A. Search costs and regret: Consumer behavior around large choice sets

Choice overload, or choice paradox, can be generally defined as the counterintuitive phenomenon of decreased consumer propensity to choose or buy any given product from a larger choice set. It has been identified in a range of settings, with some of the literature presented next.

Iyengar and Lepper's (Reference Iyengar and Lepper2000) seminal paper studied the impact of choice overload on consumer behavior. Their results suggest that extensive choice is initially more attractive, but this attractiveness is not reflected in subsequent purchasing decisions. Consumers exposed to limited choices purchase considerably more and have higher purchase satisfaction than consumers exposed to extensive choice sets. The authors suggest search costs and regret explanations as possible drivers behind the observed behavior, but do not explicitly provide a formal model.

The search cost explanation for choice has been evaluated in many empirical settings (Gabaix, Reference Gabaix, Laibson, Moloche and Weinberg2006; Reutskaja et al., Reference Reutskaja, Nagel, Camerer and Rangel2011), with the implication that choice overload can be mitigated by reducing search costs. Behavioral nudges, advertising, expert scores, and reviews have all been proposed as ways of simplifying choices (Johnson et al., Reference Johnson, Shu, Dellaert, Fox, Goldstein, Häubl, Larrick, Payne, Peters, Schkade and Wansink2012; Nelson, Reference Nelson1974; Malone and Lusk, Reference Malone and Lusk2019). Boatwright and Nunes (Reference Boatwright and Nunes2001) find that reducing the number of stock keeping units (SKUs) within a category in a grocery store leads to an overall increase in sales and higher consumer spending. The authors hypothesize that the removal of redundant items and “clutter” in the choice set reduces consumer search costs and is behind their results. Malone and Lusk (Reference Malone and Lusk2017, Reference Malone and Lusk2019) look at how the number of options on a bar menu or in a choice experiment, along with information provision, might affect choice overload in food and beverage markets. They find that choice overload is present but is remediated through the inclusion of expert rankings on the menu. They hypothesize that expert suggestions can reduce search costs by providing a reliable information signal about quality and a potential choice decision heuristic. Chernev (Reference Chernev2003) also finds that making respondents rank the importance of various attributes before making a choice reduces the preference for small choice sets by simplifying consumer choice. Schwartz et al. (Reference Schwartz, Ward, Monterosso, Lyubomirsky, White and Lehman2002) conclude that choice overload can be moderated by the seller by lowering search costs and catering to specific target markets.

In general, search costs have been successfully used in the theoretical modeling of consumer behavior in relation to choice set size. Kamenica (Reference Kamenica2008) uses choice overload as an example of contextual inference in markets, with theoretical model results showing that generally consumers who are less informed about a particular product will prefer a brand with a smaller product line (choice set) to maximize preference matching. Kuksov and Villas-Boas (Reference Kuksov and Villas-Boas2010) explicitly model the impacts of search costs and preferences for variety on consumers decisions to conduct a search or to avoid a purchase altogether. The modeling results suggest that both too many and too few alternatives can lead to a decreased propensity for any choice among alternatives. The authors suggest that too many choices might overwhelm consumers with high search or evaluation costs, while too few choices reduce the likelihood of a preference match. This suggests that modeling choices can lead to different predictions on the impact of the number of options in the choice set, highlighting the need for empirical work.

In addition to search costs, regret theory has also been proposed as an explanation for the choice paradox. Consumers might avoid an attractive ex-ante choice out of fear of being disappointed with it after the fact. In other words, consumers tend to include expectations of ex-post choice dissatisfaction in their decision making (Taylor, Reference Taylor1997). Comparatively fewer papers have focused on the impact of regret on consumer decision making.

Irons and Hepburn (Reference Irons and Hepburn2007) develop a theoretical economic model to frame regret theory in making choices over different size choice sets. They find that after a particular threshold, a higher number of choices decreases utility for agents that experience regret. Regretful agents are also less likely to engage in a product search when the number of choices is large. On the other hand, non-regretful agents benefit from more choices. Sarver (Reference Sarver2008) incorporates anticipated regret over ex-post consumer dissatisfaction with their choice into a theoretical model of choice over lottery menus. He shows that ex-post regret leads to agents limiting their choices ex-ante. Gourville and Soman (Reference Gourville and Soman2005) suggest that, for some assortment types, a larger variety increases both the regret and difficulty of comparison. They find that variety can backfire and decrease brand market share.

While most economic regret-based models imply ex-post dissatisfaction with one's choice, research in psychology suggests that consumers might experience disutility from forgoing some options even when ex-post satisfied with their initial choice. This disutility can be framed as fear of missing out (or anticipated regret) on forgone alternatives (Milyavskaya et al., Reference Milyavskaya, Saffran, Hope and Koestner2018; Solt, Rixom, and Taylor, Reference Solt, Rixom and Taylor2018; Zeelenberg, Reference Zeelenberg, Oettingen, Sevincer and Gollwitzer2018). In this framing, a larger choice set carries a larger penalty, as more choices are foregone by the consumer, which leads to higher levels of fear of missing out ex-ante and thus a reduced likelihood of any choice.

As the literature suggests, proposed explanations for choice overload can be widely described as the ones that focus on cognitive costs of decision making (Kamenica, Reference Kamenica2008; Kuksov and Villa-Boas, Reference Kuksov and Villas-Boas2010; Malone and Lusk, Reference Malone and Lusk2017, Reference Malone and Lusk2019; Iyengar and Kamenica, Reference Iyengar and Kamenica2010; Chernev, Reference Chernev2003) and the ones that focus on consumer potential regret minimization, or “fear of missing out” (Iyengar and Lepper, Reference Iyengar and Lepper2000; Sarver, Reference Sarver2008; Irons and Hepburn, Reference Irons and Hepburn2007; Milyavskaya et al., Reference Milyavskaya, Saffran, Hope and Koestner2018). Overall, the literature on the impact of search costs and regret on consumer choice behavior highlights its complexity and the need for further investigation. For example, Chernev, Böckenholt, and Goodman (Reference Chernev, Böckenholt and Goodman2015) use a meta-analysis of 99 studies and find that choice set complexity, decision task difficulty, preference uncertainty, and decision goals can all moderate the effects of choice overload. Research also suggests that contextual factors like time constraints can be a leading factor in choice and search behavior (Caplin, Dean, and Martin, Reference Caplin, Dean and Martin2011).

The question of whether search costs or regret are driving the choice overload effects is often hard to address, as information provision can minimize both search costs and the possibility of ex-post regret. For example, the most effective informational treatment for reducing choice overload in Malone and Lusk (Reference Malone and Lusk2017, Reference Malone and Lusk2019) is the provision of expert ratings for available beers. Such ratings, however, both reduce search costs and minimize the potential for feeling regret at the same time (as experts support consumer choice). Arunachalam et al. (Reference Arunachalam, Henneberry, Lusk and Norwood2009) show that in a setting where participants are allowed to voluntarily reduce their choice set size, only a small portion of respondents prefer to choose from a small choice set, suggesting that consumers themselves might either not consider large choice sets to be a problem or be unaware of their own pattern of behavior.

We contribute to the literature by explicitly testing the relative importance of regret and search costs in consumer wine choices. Manipulating the saliency of options outside the actual available choice set induces consumer regret through the feeling of missing out while keeping the actual search costs for the available choice set unchanged. On the other hand, by modifying wine sensory descriptions and food pairings suggested by wine reviewers, we affect consumer search costs but keep anticipated regret over the number of forgone options stable.

In the rest of this section, we discuss existing literature on tasting menu design and wine choice that is useful in the applied design of our experiment.

B. Wine tasting literature and drivers of consumer demand for wine

In wine-tasting rooms, printed sensory descriptions (tasting notes) are commonplace (LaTour, LaTour, and Feinstein, Reference LaTour, LaTour and Feinstein2011). The addition of sensory information has been alternatively shown to both facilitate choice and increase consumer demand, and to complicate choice and confuse consumers. Lockshin and Corsi (Reference Lockshin and Corsi2012) show that printed, elaborate taste descriptions improve the consumer's probability of choosing a wine. Bender (Reference Bender and Allhoff2008) and Ramirez (Reference Ramirez2010) also show that sensory descriptors that consumers relate to have more influence than objective information about grape varieties, geographic denominations, etc. However, other research suggests that sensory descriptions are ambiguous and represent a recreational art with highly individualistic improvisatory elements (Quandt, Reference Quandt2007; Dilworth, Reference Dilworth and Allhoff2008), with consumers likely disregarding sensory descriptors and relying more on product information such as vintage, variety, and region (Thomas et al., Reference Thomas, Gómez, Gerling and Mansfield2014). A recent study (Li, Predic, and Gómez, Reference Li, Predic and Gómez2020) also found that employing objective, rather than sensory, tasting notes are associated with better sales performance in winery tasting rooms. This suggests that expanding sensory descriptions and information about food pairings may both increase and decrease consumer search costs; either way, additional information should affect cognitive search costs.

In terms of tasting list composition, some tasting rooms create a specific wine-tasting menu that only includes wines in a flight or available for tasting, while others include more wines, highlighting the wines that are available to taste. Gómez and Kelley (Reference Gómez and Kelley2013) show that consumer flexibility in the choice of wines tasted and the variety of wines available for tasting have a high impact on winery performance. A study by Shapiro and Gómez (Reference Shapiro and Gómez2014) further suggests that the variety of the wines tasted, number of wines tasted, and other factors are important factors in consumer satisfaction in winery tasting rooms.

A. Experimental design

The lab experiment took place in an experimental economics lab at a large Pacific Northwest university. We recruited participants over 21 years old who were regular wine drinkers and had visited winery tasting rooms before. All participants received a $35 endowment that they could spend in the experiment or take home as a participation payment in its entirety if no purchases were made during the experiment.

Each session of the experiment for this study consisted of a wine-tasting stage, followed by an extended socio-demographic and wine consumption habits survey. The same six wines were presented for tasting across all treatment groups. Participants were randomly assigned to one of the four treatment groups (summarized in Table 1), and each experimental session included at most six participants, all in the same treatment. Once participants arrived in the lab, they read and signed the experimental instructions and the consent forms and were endowed with $35.

Table 1.

Experimental design and treatment group description

The tasting stage started when participants were presented with a tasting sheet on their screen. The participants were told that they could select any, none, or all of the available wines, provided they paid $1 for each wine tasted. Once the tasting selections of all participants were submitted, the experimenter served the participants with the wines they had selected for tasting. Participants were told they should taste and expectorate the wines, but expectoration was not enforced. Participants were provided with two wine glasses (one for red and white wines each), water to rinse the glasses after each tasting, and spittoons for wine waste.

The tasting sheets varied according to one of the four experimental treatment groups presented in Table 1. (1) Tasting list included only the six available wines for immediate tasting during the experiment, and presented only basic objective descriptors of the wine, such as brand, vintage, varietal, and geographical appellation on the tasting list; (2) tasting list included only six wines, but provided both basic descriptors and extended wine sensory descriptions, known as tasting notes, and food pairing recommendations; (3) tasting list included basic descriptions for both the six wines available for immediate tasting and 12 comparable wines available for tasting and purchase in local stores, but not the experiment itself; and (4) tasting list provided both basic and extended wine descriptors for both the six available and 12 comparable wines.

This approach to experimental design allows us to have treatments that would only modify either the saliency of wine options outside the immediate choice set, thus affecting consumer ex-ante regret and fear of missing out, or consumer information availability, affecting search costs and the cognitive load of selecting between wine options with either scant or detailed descriptions. This way, treatment groups (3) and (4) increase outside option saliency relative to groups (1) and (2), respectively; similarly, treatment groups (2) and (4) change the search costs and cognitive load by introducing additional information, compared to treatment groups (1) and (3), respectively.

The long wine-tasting sheet included 18 wines altogether, three of each varietal used in the study: Pinot Noir, Cabernet Sauvignon, and Syrah among red wines, and Pinot Gris, Chardonnay, and Riesling among white wines. Only one wine from each variety was available for immediate tasting. Participants in treatments (1) and (2) were presented with a short tasting of one wine from each of the aforementioned varieties. Having the same wines available in all treatments effectively limits the relevant choice set, while the inclusion of unavailable wines on the list increases the saliency of outside options. If all 18 wines were available for tasting, search costs and the fear of missing out are both likely to increase, making their impact impossible to identify separately.

Participants in all treatments saw winery names, grape varieties, regions, and vintage (basic objective descriptors) for all wines on their sheets; participants in treatments (2) and (4) were additionally provided with sensory description information and food pairing recommendations. Sensory wine descriptions were sourced from the following wine review websites: WineEnthusiast, wine.com, Cellar 503-An Oregon wine club, Portland wine company, and A-Z Wine Works.

An extensive socio-demographic and wine consumption survey was administered to all participants at the end of the experiment. A copy of the survey instrument and an example of the wine-tasting list selection sheet are provided in Appendices B and C.

A. Econometric model specifications

We use a Probit modelFootnote ¹ to understand the probability of tasting any given wine and the impact of increased outside option saliency and extended sensory and food pairing notes. The estimated model is:

(1)$$\eqalign{& Tasted_{ijt} = \alpha _0 + \alpha _1saliency_t + \alpha _2extended\;notes_t + \alpha _3bottle_j + \alpha _lx_{li} + \alpha _kz_{ki} + \epsilon _{ijt} \cr & \qquad\quad Tasted_{ijt} = \left\{{\matrix{ {1\;if\;individual\;i\;in\;treatment\;t\;choose\;to\;taste\;wine\;j\;} \cr {0\;otherwise} \cr } } \right.} $$

Here, subscript i refers to the subject, j to the wine bottle (1–6), and t to the treatment. Variable “saliency” is a dummy equal to 1 when participants see 18 wines on their tasting sheet. Variable “extended notes” is a dummy equal to 1 when basic wine information is supplemented with sensory wine descriptions and food pairing recommendations. Variable “bottle _j” captures wine-specific characteristics for each bottle j. Variable x _li is the demographic attribute l for individual i. Variable z _ki is the wine consumption habit attribute k for individual i. We cluster errors by participant ID as the errors might be correlated within individuals who taste more than one wine.

We check the robustness of our results by comparing different specifications, starting with including no demographic controls and gradually expanding the set.