. Each question had five of the ten policies generated using an incomplete block design (Bazzani et al., Reference Bazzani, Gustavsen, Nayga and Rickertsen2018), where the alternatives were randomly assigned (Boyle et al., Reference Boyle, Holmes, Teisl and Roe2001; Lusk and Norwood, Reference Lusk and Norwood2005). In each of the six tasks, they were to select their most and least preferred option.

B. Survey instrument and sampling strategy

The BWS experiment was embedded in an online survey designed and distributed through Qualtrics to a panel of cider consumers across four states: Michigan, Vermont, Washington, and Wisconsin. These states were selected because they were among the top 12 in the number of cider producers per capita (Conway, Reference Conway2020), and therefore, residents were more likely to have access to local and craft cider options. For example, Vermont has the highest number of cider producers per capita, and Washington and Michigan combine to account for more than 70% of total U.S. apple production (USApple, 2021). These states were also the focus of a larger project exploring the potential for cider-specific varieties to improve economic outcomes for small and medium-scale orchardists (Ostrom et al., Reference Ostrom, Conner, Tambet, Smith, Sirrine, Howard and Miller2022). Respondents were above the legal drinking age (21+) and had consumed cider in the past month. The data were collected from February 7 to February 23, 2023, with 688 cider consumers completing the study. Table 2 presents the summary statistics of the sample.Footnote ³

Table 2.

Sample demographics, cider consumption habits, and preferences

The survey began with several demographic questions and questions regarding cider consumption habits. This included questions on the frequency of consumption relative to other alcohols and the characteristics of cider that were important and unimportant to them when purchasing cider.Footnote ⁴ Respondents then completed the main portion of the study: the six BWS choice tasks. They were told there would be six questions about their preferences for different government policies and programs related to hard cider production. In each question, they would choose the policy they most preferred and least preferred out of five options. Figure 1 presents an example BWS question.

Figure 1.

Example best-worst scaling (BWS) task.

C. Estimation procedures

BWS estimation lies in random utility theory (McFadden, Reference McFadden and Zarembka1974), where respondent i receives indirect utility ${U_{ipt}}$ from selecting policy pair p in choice task t. In each choice task, the respondent is assumed to consider all potential policy pairs and choose the one that maximizes their utility. In other words, their selection maximizes the difference between their most preferred and least preferred policy. If J denotes the total number of policy alternatives in each choice task, then there are J(J – 1) policy pairs each respondent must consider for each of the six questions. In our setting, with five policies per choice task, there are 20 possible policy pairs. We thus estimate the multinomial logit (MNL) model:

(1)\begin{equation}{U_{ipt}} = {\beta _j} - {\beta _k} + {\epsilon _{ipt}},\end{equation}

where ${\beta _j}$ and ${\beta _k}$ are parameter estimates for best policy j and worst policy k, respectively. These parameter estimates are relative to a baseline policy normalized to zero. Then, ${\epsilon _{ipt}}$ is an independent and identically distributed (IID) and type I extreme value error term.

Since indirect utility is random, we can only model the unconditional probability that a given combination maximizes the difference in utility among the J(J – 1) pairs (Lusk and Briggeman, Reference Lusk and Briggeman2009). When alternative j is the best and k is the worst, this is formally defined as the closed-form, logit choice probability:

(2)\begin{equation}Pro{b_i}\left(\, {j \ is \ best \ and \ k \ is \ worst} \right) = {{\exp \left( {{\beta _j} - {\beta _k}} \right)} \over {\mathop \sum \nolimits_{l = 1}^J \mathop \sum \nolimits_{m = 1}^J \exp \left[ {\left( {{\beta _l} - {\beta _m}} \right) - J} \right]}}.\end{equation}

The MNL output provides marginal utility parameters for each policy alternative, where the sign suggests whether, on average, consumers prefer the policy relative to the baseline. However, an alternative and more informative approach to analyzing the output is to analyze the preference share ${\textrm{(}}{{{\Upsilon }}_j})$ of each policy. Preference shares are a way to normalize the model output as a measure of relative importance and are calculated as follows:

(3)\begin{equation}{{{Y }}_j} = {{{\textrm{exp}}\left( {\widehat {{\beta _j}}} \right)} \over {\mathop \sum \nolimits_{l = 1}^J \exp \left( {\widehat {{\beta _l}}} \right)}}.\end{equation}

By construction, the preference shares are bounded between zero and one and sum to one across the ten policies. The larger the preference share is for a given policy, the more important it is, on average. Thus, comparisons can be drawn across all policy alternatives, including the baseline.