title: "Serendipity: Chance Encounters in the Marketplace Enhance Consumer Satisfaction"
authors: "Aekyoung Kim, Felipe M. Affonso, Juliano Laran, Kristina M. Durante"
journal: "Journal of Marketing"
year: 2021
volume: 85
issue: 4
pages: "141-157"
doi: "10.1177/00222429211000344"
citation: "Kim, Aekyoung, Felipe M. Affonso, Juliano Laran, and Kristina M. Durante (2021), \"Serendipity: Chance Encounters in the Marketplace Enhance Consumer Satisfaction,\" Journal of Marketing, 85 (4), 141-57."
bibtex: |
@article{kim2021serendipity,
title={Serendipity: Chance Encounters in the Marketplace Enhance Consumer Satisfaction},
author={Kim, Aekyoung and Affonso, Felipe M. and Laran, Juliano and Durante, Kristina M.},
journal={Journal of Marketing},
volume={85},
number={4},
pages={141157},
year={2021},
publisher={SAGE Publications},
doi={10.1177/00222429211000344}
}
> **Disclaimer:** This is a machine-readable conversion of the published paper for use with AI tools. It may contain conversion errors in formatting, tables, or equations. Always verify against the [published version](https://doi.org/10.1177/00222429211000344).
**Abstract**
Despite evidence that consumers appreciate freedom of choice, they also enjoy recommendation systems, subscription services, and marketplace encounters that seemingly occur by chance. This paper proposes that enjoyment can, in some contexts, be higher than that in contexts involving choice. This occurs as a result of feelings of serendipity that arise when a marketplace encounter is positive, unexpected, and attributed to some degree of chance. A series of studies shows that feelings of serendipity positively influence an array of consumer outcomes, including satisfaction and enjoyment, perceptions of meaningfulness of an experience, likelihood of recommending a company, and likelihood of purchasing additional products from the company. The findings show that strategies based on serendipity are even more effective when consumers perceive that randomness played a role in how an encounter occurred, and not effective when the encounter is negative, the encounter occurs deterministically (i.e., planned by marketers to target consumers), and consumers perceive that they have enough knowledge to make their own choices. Altogether, this research suggests that marketers can influence customer satisfaction by structuring marketplace encounters to appear more serendipitous, as opposed to expected or entirely chosen by the consumer.
Keywords: serendipity, satisfaction, enjoyment, choice, subscription services, product recommendations
[\
]
Statement of Intended Contribution
Despite the evidence that they appreciate having freedom of choice, consumers also enjoy recommendation systems, subscription services, and marketplace encounters that seemingly occur by chance. This paper proposes that enjoyment in such contexts is the result of feelings of serendipity that arise when a marketplace encounter is positive, unexpected, and attributed to some degree of chance. This research contributes to the marketing literature by providing evidence for the characteristics and consequences of serendipity, an underutilized construct in marketing theory and practice. This research also contributes by showing that freedom to choose, which consumers many times desire, does not always lead to the highest possible consumer satisfaction. Finally, this research contributes to practice in several industries. Professionals who work in companies that use subscription services (e.g., Birchbox, Stitch Fix), product recommendations (e.g., Amazon, Spotify), and unstructured experiences (e.g., museums, amusement parks) can better understand why and when creating a serendipitous experience can bring more benefits than experiences that are expected or entirely chosen by the consumer.
Consider two consumer encounters. In one encounter, a consumer is listening to a music streaming service and a song they love comes across the speakers. In a second encounter, the consumer is surfing TV channels on Friday night and arrives at channel 131 to find that The Dark Knight, their favorite movie, is about to begin. These encounters are extremely pleasant when they occur in a consumer's life, and may be even more enjoyable than something the consumer personally picks (e.g., choosing a song to listen to or movie to watch). Such occurrences are also increasingly common in the marketplace. For example, subscription product delivery has proliferated in recent years and often involves receiving products (e.g., clothes, wine) periodically without prior knowledge of the items in the shipment. Other occurrences include recommendation systems that make selections for consumers (e.g., songs, videos) and attractions with unstructured experiences that are not previously defined (e.g., seeing works of art in a museum). These encounters may also occur in contexts where the consumer knows what to expect but something unexpected happens, such as a tasting of a consumer's favorite cheese in the grocery store precisely on the day the consumer decided to go shopping.
It is unclear, however, why such marketplace encounters are so enjoyable and whether marketers may be able to create, influence, and enhance these kinds of encounters. We contend that some encounters that do not involve deliberate choice are enjoyable because the way they happen generates feelings of *serendipity*. Serendipity in the marketplace is the set of feelings resulting from a product, service, or experience that is positive, unexpected, and attributed to some degree of chance. We posit that rather than being random encounters, marketers may have control over consumers' perceptions of how the encounter happened (e.g., "I chose it" vs. "There was some chance involved in how it happened."). By altering certain aspects of consumer encounters to generate feelings of serendipity, marketers can influence an array of outcomes, such as satisfaction, enjoyment, meaningfulness of the experience, and willingness to pay, which we collectively call consumer outcomes (see Figure 1). A series of studies in multiple domains (online subscription services, works of art, movies, food consumption, and music) tests the idea that feelings of serendipity positively influence such consumer-relevant outcomes.
Drawing from research on consumer reactions to unexpected events (Heilman et al. 2002), findings on how personal choice may not always generate the highest level of satisfaction (Botti and McGill 2006), and qualitative work in the area of recommendation systems (Kotkov et al. 2016), this research contributes to marketing literature and practice in several ways. First, we provide evidence for the characteristics and consequences of serendipity, an underutilized construct in marketing theory and practice. There is research on how surprising consumers may generate positive or negative reactions, but surprise is only one property of encounters that generate feelings of serendipity. By incorporating the role of chance, we can make different recommendations that are not part of the surprise literature, such as how randomness may increase the enjoyment of experiences. Second, this research shows that freedom to choose, which consumers many times desire, does not always lead to the highest consumer satisfaction. This is because choice involves elaboration and the use of limited cognitive resources already taxed by vast amounts of daily information exposure. We show that the absence of choice may actually increase satisfaction with an encounter, and how different properties of such encounters influence satisfaction. Third, this research contributes to practice in several industries. Professionals who work in companies that use subscription services (e.g., Birchbox, Stitch Fix), product recommendations (e.g., Amazon, Spotify), and unstructured experiences (e.g., museums, amusement parks) can better understand why and when creating serendipitous encounters can bring more benefits than encounters that are expected or entirely chosen by the consumer. Feelings of serendipity are akin to the famous adage of being in the right place at the right time, once popularized by Humphrey Bogart in the classic film *Casablanca*, "Of all the gin joints in all the towns in all the world, she walks into mine." Our conceptualization and findings will provide marketers with insights on how to build some of this magic into marketplace encounters.
**THE PLEASURES OF SERENDIPITY**
The eighteenth-century writer Horace Walpole originally coined serendipity to describe, in a Persian fairy tale, the idea of people "always making discoveries, by accidents and sagacity, of things they were not in quest of" (Walpole 1754, p. 407). Today, serendipity is defined as "finding valuable or agreeable things not sought for" (Merriam-Webster Dictionary 2020), looking for something and finding something else that is actually more suitable to one's needs (Parker 2008), and a positive and unexpected discovery (Herlocker et al. 2004). Despite these definitions, we lack an understanding of serendipity that applies more directly to marketing-relevant phenomena.
Serendipity in the marketplace refers to the set of feelings resulting from an encounter involving a chance finding of a product, service, or experience not directly chosen by the consumer. It happens when a consumer is not looking for anything specific or looking for something and discovers something else (Cunha 2005; McCay-Peet and Toms 2010). Thus, serendipity is an unexpected event that occurs either when the consumer is in a passive state, not trying to discover anything, or an active state, trying to find something of value. As such, we propose that feelings of serendipity in the consumer domain result from an encounter that is (a) positive, (b) unexpected, and (c) involving some degree of chance (Makri and Blandford 2012; Matt et al. 2014; McCay-Peet and Toms 2010).
**The Properties of Serendipity**
Unexpectedness is the cognitive process responsible for the feeling of surprise (Reisenzein, Horstmann, Schützwohl 2019), which can be a positive or negative emotional reaction (Faraji-Rad and Pham 2017; Mellers et al. 1997). An unexpected surprise that is negative can enhance negative reactions, but when a surprise is positive (i.e., brings value to the consumer and generates positive emotions) it enhances satisfaction (Lindgreen and Vanhamme 2003; Westbrook and Oliver 1991). For example, consumers experienced greater enjoyment from winning a smaller, unexpected amount of money compared to a larger, but expected amount of money (Mellers et al. 1997). In addition, surprise incentives (e.g., coupons) are viewed positively and lead to increased spending on unplanned purchases (Heilman, Nakamoto, and Rao 2002; Valenzuela, Strebel, and Mellers 2010).
While serendipitous events are unexpected, surprise is not the only component of serendipity. For an event to generate feelings of serendipity, it must be attributed to some chance or, in the case of chance that leads to positive experiences, luck. This occurs because one consequence of feeling surprised is the search for attribution (Reisenzein et al. 2019; Stiensmeir-Pelster, Martini, and Reisenzein 1995). For example, you may be surprised to receive a free cup of coffee when entering your favorite coffee shop, but then see a sign that says they are giving away coffee to their loyalty club patrons. However, if there is no sign, a consumer may attempt to infer what happened by making an attribution to chance ("They are randomly selecting people to receive a free coffee"), and the search for attribution ends (Feather and Simon 1971; Reisenzein et al. 2019). In this view, while receiving a cup of coffee with a clear attribution (you are a loyalty club member) is nice, perceiving that you were lucky to be selected may be more satisfying.
Thus, when marketers deliver a product, service, or experience in a way that is positive, unexpected, and involving chance, this will generate congruent feelings. Consumers will feel that the encounter was a good surprise, make attributions to chance, and feel lucky that it happened. We collectively call these "feelings of serendipity." These feelings, in turn, can influence an array of consumer-relevant outcomes, such as satisfaction with the entire experience with a company. In addition, feelings of serendipity could make such experiences feel more meaningful, as people attach more meaning to events that they perceive to occur by chance or luck. In fact, ascribing meaning to chance events is most prevalent for positive experiences (King et al. 2006; Krantz 1998; Shanahan and Porfeli 2006). For example, a positive event attributed to chance may lead people to think about ways in which things could have happened less positively (Kray et al. 2010), making them think the event was "meant to be," and that "there is a reason why it occurred." If feelings of serendipity enhance satisfaction and meaning, serendipity should influence other outcomes, such as likelihood of recommending a product or service, willingness to pay, and willingness to buy additional products or services from a company.
**Serendipity in the Marketplace**
Figure 1 presents a summary of the influence of serendipity on consumer-relevant outcomes. Serendipitous events happen without people choosing them or knowledge that they are going to happen. In our earlier example, the person chose to go to the coffee shop that day, but did not choose to get a free cup of coffee. A consumer chooses to listen to a music streaming service, but does not choose the song that plays once they start listening. These examples raise the question of when serendipity occurs and when marketers can take advantage of it.
Serendipity occurs when, at the time of purchase or consumption, an encounter results in the feelings mentioned above (i.e., a good surprise, luck, attributions to chance). In such contexts, the product, service, or experience may be judged as quite positive (Matt et al. 2014). Consider, for example, subscription services (e.g., Stitch Fix). The consumer chooses the company and whether they want to receive a box with clothes (or other products) chosen by the company based on a profile they fill out. While the company has information about the consumer's preferences, it is never perfect information, and the consumer is not making the choice of which products to receive. We propose that a large part of the appeal of such subscription services is that there is an element of surprise and chance (i.e., there is randomness in the process). When the products received are good ("positive"), this generates feelings of serendipity, increasing satisfaction compared to when consumers choose their products.
A similar example is streaming and other recommendation systems, which are the focus of much of the literature on serendipity (Kotkov, Wang, and Veijalainen 2016). These recommendations are based on consumers' preferences, but the recommendations are unexpected and there is a chance component to them. Thus, when the recommendation is good, the entire experience is more enjoyable because the context generates feelings of serendipity (Leong, Vetere, and Howard 2008; Melo and Carvalhais 2013; Zhang et al. 2012). If a consumer places their favorite songs on a playlist and chooses to play one of them, the listening experience will not generate feelings of serendipity, and enjoyment may not be as high. Supporting these predictions, qualitative research on how consumers listen to music suggests that when consumers perceive that they encountered songs and information unexpectedly and by chance, they indicate that their listening experience was better (Leong et al. 2008; Celma 2010). The implication is that consumers appreciate choosing (Brehm 1972; Sharot, De Martino, and Dolan 2009), but serendipitously encountering a product can be more enjoyable as long as the product brings value to the consumer.
As an additional example, consider experiences such as going to a museum. If a consumer knows beforehand which work of art (e.g., a painting) they plan to see, this may be enjoyable but not serendipitous. Alternatively, if the museum places beautiful paintings in locations where consumers may find them by surprise (e.g., immediately upon turning a corner), the experience may become serendipitous and even more enjoyable. Finally, consider sampling at a supermarket. While the consumer has chosen to go grocery shopping, and may even expect that there will be some sampling opportunities, coming across a sampling of a favorite varietal of wine can generate feelings of serendipity, which may lead to heightened enjoyment of the wine and a decision to buy it. The relationship illustrated in these examples, depicted with solid lines in Figure 1, leads to our focal hypotheses:
> H1a: A marketplace encounter that is positive, unexpected, and involves some degree of chance will improve consumer outcomes compared to an encounter that the consumer directly chooses.
>
> H1b: The effect of a marketplace encounter that is positive, unexpected, and involves some degree of chance on consumer outcomes is mediated by feelings of serendipity.
**Attenuating and Enhancing the Effects of Serendipity**
There are different ways in which marketers can attenuate or enhance the effects of serendipity (Figure 1). Marketers can manipulate variables that make the properties of serendipity more or less salient or variables that impact how desirable serendipity is. In terms of the properties of serendipity, the current research manipulates the valence of the encounter (i.e., positive vs. negative; Study 2) and how random consumers perceive the encounter to be (i.e., the encounter was the result of random events vs. planned by the marketer; Study 3). Manipulating valence and perceived randomness allows us to investigate the premise that, for an encounter to be serendipitous, it needs to be a good surprise and attributed to some degree of chance, respectively. In terms of how desirable serendipity is, we manipulate the amount of diagnostic information consumers receive about the product option, which can dampen the serendipity effect (Study 4). We show that sometimes all the properties of serendipity are present, but marketers should be careful to not provide information that can make serendipity undesirable, leading consumers to prefer making their own choice. Altogether, these variables directly address what makes serendipitous encounters so enjoyable the properties must be present and serendipity must be desirable.
First, the encounter must be positive. While surprise can make an experience more positive due to its unexpected nature (Goldsmith and Amir 2010), it cannot make all experiences more positive (Laran and Tsiros 2013). In fact, a surprise can amplify negative affect when the event does not have utility to the consumer or imposes a cost (Kim and Mattila 2010). This means that negative encounters are not serendipitous, even if the element of surprise is present. Thus, we predict that when an unexpected encounter is positive, it will generate feelings of serendipity and improve consumer outcomes compared to when the encounter is chosen by the consumer. When an unexpected encounter is negative, it will not generate feelings of serendipity, and may diminish consumer outcomes compared to when the consumer chooses the encounter. Formally:
> H2: Feelings of serendipity and improved consumer outcomes will occur when an encounter is positive (i.e., brings value to the consumer), but not when it is negative.
Second, influencing the perceived amount of chance involved in an encounter should result in different perceptions of serendipity and outcomes. This can be done in different ways. For example, consider a consumer who receives a song recommendation. In one approach, the song is described as having been randomly selected from a playlist of 100 great songs, and in another situation, the same playlist is described as having 10 songs. Assuming the song is good, the consumer should have more intense feelings of serendipity when it came from the playlist featuring 100 songs, as there was a lower probability that this specific song would have been selected (the consumer feels "luckier"). In a second approach, consumers know that a marketer is responsible for the product, service, or experience (i.e., most marketplace encounters do not occur completely by chance). However, there can still be a surprise and a chance component to the encounter as the marketer does not have perfect information about the consumer's preferences. This implies that the more salient it is that the marketer played a role (e.g., "We chose this carefully to match your preferences"), the less attribution to chance there will be. Thus, we predict that the more consumers perceive that an encounter was the result of randomness (vs. selected deterministically), the more feelings of serendipity it will generate, and the more satisfying it will be. Alternatively, if it is salient that an encounter was planned by a marketer in order to target consumers, the encounter will not generate feelings of serendipity and will not be as satisfying compared to when the presence of the marketer is not salient. We formally hypothesize:
> H3: An increase in the perceived amount of randomness involved in an encounter will increase feelings of serendipity and improve consumer outcomes, whereas the perception that an encounter was selected deterministically will diminish feelings of serendipity and consumer outcomes.
Third, sometimes an encounter successfully generates feelings of serendipity, but the effect of having these feelings on satisfaction is moderated by whether serendipity is desirable or not. Consider services that offer recommendations (e.g., music), which can be successful if their recommendations generate feelings of serendipity. This success may depend on consumers' perception that they have enough knowledge to make their own choices. We predict that when consumers receive a high amount of diagnostic information about a recommendation service and the products it offers, consumers will not desire serendipity, and will be more satisfied when they make their own choices. When information is diagnostic, it is directly relevant to choice, as it can inform which option(s) is (are) superior to other options in a choice set (Feldman and Lynch 1988). When consumers perceive that they have enough relevant information to make an informed choice themselves, feelings of serendipity should not translate into increased satisfaction, as the consumer should believe that they could have made a better choice based on the knowledge they have. This prediction is consistent with research demonstrating that choice can be quite desirable (Brehm 1972; Sharot et al. 2009), especially when consumers have enough information to make a satisfying choice (Botti and McGill 2006). Thus:
> H4: Providing a high amount of diagnostic information to consumers will make serendipity less desirable, diminishing consumer outcomes compared to when consumers make their own choices.
STUDY 1: SERENDIPITY IN ONLINE SUBSCRIPTION SERVICES
Study 1 examines real purchase experiences. We identified four subscription service companies where consumers have the option to choose the products themselves or have the products selected for them. This allowed us to understand the role of serendipity using the natural dichotomy that occurs in subscription services. To do so, we asked participants to describe their experiences with the companies and indicate their satisfaction with the products they purchased, meaningfulness of the consumption experience, willingness to recommend the service, and willingness to extend the subscription. We also measured feelings of serendipity. Consistent with H1a and H1b, we expected that participants who had the products selected for them (vs. chose the products themselves) would be more satisfied with the products, and that feelings of serendipity would drive this effect.
Method
*Participants and Design*. Eight hundred and twenty-nine participants were recruited from Amazon's Mechanical Turk and paid a small monetary compensation. After eliminating 18 outliers based on the overall time spent responding to the survey questions (above 3 SDs from the mean; Meade and Craig 2012), the final sample size was 811 (43.3% men; 18 to 71 years, M = 34.90, SD = 9.96). We used the same exclusion criterion for all studies, but also report the main results without employing exclusions in the Web Appendix. The results of all studies are virtually unchanged without excluding participants. This study had a 2 (condition: personal choice vs. serendipity) × 4 (company replicate) between-subjects design.
*Procedure*. Participants were told that the survey was about their consumption experiences from specific companies. Participation in the study was contingent upon whether the participant indicated that over the last month they had an experience with one of the companies we selected for the study (Birchbox \[www.birchbox.com\], Stitch Fix \[www.stitchfix.com\], The Tie Bar \[www.thetiebar.com\], and FabFitFun \[www.fabfitfun.com\]). Participants were randomly assigned to one of two conditions. Participants in the serendipity condition were told: "Please examine the companies listed below and indicate a company that recently (within the past month) selected products for you and sent them to you as a box you received in the mail." Participants in the personal choice condition were told: "Please examine the companies listed below and indicate a company where you recently (within the past month) selected products from and received your purchase in the mail." In both conditions, if a participant indicated they had not received products from any of the companies listed, they were redirected out of the survey. We programmed the survey with fixed quotas per condition per company used. This allowed us to collect a similar number of participants per condition per company. Once we achieved a certain number of participants for one company (e.g., 200 Birchbox participants with 100 in the serendipity and 100 in the personal choice condition), the survey automatically hid Birchbox from the company selection list.
After indicating a company, participants were told that we were interested in how people process moments of their life and that we would like to know about the recent consumption experience in more detail: "Please think about the time when you recently received a package from \[company name was inserted here\]. Take a minute to remember what it felt like to receive the products and then describe the products and how you felt when you opened the box." After this writing task, participants were asked: "How satisfied are you with the products you received?" (1 = "not at all satisfied," and 7 = "very satisfied"). Participants also responded to four items measuring feelings of serendipity: "I feel that the products I received from the company were a good surprise," "I feel lucky to have come across these products," "I feel that these products were an unexpected discovery," and "I feel that there was some element of chance involved in having received these exact products" (1 = "strongly disagree," and 7 = "strongly agree"). We combined the items to form a serendipity index (α = .83). Participants also responded to questions about the meaningfulness of the experience: "The experience with the products I received was meaningful," "The experience with the products was more meaningful than regular consumption experiences," "I felt that the fact that I got these products was "meant to be", and "These products are meaningful to me" (1 = "strongly disagree," and 7 = "strongly agree"). We combined the items to form a meaningfulness index (α = .91). We also asked "how likely are you to purchase an additional 6-month subscription from \[company name\]?," and "How likely are you to recommend \[company name\] subscription service to a friend?" (1 = "not likely at all," and 7 = "very likely).
Participants then responded to two measures related to expectations: "how high were your expectations about the products before you got them?" (1 = "very low," and 7 = "very high") and "how satisfied did you expect to be with the products before you got them?" (1 = "not at all," and 7 = "very satisfied"). We combined these items to form an expectation index (r = .63). This measure allowed us to rule out the alternative explanation that serendipity leads to positive consumer outcomes due to higher expectations when consumers order their own products. We also measured a series of control items designed to check for the robustness of the effects and test the influence of alternative factors on the results. The items measured product cost, product type, perceived quality, number of products, shipping period, time from the purchase, general attitudes toward the company, and satisfaction with the purchase process. The Web Appendix presents the complete procedure and materials of this (Web Appendix A) and all subsequent studies.
**Results**
*Satisfaction.* A 2 (condition) × 4 (company) ANOVA revealed a main effect of condition, such that participants reported higher satisfaction in the serendipity (M = 6.01, SD = 1.10) than in the personal choice condition (M = 5.55, SD = 1.50; F(1, 803) = 25.58, *p* \< .001). There was no interaction between condition and the company replicates (F(3, 803) = .12, *p* = .946), and the effect of condition was significant for each company. There was a main effect of the company replicate (F(3, 803) = 3.55, *p* = .014). Because this main effect does not change the interpretation of the results, we report additional details, along with the results for each company, in Web Appendix A. In all studies reported in the main text, there were no interactions with the replicates we used (different companies, paintings, videos, and songs), and therefore we collapsed across replicates for all analyses. We present the analysis of replicates of each study in the Web Appendix. We also tested whether the effect would hold when we included each control variable we measured in the analysis for each outcome. None of the covariates changed the results, which indicates that they cannot explain the influence of condition on satisfaction (see Web Appendix A).
*Willingness to Recommend*. A 2 (condition) × 4 (company) ANOVA revealed a main effect of condition, such that participants reported higher willingness to recommend in the serendipity (M = 5.88, SD = 1.29) than in the personal choice condition (M = 5.25, SD = 1.73; F(1, 803) = 34.30, *p* \< .001).
*Willingness to Extend the Subscription*. A 2 (condition) × 4 (company) ANOVA revealed a main effect of condition, such that participants reported higher willingness to extend the subscription in the serendipity (M = 5.45, SD = 1.60) than in the personal choice condition (M = 4.83, SD = 1.89; F(1, 803) = 25.82, *p* \< .001).
*Meaningfulness.* A 2 (condition) × 4 (company) ANOVA revealed a main effect of condition, such that participants reported higher meaningfulness in the serendipity (M = 5.12, SD = 1.34) than in the personal choice condition (M = 4.72, SD = 1.48; F(1, 803) = 16.48, *p* \< .001).[^1]
*Mediation by Feelings of Serendipity.* A 2 (condition) × 4 (company) ANOVA revealed a main effect of condition, such that participants reported greater feelings of serendipity in the serendipity (M = 5.50, SD = 1.02) than in the personal choice condition (M = 5.07, SD = 1.36; F(1, 803) = 25.89, *p* \< .001). We conducted bootstrapping mediation analyses (Process Model 4; Hayes 2018) using condition (serendipity vs. personal choice) as the independent variable and serendipity as the mediator for each of the outcomes.
The indirect effect of serendipity was significant for satisfaction (β = .32, SE = .07, 95% CI: \[.19 to .47\]), meaningfulness (β = .35, SE = .07, 95% CI: \[.21 to .50\]), willingness to recommend (β = .37, SE = .08, 95% CI: \[.22 to .52\]), and willingness to extend the subscription (β = .36, SE = .07, 95% CI: \[.22 to .50\]).
*Expectations.* A 2 (condition) × 4 (company replicate) ANOVA did not reveal main effects of condition (M~serendipity~ = 5.34, SD~serendipity~ = 1.10; M~personalchoice~ = 5.27, SD~personalchoice~ = 1.05; F(1, 803) = .91, *p* = .340), showing that the findings cannot be explained by consumers creating higher expectations when they choose the products they will receive.
**Discussion**
Using retrospective reports from real purchase experiences, Study 1 found that having products sent by a subscription service without knowing what the specific products are leads to more positive consumer responses than personally choosing products. This effect, which supports H1a and H1b, was due to feelings of serendipity, and not by an increase in expectations about the products. Further, the effect held while controlling for an array of factors that may influence consumer responses (i.e., product cost, product type, perceived quality, number of products, shipping period, time from the purchase, general attitudes toward the company, and satisfaction with the purchase process), meaning that the effect of serendipity is robust and goes beyond any possible effect of such factors.
STUDY 2: THE ROLE OF EXPERIENCE VALENCE
We have theorized that the experience needs to be positive for serendipity to occur. In Study 2, we examine the role of valence in the context of seeing a painting in a museum. Participants either chose which painting they wanted to see or had a painting randomly chosen for them. We also had a baseline condition in which participants simply saw a painting. The baseline condition allowed us to determine whether the focal effect occurs because of a positive effect in the serendipity condition or a negative effect in the choice condition. Consistent with H2, we predicted that having a painting randomly chosen would increase enjoyment relative to a baseline and a personal choice condition when the painting was attractive (i.e., positive valence), but not when it was unattractive (i.e., negative valence). This context is relevant to practice as serendipity can also occur when consumers have unstructured experiences such as those that occur in a museum or an amusement park. Thus, findings in this context may help managers configure these experiences in a way that generates feelings of serendipity.
**Method**
*Participants and Design*. Four hundred and sixty-two participants were recruited from Amazon's Mechanical Turk and paid a small monetary compensation. After eliminating fifteen outliers based on the time spent responding to the survey questions (see criterion in Study 1), the final sample size was 447 (46% men; 18 to 89 years, M = 39.33, SD = 13.69). This study had a 3 (condition: baseline, personal choice, serendipity) × 2 (valence: positive vs. negative) × 2 (painting replicate) between-subjects design.
*Procedure*. Participants were told that we were interested in how people respond in different situations. Participants in the personal choice condition were asked to "Imagine you enter an art gallery. Two of the paintings the gallery features appear below." Participants then saw the titles of two paintings by Gerald Chodak: Moving Around and On the Border. The titles were the same in the positive and negative valence condition, but the paintings were either attractive or unattractive, depending on the condition. The order of presentation of the titles on the screen was counterbalanced. Participants were then asked to "select one of these two paintings to view" and clicked on a continue button to proceed. Once participants proceeded to the next page, they saw the painting they chose (see all the paintings in Web Appendix B, as well as the results of a pretest showing that the positive paintings were indeed seen as more positive). Participants in the serendipity condition went through the same procedure, but instead of being asked to select one of the paintings when they saw the information about the paintings, they were simply asked to click on a continue button to proceed. Once they proceeded, they were told "Imagine that you walk down a hallway in the art gallery and turn a corner. Just as you turn the corner, you happen to find this painting on the wall," and one of two paintings, selected randomly, was presented. Participants in the baseline condition were told to "Imagine you enter an art gallery. You will see and rate a painting on the next page." Once participants proceeded to the next page, one of the two paintings was randomly presented to them.
After viewing the painting, participants were asked: "How much did you enjoy the painting?" (0 = "I hated it," and 100 = "I loved it"). Participants also responded to questions about their feelings of serendipity (α = .78): "Getting to see the painting I just saw was a good surprise," "I came across this painting by luck," and "This painting was an unexpected discovery" (1 = "strongly disagree," and 7 = "strongly agree").
We also measured alternative explanations. Participants reported their attachment to the alternative option: "To what extent did you feel attached to the other option?" (1 = "not at all attached," and 7 = "very much attached"). This question examined whether participants enjoyed the chosen option less in the positive valence condition because they were attached to the option they did not choose (Carmon, Wertenbroch, and Zeelenberg 2003). Participants also answered a question about regret: "To what extent did you feel regretful about the painting you saw?" (1 = "not at all regretful," and 7 = "very much regretful"). Moreover, they indicated "How much did you scrutinize the painting?" (1 = "not at all," and 7 = "very much"), allowing us to verify whether participants scrutinized the paintings to different degrees across conditions. Participants then answered questions about stress and frustration: "How stressed were you with the painting selection process?" (1 = "not at all stressful," and 7 = "very much stressful"), and "How frustrated were you with the painting selection process?" (1 = "not at all frustrated," and 7 = "very much frustrated"), allowing us to verify whether choosing versus not choosing, and seeing a negative versus a positive painting, generated negative feelings that could explain the results.
**Results**
*Enjoyment.* A 3 (condition) × 2 (valence) ANOVA revealed a main effect of valence (F(1, 441) = 32.21, *p* \< .001), such that participants enjoyed the positive paintings (M = 63.42, SD = 26.12) more than the negative paintings (M = 47.68, SD = 33.16). There was no main effect of condition (F(2, 441) = .209, *p* = .812). The interaction was significant (F(2, 441) = 3.92, *p* = .021; see Figure 2). When participants saw a positive painting, there was a marginally significant effect of condition (F(2, 441) = 2.83, *p* = .060). Participants reported higher enjoyment in the serendipity (M = 70.08, SD = 23.76) than in the personal choice (M = 59.97, SD = 24.49; F(1, 441) = 4.32, *p* = .038) and baseline conditions (M = 60.17, SD = 28.85; F(1, 441) = 4.17, *p* = .042). There was no difference between the baseline and the personal choice conditions (F \< 1). When participants saw a negative painting, there was no effect of condition (F(2, 441) = 1.35, *p* = .261). We did find that enjoyment was lower in the serendipity (M = 42.88, SD = 32.22) than in the baseline condition (M = 50.90, SD = 31.36; F(1, 441) = 2.59, *p* = .108). While this difference is not statistically significant, the pattern of results suggests that serendipity can be potentially harmful for negative experiences. None of the results for the measured alternative explanations could explain the effects on enjoyment (see Web Appendix B for detailed analyses).
*Mediation by Feelings of Serendipity*. A 3 (condition) × 2 (valence) ANOVA revealed an effect of valence (F(1, 441) = 6.51, *p* = .011), such that participants reported greater feelings of serendipity when the painting was positive (M = 4.53, SD = 1.41) than when it was negative (M = 4.17, SD = 1.55). There was also an effect of condition (F(2, 441) = 5.30, *p* = .005), such that serendipity was higher in the serendipity (M = 4.65, SD = 1.37) than in the baseline condition (M = 4.07, SD = 1.54; F(1, 441 = 10.60, *p* = .001). There was no difference between the serendipity and personal choice (M = 4.34, SD = 1.51) conditions (F(1, 441) = 2.69, *p* = .102). In addition, serendipity was marginally higher in the personal choice than in the baseline condition (F(1, 441) = 2.75, *p* = .098). These effects were qualified by a marginally significant interaction (F(2, 441) = 2.39, *p* = .093). When participants saw a positive painting, there was an effect of condition (F(2, 441) = 7.19, *p* = .001). Serendipity was higher in the serendipity (M = 5.02, SD = 1.33) than in the personal choice (M = 4.42, SD = 1.29; F(1, 441) = 6.31, *p* = .012) and baseline conditions (M = 4.13, SD = 1.48; F(1, 441) = 13.80, *p* \< .001), which did not differ from each other (F(1, 441) = 1.42, *p* = .235). When participants saw a negative painting, there was no effect of condition (F \< 1).
We conducted a bootstrapping analysis for moderated mediation using the three conditions (baseline, personal choice, serendipity) as multicategorical independent variables, valence as the moderator, feelings of serendipity as the mediator, and enjoyment as the dependent variable (Hayes 2018; PROCESS model 8). When participants saw a positive painting, the pathway to enjoyment through feelings of serendipity was significant when comparing the serendipity to the personal choice (β = 8.85, SE = 3.06, 95% CI: \[2.77 to 14.87\]) and baseline (β = 13.05, SE = 3.36, 95% CI: \[6.46 to 19.71\]) conditions. However, when participants saw a negative painting, the pathway to enjoyment through feelings of serendipity was not significant when comparing serendipity both with the personal choice (β = .62, SE = 3.62, 95% CI: \[-7.77 to 6.63\]) and baseline (β = 3.35, SE = 3.56, 95% CI: \[-3.60 to 10.47\]) conditions.
Discussion
Study 2 replicates the positive effect of serendipity, showing that the effect is not only evident in online subscription contexts, but also in experiences such as those that occur when consumers visit a museum (i.e., art consumption). Importantly, this effect only happens when the experience is positive, which supports H2. The effect was due to feelings of serendipity, and could not be explained by feelings of attachment to the alternative option or negative feelings during the painting selection process.
**STUDY 3: DEGREE OF RANDOMNESS IN RECOMMENDATION SYSTEMS**
Study 3 investigates the role of attributions to chance in determining feelings of serendipity and its outcomes (H3). We propose that when it is salient that a marketer carefully planned a product encounter, there will not be an attribution to chance, and feelings of serendipity will not arise. To test this proposition, we simulated a movie trailer recommendation platform ("Movie Trailer Zone") which ostensibly learned about a consumer's movie preferences in order to build a profile. Once this profile was built, the platform recommended a movie trailer fitting the consumer's preferences, helping with the decision about which movie to watch. It was important to investigate serendipity in this context as recommendation services depend heavily on how satisfied consumers are with products that are chosen for them.
We designed the conditions to generate a high versus low attribution to chance by manipulating the degree of randomness in the (a) initial movie selection, by varying whether the movie was selected from a pool of 100 versus 10 curated movie trailers, respectively, and (b) final movie selection, by varying whether the movie selection was described as randomly drawn from the movie pool versus carefully selected by a marketer out of the movie pool, respectively. We predicted that, when the final movie selection was a random process, an increased amount of randomness (making a selection from 100 vs. 10 options) would increase feelings of serendipity and enjoyment. Alternatively, when the final movie selection was described as being made by a marketer, there would be less attribution to chance (i.e., the selection would be deemed determinist), which would attenuate feelings of serendipity and enjoyment.
**Method**
*Participants and Design.* Four hundred participants were recruited from Amazon's Mechanical Turk and paid a small monetary compensation. After eliminating eleven outliers based on the criterion outlined in Study 1, the final sample size was 389 (59.6% men; 18 to 78 years, M = 38.11, SD = 11.56). This study had a 2 (degree of randomness in the initial selection: high vs. low) × 2 (degree of randomness in the final selection: high vs. low) × 5 (movie trailer replicate) between-subjects design. Participants did not make a personal choice in any of the conditions.
*Procedure.* Participants were told that we were interested in consumers' responses to a recently launched platform ("Movie Trailer Zone") that allows members to receive curated movie trailer recommendations, and that they would perform a task that resembled how the platform was used. The procedure was designed to mimic common user experiences in streaming platforms such as Hulu or Netflix. Participants initially saw a short description of the platform, asking them to create an account. In order to increase immersion, we asked participants to provide a username for their profile and proceed to the profile building task. Once participants proceeded, we showed the username they had chosen and a list with 50 movie posters, asking them to choose 5 movies they liked. We indicated that this would help us (i.e., the platform) find trailers for the movies we thought they might like. As in Hulu or Netflix, participants could choose more than five movies if they wanted to. Once participants clicked next, they saw a loading spinner icon so that they would think there was a platform building a customized profile in the background. Participants were told we were examining their preferences in the background, and that this was a necessary step to make a movie trailer recommendation. We asked them a few filler questions and, once they finished, they saw another loading spinner icon with a "Preparing your recommendation..." message. The page advanced after 10 seconds, and the screen showed a recommendation. At this point, we administered different manipulations.
Participants read that based on their profile, we (i.e., the platform) had curated a selection of movie trailers matching their preferences and that we selected one movie trailer out of this curated selection of 100 (randomness in the initial selection: high) or 10 (randomness in the initial selection: low) movie trailers. Here, we also manipulated the perceived degree of randomness in the final selection. Participants in the "high degree of randomness in the final selection" condition read that "We have examined your preferences with the help of our system and have randomly selected one movie trailer that you may enjoy." Participants in the "low degree of randomness in the final selection" condition read that "We have examined your preferences with the help of our system and have carefully selected one movie trailer that you may enjoy." Thus, it was less likely that participants would make attributions to chance in the latter condition, as the instructions made it salient that a marketer planned the experience to target them (i.e., the final selection was deterministic). In addition to the recommendation message, participants saw 60 seconds of the selected movie trailer. Participants saw one of five randomly selected trailers (the movies were Chronicle, Last Stand, Lawless, Priceless, and Wildlife), which we used to ensure the effect was robust across different content. We used these movies based on a pretest that showed a similar baseline level of enjoyment across them (see Web Appendix C).
After viewing the movie trailer, we asked: How much did you enjoy the movie trailer? (0 = "I hated it," and 100 = "I loved it"). We also asked participants whether they wanted to sign up to receive more information about the movie trailer recommendation platform (yes vs. no, and if yes, they had to provide an e-mail). To ensure we could assess actual interest in the platform, we used an online e-mail validation tool to clean invalid or inexistent e-mail addresses. Hence, we had two indicators of interest: answering yes vs. no, and the presence (vs. not) of a valid e-mail.
Participants also responded to items assessing feelings of serendipity (α = .79): "Getting to watch this movie trailer was a good surprise," "The movie trailer was an unexpected discovery," "I came across this movie trailer by luck," and "Based on how the service works, there was a low chance that I would be watching the specific movie trailer that was selected for me." As a manipulation check, we measured: "The movie trailer was selected through a random process" (1 = "strongly disagree," and 7 = "strongly agree"). Finally, we measured the following alternative explanations: regret, stress, frustration, and scrutinizing.
**Results**
*Manipulation Check.* A 2 (initial selection randomness) × 2 (final selection randomness) ANOVA revealed a main effect of final selection randomness (F(1, 385) = 32.51; *p* \< .001), such that participants perceived the movie selection process as more random when the movie was randomly (M = 4.51, SD = 1.85) rather than deterministically selected (M = 3.42, SD = 1.98). With regards to randomness in the initial selection, there was no main effect (F \< 1). This was expected given that there should be no effect of initial selection randomness within the low final selection randomness condition (F \< 1). Thus, the key contrast was the effect of initial selection randomness within the high final selection randomness condition. In this condition, participants perceived the selection process as marginally more random when the degree of randomness in the initial selection was high (M = 4.77, SD = 1.71) than when it was low (M = 4.28, SD = 1.94; F(1, 385) = 3.40, *p* = .066).
*Enjoyment.* A 2 (initial selection randomness) × 2 (final selection randomness) ANOVA on enjoyment revealed a main effect of randomness in the final selection (F(1, 385) = 40.08, *p* \< .001), such that enjoyment was higher when the movie was randomly (M = 73.45, SD = 22.10) rather than deterministically selected (M = 56.63, SD = 30.76). There was no main effect of randomness in the initial selection (F(1, 385) = 2.50, *p* = .114). The interaction was significant (F(1, 385) = 4.05, *p* = .045; see Figure 3). When there was a high degree of randomness in the final selection, enjoyment was greater in the high randomness (M = 78.41, SD = 19.28) than in the low randomness in the initial selection condition (M = 68.77, SD = 23.61; F(1, 385) = 6.80, *p* = .009). When there was a low degree of randomness in the final selection, there was no difference between the high (M = 56.03, SD = 31.61) and low (M = 57.19, SD = 30.11) randomness in the initial selection conditions (F \< 1). None of the results for the measured alternative explanations could explain the pattern of results on enjoyment and the the following dependent variables (see Web Appendix C for detailed analyses).
*Interest in the Platform (Yes vs. No).* A 2 (initial selection randomness) × 2 (final selection randomness) logistic regression revealed a significant interaction between initial and final selection randomness (ß = 1.28, *Wald* = 5.13, *p* = .024), such that participants in the high randomness in the final selection condition were more likely to provide their email for signup when randomness in initial selection was high (P = 34.3%) than when it was low (P = 15.2%; *Wald* = 10.05, *p* = .002). For participants in the low randomness in the final selection condition, there was no difference between the high (P = 11.2%) and low initial randomness conditions (vs. P = 13.5%; *Wald* = .23, *p* = .64).
*Interest in the Platform (Presence of Valid E-Mail).* While a 2 (initial selection randomness) × 2 (final selection randomness) logistic regression did not reveal a significant interaction between initial and final selection randomness (ß = 1.05, *Wald* = 2.59, *p* = .108), the results were consistent with our predictions. Participants in the high randomness in the final selection condition were more likely to provide their email for signup when randomness in initial selection was high (P = 21.2%) than when it was low (P = 11.4%; *Wald* = 3.60, *p* = .058). For participants in the low randomness in the final selection condition, there was no difference between the high (P = 7.9%) and low initial randomness conditions (vs. P = 10.4%; *Wald* = .36, *p* = .548).
*Mediation by Feelings of Serendipity.* A 2 (initial selection randomness) × 2 (final selection randomness) ANOVA revealed a main effect of randomness in the final selection (F(1, 385) = 64.90, *p* \< .001), such that participants reported greater feelings of serendipity when the movie was randomly (M = 4.99, SD = 1.14) rather than deterministically selected (M = 3.89, SD = 1.58). There was also an effect of degree of randomness in the initial selection (F(1, 385) = 4.75, *p* = .030), such that participants reported greater feelings of serendipity when the movie trailer was selected out of a pool of 100 movies (M = 4.63, SD = 1.53) than when it was selected out of a pool of 10 movies (M = 4.31, SD = 1.39). The interaction was significant (F(1, 385) = 4.74, *p* = .030). When there was a high degree of randomness in the final selection, participants reported greater feelings of serendipity in the high randomness (M = 5.30, SD = 1.09) than in the low randomness in the initial selection condition (M = 4.70, SD = 1.11; F(1, 385) = 9.98, *p* = .002). When there was a low degree of randomness in the final selection, there was no difference between the high (M = 3.89, SD = 1.62) and low (M = 3.89, SD = 1.55) initial selection randomness conditions (F \< 1).
We conducted a bootstrapping moderated mediation analysis using the degree of randomness in the initial selection as the independent variable, degree of randomness in the final selection as the moderator, and serendipity as the mediator for each of the outcomes we measured (Process Model 8; Hayes 2018). For the enjoyment measure, the index of moderated mediation was significant (Index = 8.23; 95% CI: \[.69 to 16.08\]). When there was a high degree of randomness in the final selection, the pathway to enjoyment through feelings of serendipity was significant (β = 8.23, SE = 2.17, 95% CI: \[4.07 to 12.57\]). When there was a low degree of randomness in the final selection, the pathway to enjoyment through feelings of serendipity was not significant (β = .00, SE = 3.23, 95% CI: \[-6.34 to 6.33\]). A similar pattern emerged for both measures of interest in the platform (see Web Appendix C).
**Discussion**
Study 3 supports H3, and makes important contributions for the understanding of serendipity. First, feelings of serendipity only occur when the product encounter does not involve highly deterministic components. When it is salient to the consumer that the marketer controlled the ultimate selection of the experience, the experience cannot be attributed to chance, making it less enjoyable. This means that as long as the presence of the marketer (or other nonrandom component) is not made salient, consumers may attribute the selection to chance, increasing serendipity. In addition, increasing the perceived amount of randomness involved in the initial selection of a product experience had a positive effect, which provides theoretical insight about what makes an experience serendipitous and offers marketers another tool to increase serendipity, enjoyment, and interest. In order to provide further theoretical and practical insight, Web Appendix C-1 presents an additional study examining the role of chance in encounters involving serendipity in the food domain.
**STUDY 4: THE ROLE OF INFORMATION**
Study 4 examined information as a moderator of the effect of serendipity on consumer satisfaction. We predicted that feelings of serendipity would not translate to higher satisfaction when consumers are presented with enough diagnostic information that makes them perceive they have the knowledge to make their own choices (H4). To test this prediction, we investigated the context of a recommendation service, similar to Study 2, but now used an existing company that provides a more functional service. We introduced consumers to a service called Brain.fm, which features functional music that can enhance focus. Functional music is used for many specific purposes, including concentration, relaxation, and meditation. We presented consumers with information about what improves a song's ability to increase concentration. In one condition, this information was nondiagnostic to whether a song is functionally effective or not, whereas in the other condition it was diagnostic. In the nondiagnostic information condition, consumers should be more satisfied with a song when the encounter occurred serendipitously than when they made their own choice, and feelings of serendipity should predict satisfaction. Alternatively, in the diagnostic information condition, they should be more satisfied with a song when they made their own choice than when the encounter occurred serendipitously. In this condition, even when feelings of serendipity are high, this should not translate to satisfaction, and satisfaction should be driven by consumers' perceived knowledge to make their own decisions in the product category.
**Method**
*Participants and Design*. Four hundred participants were recruited from Amazon's Mechanical Turk and paid a small monetary compensation. After eliminating seven outliers based on the criterion outlined in Study 1, the final sample size was 393 (55.2% men; 19 to 79 years, M = 40.66, SD = 13.56). This study had a 2 (condition: personal choice vs. serendipity) × 2 (information: nondiagnostic vs. diagnostic) × 5 (song replicate) between-subjects design.
*Procedure.* Participants were told that they would complete a study about Brain.fm a functional music platform backed by scientific research to help listeners focus and concentrate. Then, participants read a section named "What is functional music, anyway?", and that Brain.fm develops music to improve concentration. Participants were also told that functional music involves tempo, pitch, neural phase-locking value, induced brain wave, 3D externalized sound, and brain modulation rate. We included this information so participants could understand whether the information they were about to receive was diagnostic or not.
In the nondiagnostic information condition, participants read about three attributes of functional music that were not relevant to the superiority of one song over others (e.g., "initial composition - humans compose the musical content"). In the diagnostic information condition, participants read about three attributes that were relevant to the superiority of one song over others (e.g., "neural phase-locking value - refers to the extent to which populations of neurons engage in various kinds of coordinated activity"), and saw the range of values that would make a song highly functional. Key to the manipulation, all participants then saw the title, neural phase-locking value, induced brain wave, and brain modulation rate of five songs, the latest three pieces of information being the three attributes participants in the diagnostic condition had just learned about. Thus, each song had three attributes, and participants in the diagnostic (nondiagnostic) condition had just been presented (not presented) with information about which attribute values made a song highly functional.
Participants in the personal choice condition were asked to "choose one of the five songs available for a listening sample," whereas participants in the serendipity condition were told we would randomly select one song for them to listen to on the next page. Once participants proceeded to the next page, the song started playing. We fixed the listening page to auto-advance after 60 seconds, so every participant would listen to the same amount of music. To increase immersion and realism, we used existing functional music and imagery associated with the Brain.fm service throughout the survey.
After participants listened to the song, we asked: "how satisfied are you with the song you just listened to?" and "how satisfied are you with the song listening experience in general?" (1 = "not at all satisfied," and 7 = "very satisfied"), which formed a satisfaction index (r = .85). We also asked about interest in the platform ("how interested would you be in subscribing to Brain.fm's platform?" \[1 = "not at all," and 7 = "very interested"\]) and willingness to recommend ("how likely are you to recommend Brain.fm's subscription service to a friend?" \[1 = "not likely at all," and 7 = "very likely"\]). In addition, we assessed willingness to pay: "Brain.fm has several subscription options, and such as other platforms (e.g., Spotify, Apple Music), the monthly plan costs between \$5 and \$15. How much are you willing to pay for a one-month subscription to Brain.fm?," with a slider scale ranging from 5 to 15.
Participants then responded to a serendipity measure. We told them to consider their experience with Brain.fm and how the platform works, and asked: "Getting to experience this one song I just listened to ended up being a good surprise," "Considering the song selection process, I feel lucky to have come across the song I listened to," "From what it could have been, I feel that the song I listened to was an unexpected discovery," and "I feel that there was some element of chance involved in having experienced this specific song I just listened to" (1 = "strongly disagree," and 7 = "strongly agree"). We combined the items to form a serendipity index (α = .89). We also measured participants' perceived knowledge to make a choice using four-items (e.g., "From the information provided about functional music, I was knowledgeable enough to choose a song to listen to,"), and verified whether perceived knowledge mediated the results when diagnostic information was presented, which it did. These items and their analyses are presented in Web Appendix D. Finally, we measured regret, scrutinizing, and expectations.
Finally, we asked two manipulation check questions: "to which extent did you make your own song choice?" (1 = "not at all," and 7 = "very much") and "how much information was provided about what attributes are necessary for a good functional song?" (1 = "not much," and 7 = "very much").
**Results**
*Manipulation Checks*. Participants in the personal choice condition indicated making their own choice (M = 5.51, SD = 1.45) to a greater extent than those in the serendipity condition (M = 1.70, SD = 1.33; F(1, 389) = 735.08, *p* \< .001). There was no interaction between condition and information (F \< 1). Participants in the diagnostic information condition indicated that information about what attributes are necessary for a good functional song were provided to a greater extent (M = 5.49, SD = 1.37) than in the nondiagnostic information condition (M = 4.20, SD = 1.74; F(1, 389) = 66.34, *p* \< .001).
*Satisfaction*. A 2 (condition) × 2 (information) ANOVA revealed a main effect of information, such that satisfaction was higher when the information was nondiagnostic (M = 5.18, SD = 1.62) rather than diagnostic (M = 4.65, SD = 1.74; F(1, 389) = 10.23, *p* = .001). There was no main effect of condition (F(1, 389) = .028, *p* = .868). The interaction was significant (F(1, 389) = 23.01, *p* \< .001; see Figure 4). When the information was nondiagnostic, satisfaction was greater in the serendipity (M = 5.57, SD = 1.29) than in the personal choice condition (M = 4.80, SD = 1.82; F(1, 389) = 10.64, *p* = .001). When the information was diagnostic, satisfaction was greater in the personal choice (M = 5.07, SD = 1.61) than in the serendipity condition (M = 4.25, SD = 1.77; F(1, 389) = 12.41, *p* \< .001). None of the results for the measured alternative explanations could explain the pattern of results on this and the other dependent variables (see Web Appendix D for detailed analyses).
*Interest in the Platform.* A 2 (condition) × 2 (information) ANOVA on interest did not reveal main effects of condition (F(1, 389) = .028, *p* = .867) or information (F(1, 389) = .110, *p* = .741). The interaction was significant (F(1, 389) = 22.56, *p* \< .001). When the information was nondiagnostic, interest was greater in the serendipity (M = 4.63, SD = 1.80) than in the personal choice condition (M = 3.65, SD = 2.01; F(1, 389) = 12.00, *p* = .001). When the information was diagnostic, interest was greater in the personal choice (M = 4.53, SD = 2.02) than in the serendipity condition (M = 3.61, SD = 2.05; F(1, 389) = 10.58, *p* = .001).
*Willingness to Recommend.* A 2 (condition) × 2 (information) ANOVA on willingness to recommend did not reveal main effects of condition (F(1, 389) = .026, *p* = .872) or information (F(1, 389) = .651, *p* = .420). The interaction was significant (F(1, 389) = 34.37, *p* \< .001). When the information was nondiagnostic, willingness to recommend was greater in the serendipity (M = 4.65, SD = 1.77) than in the personal choice condition (M = 3.59, SD = 1.86; F(1, 389) = 16.13, *p* \< .001). When the information was diagnostic, willingness to recommend was greater in the personal choice (M = 4.53, SD = 1.86) than in the serendipity condition (M = 3.41, SD = 1.87; F(1, 389) = 18.28, *p* \< .001).
*Willingness to Pay.* A 2 (condition) × 2 (information) ANOVA on willingness to recommend did not reveal main effects of condition (F(1, 389) = .001, *p* = .982) or information (F(1, 389) = .002, *p* = .968). The interaction was significant (F(1, 389) = 8.74, *p* = .003). When the information was nondiagnostic, willingness to pay was greater in the serendipity (M = 8.42, SD = 3.45) than in the personal choice condition (M = 7.39, SD = 3.11; F(1, 389) = 4.41, *p* = .036). When the information was diagnostic, willingness to pay was greater in the personal choice (M = 8.42, SD = 3.86) than in the serendipity condition (M = 7.42, SD = 3.86; F(1, 389) = 4.34, *p* = .038).
*Mediation by Feelings of Serendipity.* A 2 (condition) × 2 (information) ANOVA on feelings of serendipity revealed a main effect of condition, such that participants in the serendipity condition (M = 4.67, SD = 1.57) reported greater feelings of serendipity than those in the personal choice condition (M = 3.96, SD = 1.70; F(1, 389) = 19.64, *p* \< .001). The interaction (F(1, 389) = .357, *p* = .550) and main effect of information (F(1, 389) = 2.72, *p* = .100) were not significant.
We conducted a bootstrapping moderated mediation analysis using PROCESS model 15 (Hayes 2018), with the moderator influencing the indirect path post-mediator. We used condition (serendipity vs. personal choice) as the independent variable, information as the moderator, and feelings of serendipity as the mediator for each of the outcomes we measured. For the satisfaction measure, the index of moderated mediation was significant (Index = -.42; 95% CI: \[-.70 to -.20\]). When the information was nondiagnostic, the pathway to satisfaction through feelings of serendipity was positive (β = .49, SE = .12, 95% CI: \[.26 to .75\]). When the information was diagnostic, the pathway to satisfaction through feelings of serendipity was not significant (β = .08, SE = .06, 95% CI: \[-.03 to .20\]). These results suggest that feelings of serendipity drive satisfaction when consumers do not think they have all the knowledge necessary to make a choice themselves. A similar pattern emerged for the interest, willingness to recommend, and willingness to pay measures (see Web Appendix D).
Discussion
Supporting H4, Study 4 demonstrates that when consumers perceive that they have enough knowledge to make their own choices, they are more satisfied with product encounters that they choose than those that occur serendipitously. Feelings of serendipity were still high, but serendipity was simply not as desirable when consumers perceived they had the information they needed to make a choice. These findings are theoretically and managerially important as they show that sometimes consumers may experience feelings of serendipity, which are not negative, but still prefer to make their own choices. In addition, marketers should be careful to not provide too much diagnostic information that can lead consumers to believe that others should not choose for them. In order to provide further theoretical and practical insight, Web Appendix D-1 presents an additional study examining the role of information in encounters involving serendipity.
**GENERAL DISCUSSION**
This paper developed and tested a conceptualization of the role of serendipity in the marketplace. We proposed that feelings of serendipity arise when a consumer encounter is positive, unexpected, and attributed to some degree of chance. The results of four main studies and two supplemental studies support our conceptualization. In multiple domains (online subscription services, works of art, movies, food consumption, and music), the presence of serendipity (Studies 1 to 4, C-1, and D-1) positively influenced satisfaction, enjoyment, meaningfulness, willingness to pay, willing to recommend, and interest. This effect was attenuated when the encounter was negative (Study 2), when a product recommendation was deterministic (i.e., carefully controlled by a marketer; Study 3), and when consumers believed they had enough knowledge to make their own choices (Study 4). In contrast, the effect was enhanced when consumers believed there was a high degree of randomness involved in the selection of the experience, which increased attributions of the experience to chance (Study 3).
**Theoretical Implications**
This research has implications for the literature on serendipity. Some research has examined consumers' appreciation for online recommendations (Ge et al. 2010), but has not provided much evidence for how well recommendations work compared to personal choices and what it is about these recommendations that consumers see positively. We show that feelings of serendipity associated with a recommendation make the consumer experience more positive compared to having a personal choice, and that these feelings can influence a large set of consumer-relevant outcomes. This implies that, instead of simply making recommendations that try to match previous behavior and stated preferences, online recommendation services should design experiences that appear to involve chance, as this will make consumers more satisfied.
The role of chance uncovered in this research informs the literature on surprise in the marketplace. Surprising events can be positive or negative (Calvo and Castillo 2001; Loewenstein 1994) and the current research suggests that research investigating consumer responses to unexpectedness must consider the degree of chance involved. Responses to positive surprises may not be as positive if the consumer is aware that the surprise was carefully planned by a marketer. This implies that the literature on surprise should manipulate or measure the perceived amount of chance that led to something unexpected happening, as this may provide knowledge on why surprises are sometimes so positive (attributed to chance, "this was meant to be") and sometimes not (attribute to specific events, "I know exactly how this happened").
The current work also contributes to the literature on uncertainty and how it influences consumption. Uncertainty and low control are associated with stress (Durante and Laran 2016), and lead consumers to engage in behaviors to regain control (Cutright and Samper 2014; VanBergen and Laran 2016). Research investigating how uncertainty and low control influence choice and consumption should be cautious about the role of chance. Leading consumers to believe that something happened as a result of random events could backfire if the encounter turns out to be negative. The result could lead not only to lower consumer satisfaction, but also magnified negative consequences for a consumer's wellbeing. If an encounter becomes negative (e.g., a movie that leads a consumer to remember traumatic life events, a product the consumer is allergic to), an attribution to chance may generate a strong emotional reaction, which would otherwise be attenuated if the experience was attributed to a specific source. This does not mean that the study of serendipity should be limited because of the possibility that consumers are averse to uncertainty and low control. Instead, serendipity needs to be further explored, and be accompanied by a clear understanding of what brings positive value to the consumer.
Moreover, the current work shows that serendipity occurs when the consumer does not choose a specific product or experience, which means that our findings inform work on consumer preference for choice (Botti and Iyengar 2006; Botti and McGill 2006; Chernev 2003). Prior work shows that people generally like choosing and having more options, and that personal control over choice can increase satisfaction. Alternatively, the current research aligns with the smaller set of evidence on how choice does not always lead to greater satisfaction. This implies that the assumption that choice is preferred over not having choice must be revised to include considerations about how much serendipity is involved when there is no choice. Being able to choose may be preferred, but the absence of choice can lead to greater satisfaction when the consumption context is positive, unexpected, and involves attributions to chance.
Finally, this research has implications for the literature on how the absence of deliberative choice influences well-being (Iyengar and Lepper 1999; Raghunathan and Irwin 2001). Relatively little marketing research has examined how conditions that lack deliberate action influence consumer outcomes. Here, the lack of deliberative choice made products and experiences seem more meaningful. This is important because part of being happy is the feeling that there is meaning in life (Lambert et al. 2013). A lack of meaning is averse, which leads people to immediately engage in meaning restoration (Heine et al. 2006). This means that, instead of focusing on understanding how consumers actively seek meaning, researchers may help consumers by putting more emphasis on events that generate meaning without the need for consumers to actively search for it. Our findings show that marketers can structure several contexts, across industries, to imbue experiences with meaning.
**Practical Implications**
These findings have important implications for marketers and consumers. Consider the different domains in which marketers sell based on recommendations. Marketers may want to emphasize the number of options available and how the encounter with a specific option is a result of chance or randomness. Enhancing perceptions of chance engenders the sense that an experience was "meant to be" given the number of alternative outcomes and increases satisfaction with the recommendation. This strategy is most effective, of course, when the recommendations are positive, and provides an alternative to the view that consumers appreciate knowing that a marketer has specifically tailored an option to them.
In fact, the current research suggests that decreasing the salience that there is a marketer behind the recommendations enhances enjoyment. Thus, marketers should avoid framing an experience with communication suggesting that the firm has "made this selection carefully for you after examining your preferences." This is important because much marketing communication highlights the targeting process by informing consumers that a product was selected for them based on what the company knows about their preferences. The recommendations may be good, but such emphasis decreases the likelihood that consumers will experience feelings of serendipity, as attribution to chance and luck is replaced by an attribution to being watched and targeted by a specific firm. Of course, recommendations are by definition made by the company behind a product, but this fact need not be salient at the time a recommendation is made.
In addition, companies may consider enhancing consumer experiences by providing more opportunities for serendipitous encounters. For example, consumers may enjoy some unexpected events more as part of vacation packages relative to events they personally choose to experience. These events could be partially planned, as when consumers know there will be unexpected activities but do not know when or what it is, or completely unplanned, as when consumers are given free time as part of a travel package but are surprised by an activity that feels serendipitous. While vacation packages mostly involve activities chosen by the travel agency (i.e., the marketer), these are typically previously determined and known by consumers, which our results indicate may not always generate the highest level of enjoyment and satisfaction.
Another valuable insight is that distancing the consumer from the controlled act of choice can systematically enhance and sustain enjoyment over longer periods of time. This insight can inform strategies for promotion tactics such as induced trials via sampling, mailers, and event marketing, whereby marketers can take extra steps to imbue such situations with serendipity. For example, when companies send small product samples to consumers via mail, they typically provide a lot of information about the product, its benefits, and why the consumer is receiving the product. Our findings indicate that providing less information, leaving room for thoughts about how there may have been some chance involved in receiving that exact sample product, may increase enjoyment and the likelihood that the consumer will buy the product.
The idea of providing less information about the recommendation mechanism also has implications for online recommendations. These are typically based on consumers' profiles, preferences, and previous behavior (Lee, Liu, and Lu 2002). There are varying levels of satisfaction with these recommendations, and our findings indicate that this variation can be partially explained by how much diagnostic information consumers have about how the services work, how the selections are made, and the options themselves. Making recommendations is a well-advised strategy as long as the information provided to consumers does not make them believe that they know enough to make their own choices. In these cases, consumers still have feelings of serendipity, but these feelings do not translate to higher satisfaction with the recommendation and the experience as a whole. This means that services that use recommendations can still provide important information to consumers, but must be careful not to provide too much diagnostic information that will decrease the appeal of an experience that the consumer does not choose.
Moreover, the positive effect of serendipity was mitigated when the product had a negative valence. This suggests that for products that may generate negative affect, from solemn movies to more critical experiences such as medical services, attempts to imbue the experience with serendipity would likely result in a stronger negative appraisal of the experience. Negative experiences do not benefit from serendipity, and serendipity can even exacerbate the negativity, as was the case in Study 2.
An interesting question is whether serendipity will translate to increased satisfaction when the product encounter and consumption do not occur at the same time. There are subscription services, such as those for books or wine, where consumers receive a product and only consume it later. Study 1 provides some insight into this question, as consumers who experienced feelings of serendipity showed higher satisfaction up to a month after receiving the products. However, people wear clothes over and over again, which means that eventually the effect of serendipity should fade away. For products that are consumed once, separated from the initial product encounter, we speculate that feelings of serendipity will still have an effect on satisfaction. The surprise may not be present anymore, but attributions to chance, and the feeling that the consumer was lucky, should still have an influence on the enjoyment of the experience. This is an important extension that future research could explore.
The findings also have implications for when consumers buy gifts. Many times the receiver knows they will receive a gift, and sometimes a person receives a gift by surprise. We all know receiving a surprise gift is positive, but our findings show a way to make the experience even more positive for the gift-receiver. The gift-giver could communicate that there were many options to choose from or that one option was selected based on intuition rather than much elaboration. This may generate feelings of serendipity for the receiver of the gift and increase satisfaction. This recommendation is interesting as we tend to believe that signaling effort and planning behind the choice of a gift makes the receiver more satisfied with the gift. However, this belief ignores the possibility that serendipity can also enhance gift-giving, which is an avenue for future research.
**Limitations**
The benefits of serendipity emerged across many categories, but it is possible that the effects would not emerge for certain products and experiences. The effects may be less evident for durable goods that are relatively expensive (cars, appliances), for products or services that require an extended amount of information-gathering before being experienced (surgical procedures, medical treatments), or for products that have a less hedonic orientation than those we investigated. In addition, in the context of ordinary experiences that occur regularly (e.g., an ice cream vendor that visits one's neighborhood nightly), consumers may expect to find them, which decreases the likelihood that consumers will have feelings of serendipity related to these experiences. Future research is needed to examine whether product type or the frequency with which an experience occurs alters the serendipity effect.
Further, it is unclear if the results would hold for products associated with strong, pre-existing brand preferences (e.g., colas). It is possible that the effects are specific to the experience of products or services for which pre-existing preferences or brand loyalty are not strong (Chernev 2003). This is a common phenomenon in consumer decision making research, as strong previous attitudes and preferences may be immune to the effect of marketer-driven manipulations. This is not something we tested in the current investigation, but could be explored in future research.
Despite these limitations, we consistently found that serendipity leads to positive outcomes. Complementing previous research on the positive power of uncontrolled events (Kray et al. 2010; Morewedge et al. 2014), the effects emerged because people perceived such events to involve surprises and chance. Given that little marketing research has examined consumer outcomes when events are not deliberately orchestrated, future work is poised to build on these findings to further consider the varied effects that can emerge when the absence of choice signals the presence of serendipity.
**REFERENCES**
Botti, Simona and Ann L. McGill (2006), "When Choosing Is Not Deciding: The Effect of Perceived Responsibility on Satisfaction," *Journal of Consumer Research*, 33 (2), 21119.
Botti, Simona and Sheena S. Iyengar (2006), "The Dark Side of Choice: When Choice Impairs Social Welfare," *Journal of Public Policy and Marketing*, 25 (1), 2438.
Brehm, Jack W. (1972), *Responses to Loss of Freedom: A Theory of Psychological Reactance*. Morristown, NJ: General Learning Press.
Calvo, Manuel G. and M. Dolores Castillo (2001), "Selective Interpretation in Anxiety: Uncertainty for Threatening Events," *Cognition and Emotion*, 15 (May), 299320.
Carmon, Ziv, Klaus Wertenbroch, and Marcel Zeelenberg (2003), "Option Attachment: When Deliberating Makes Choosing Feel Like Losing," *Journal of Consumer Research*, 30 (1), 1529.
Celma, Oscar (2010), "Music Recommendation," in *Music Recommendation and Discovery: The Long Tail, Long Fail, and Long Play in the Digital Music Space*, Oscar Celma, ed. Berlin, Germany: Springer-Verlag Berlin Heidelberg, 4385.
Chernev, Alexander (2003), "When More Is Less and Less Is More: The Role of Ideal Point Availability and Assortment in Consumer Choice," *Journal of Consumer Research*, 30 (2), 17083.
Cunha, Miguel Pina (2005), "Serendipity: Why Some Organizations Are Luckier than Others," working paper, Universidade Nova de Lisboa, FEUNL Working Paper Series.
Cutright, Keisha M. and Adriana Samper (2014), "Doing It the Hard Way: How Low Control Drives Preferences for High-Effort Products and Services," *Journal of Consumer Research*, 41 (3), 73045.
Durante, Kristina and Juliano Laran (2016), \"The Effect of Stress on Consumer Saving and Spending,\" *Journal of Marketing Research*, 53 (October), 814-28.
Faraji-Rad, Ali, and Michel Tuan Pham (2017), "Uncertainty Increases the Reliance on Affect in Decisions," *Journal of Consumer Research*, 44 (1), 121.
Feather, Norman T. and Jerrold G. Simon. (1971), "Attribution of Responsibility and Valence of Outcome in Relation to Initial Confidence and Success and Failure of Self and Other," *Journal of Personality and Social Psychology*, 18 (2), 17388.
Feldman, Jack M. and John G. Lynch (1988), "Self-Generated Validity and Other Effects of Measurement on Belief, Attitude, Intention, and Behavior," *Journal of Applied Psychology*, 73 (3), 421-35.
Ge, Yong, Hui Xiong, Alexander Tuzhilin, Keli Xiao, Marco Gruteser, and Michael Pazzani (2010), "An Energy-Efficient Mobile Recommender System," in *Proceedings of the 16th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining*, Bharat Rao and Balaji Krishnapuram, eds. New York, NY: Association for Computing Machinery, 899908.
Goldsmith, Kelly and On Amir (2010), "Can Uncertainty Improve Promotions?" *Journal of Marketing Research*, 47 (6), 107077.
Heilman, Carrie M., Kent Nakamoto, and Ambar G. Rao (2002), "Pleasant Surprises: Consumer Response to Unexpected In-Store Coupons," *Journal of Marketing Research*, 39 (2), 24252.
Herlocker, Jonathan L., Joseph A. Konstan, Loren G. Terveen, and John T. Riedl (2004), "Evaluating Collaborative Filtering Recommender Systems," *ACM Transactions on Information Systems (TOIS)*, 22 (1), 553.
Iyengar, Sheena S. and Mark R. Lepper (1999), "Rethinking the Value of Choice: A Cultural Perspective on Intrinsic Motivation," *Journal of Personality and Social Psychology*, 76 (3), 34966.
Kim, Min Gyung and Anna S. Mattila (2010), "The Impact of Mood States and Surprise Cues on Satisfaction," *International Journal of Hospitality Management*, 29 (3), 43236.
King, Laura A., Joshua A. Hicks, Jennifer L. Krull, and Amber K. Del Gaiso (2006), "Positive Affect and the Experience of Meaning in Life,\" *Journal of Personality and Social Psychology*, 90 (1), 17996.
Krantz, David L. (1998), "Taming Chance: Social Science and Everyday Narratives," *Psychological Inquiry* 9 (2), 87-94.
Kray, Laura J., Linda G. George, Katie A. Liljenquist, Adam D. Galinsky, Philip E. Tetlock, and Neal J. Roese (2010), "From What Might Have Been to What Must Have Been: Counterfactual Thinking Creates Meaning," *Journal of Personality and Social Psychology*, 98 (1), 10618.
Hayes, Andrew F. (2018), *An Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach*, 2nd ed. New York, NY: Guilford.
Heine, Steven J., Travis Proulx, and Kathleen D. Vohs (2006), "The Meaning Maintenance Model: On the Coherence of Social Motivations," *Personality and Social Psychology Review*, 10 (2), 88110.
Kotkov, Denis, Shuaiqiang Wang, and Jari Veijalainen (2016), "A Survey of Serendipity in Recommender Systems," *Knowledge-Based Systems*, 111, 18092.
Lambert, Nathaniel M., Tyler F. Stillman, Joshua A. Hicks, Shanmukh Kamble, Roy F. Baumeister, and Frank D. Fincham (2013), "To Belong Is to Matter: Sense of Belonging Enhances Meaning in Life," *Personality and Social Psychology Bulletin*, 39 (11), 141827.
Laran, Juliano and Michael Tsiros (2013), "An Investigation of the Effectiveness of Uncertainty in Marketing Promotions Involving Free Gifts," *Journal of Marketing*, 77 (2), 11223.
Lee, Wei-Po, Chih-Hung Liu, and Cheng-Che Lu (2002), "Intelligent Agent-Based Systems for Personalized Recommendations in Internet Commerce," *Expert Systems with Applications*, 22 (4), 27584.
Leong, Tuck, Frank Vetere, and Steve Howard (2008), "Abdicating Choice: The Rewards of Letting Go," *Digital Creativity*, 19 (4), 23343.
Lindgreen, Adam, and Joëlle Vanhamme (2003), "To Surprise or Not to Surprise Your Customers: The Use of Surprise as a Marketing Tool," *Journal of Customer Behaviour*, 2 (2), 21942.
Loewenstein, George (1994), "The Psychology of Curiosity: A Review and Reinterpretation," *Psychological Bulletin*, 116 (July), 7598.
Meade, Adam W. and S. Bartholomew Craig (2012), "Identifying Careless Responses in Survey Data," *Psychological Methods*, 17 (3), 43755.
Makri, Stephann, and Ann Blandford (2012), "Coming across Information SerendipitouslyPart 1: A Process Model," *Journal of Documentation*, 68 (5), 684705.
Matt, Christian, Alexander Benlian, Thomas Hess, and Christian Weiß (2014), "Escaping from the Filter Bubble? The Effects of Novelty and Serendipity on Users' Evaluations of Online Recommendations," in *Proceedings of the 35th International Conference on Information Systems (ICIS2014)*, Michael D. Myers and Detmar W. Straub, eds. Auckland, New Zealand: Association for Information Systems, 150320.
McCay-Peet, Lori, and Elaine G. Toms (2010), "The Process of Serendipity in Knowledge Work," in *Proceedings of the Third Symposium on Information Interaction in Context*, Nicholas J. Belkin, ed. New York, NY: Association for Computing Machinery, 37782.
Mellers, Barbara A., Alan Schwartz, Katty Ho, and Ilana Ritov (1997), "Decision Affect Theory: Emotional Reactions to the Outcomes of Risky Options," *Psychological Science*, 8 (6), 42329.
Melo, Ricardo, and Miguel Carvalhais (2013), "The Design of Horacle: Inducing Serendipity on the Web," in *xCoAx 2013: Proceedings of the First Conference on Computation, Communication, Aesthetics and X*, Mario Verdicchio and Miguel Carvalhais, eds., Porto, Portugal: Universidade do Porto, 18391.
Merriam-Webster Dictionary (2020), "Serendipity," (accessed May 22, 2020), https://www.merriam-webster.com/dictionary/serendipity.
Morewedge, Carey K., Colleen E. Giblin, and Michael I. Norton (2014), "The (Perceived) Meaning of Spontaneous Thoughts," *Journal of Experimental Psychology. General*, 143 (4), 174254.
Parker, Mina (2008), *Silver Linings: Meditations on Finding Joy and Beauty in Unexpected Places*. Newburyport, MA: Conari Press.
Raghunathan, Rajagopal and Julie R. Irwin (2001), "Walking the Hedonic Product Treadmill: Default Contrast and Mood-Based Assimilation in Judgments of Predicted Happiness with a Target Product," *Journal of Consumer Research*, 28 (3), 35568.
Reisenzein, Rainer, Gernot Horstmann, and Achim Schützwohl (2019), "The Cognitive‐Evolutionary Model of Surprise: A Review of the Evidence," *Topics in Cognitive Science*, 11 (1), 5074.
Rindfleisch, Aric, James E. Burroughs, and Nancy Wong (2009), "The Safety of Objects: Materialism, Existential Insecurity, and Brand Connection," *Journal of Consumer Research*, 36 (1), 116.
Shanahan, Michael J. and Erik J. Porfeli (2006), "Chance Events in the Life Course," *Advances in Life Course Research*, 11, 97119.
Sharot, Tali, Benedetto De Martino, and Raymond J. Dolan (2009), "How Choice Reveals and Shapes Expected Hedonic Outcome," *Journal of Neuroscience*, 29 (12), 376065.
Stiensmeier-Pelster, Joachim, Alice Martini, and Rainer Reisenzein (1995), "The Role of Surprise in the Attribution Process," *Cognition and Emotion*, 9 (1), 531.
Valenzuela, Ana, Barbara Mellers, and Judi Strebel (2010), "Pleasurable Surprises: A Cross-Cultural Study of Consumer Responses to Unexpected Incentives," *Journal of Consumer Research*, 36 (5), 792805.
VanBergen, Noah and Juliano Laran (2016), \"Loss of Control and Self-Regulation: The Role of Childhood Lessons,\" *Journal of Consumer Research*, 43 (December), 534-48.
Walpole, Horace (1754), Letter to Horace Mann, 28 January 1754, in *Horace Walpole's Correspondence: Yale Edition \[Electronic Version\],* Volume 20, p. 407-08, edited by Wilmarth S. Lewis, The Lewis Walpole Library, Yale University, https://images.library.yale.edu/hwcorrespondence/page.asp?vol=20&seq=435.
Westbrook, Robert A., and Richard L. Oliver (1991), "The Dimensionality of Consumption Emotion Patterns and Consumer Satisfaction," *Journal of Consumer Research*, 18 (1), 8491.
Zhang, Yuan Cao, Diarmuid Ó. Séaghdha, Daniele Quercia, and Tamas Jambor (2012), "Auralist: Introducing Serendipity into Music Recommendation," in *Proceedings of the Fifth ACM International Conference on Web Search and Data Mining*, Eytan Adar and Jaime Teevan, eds. New York, NY: Association for Computing Machinery, 1322.
**FIGURE 1: THE ROLE OF SERENDIPITY IN THE MARKETPLACE**
*Conceptual framework diagram with boxes and arrows arranged left to right. On the far left, a box labeled "High Level of Serendipity in an Encounter" connects via a rightward arrow to a central box labeled "Feelings of Serendipity" (this path represents H1b). The central box connects via a rightward arrow to a box on the far right labeled "Consumer Outcomes," which lists seven bullet items: Satisfaction, Enjoyment, Meaningfulness, Willingness to pay, Willingness to recommend, Purchase intentions, and Interest (this path represents H1a). Three moderator boxes with arrows pointing into the main pathway represent boundary conditions: (1) "Positive Versus Negative Encounter" (H2) positioned above and pointing down to the left-to-center path, (2) "Random Versus Deterministic Encounter" (H3) positioned below and pointing up to the left-to-center path, and (3) "Amount of Diagnostic Information Provided" (H4) positioned above and pointing down to the center-to-right path. Solid lines depict the focal mediation relationships (H1a, H1b), while dashed lines depict the moderating relationships (H2, H3, H4).*
*Note: Figure described by AI (Claude Opus) from published image. Values cross-referenced against paper text.*
**FIGURE 2: STUDY 2 RESULTS**
*Bar chart with the y-axis labeled "Enjoyment" (scale 0 to 100) and two clusters of three bars each on the x-axis: "Positive Valence" (left cluster) and "Negative Valence" (right cluster). Within each cluster, three bars represent Baseline (white), Personal choice (hatched gray), and Serendipity (solid gray), as indicated by a legend at the bottom. Error bars show +/- 1 SE on each bar. For the Positive Valence cluster: Baseline = 60.17, Personal Choice = 59.97, Serendipity = 70.08. Asterisk brackets connect the Serendipity bar to both the Personal Choice bar (p = .038) and the Baseline bar (p = .042), indicating significantly higher enjoyment for Serendipity. The Baseline and Personal Choice bars are nearly identical in height and do not differ significantly. For the Negative Valence cluster: Baseline = 50.90, Personal Choice = 48.54, Serendipity = 42.88. No significance brackets appear, consistent with the nonsignificant effect of condition for negative paintings (p = .261). The error bars are visibly larger for the negative paintings than for the positive paintings, reflecting higher variance. The visual pattern clearly shows a crossover: serendipity boosts enjoyment for positive experiences but does not help (and may even hurt) for negative ones. Text means from the Results section confirm: Positive paintings, Serendipity M = 70.08 (SD = 23.76), Personal Choice M = 59.97 (SD = 24.49), Baseline M = 60.17 (SD = 28.85). Negative paintings, Serendipity M = 42.88 (SD = 32.22), Baseline M = 50.90 (SD = 31.36); the Personal Choice negative mean is not reported in the text but appears as 48.54 in the figure.*
*Note: Figure described by AI (Claude Opus) from published image. Values cross-referenced against paper text.*
*Notes*: Error bars = +/− 1 SE. ^†^*p* \< .10. \**p* \< .05. \*\**p* \< .01. Unbracketed comparisons are not significantly different from one another.
**FIGURE 3: STUDY 3 RESULTS**
*Bar chart with the y-axis labeled "Enjoyment" (scale 0 to 100, gridlines at every 10 units starting from 40) and two clusters of two bars each on the x-axis. The left cluster is labeled "High Final Selection Randomness (Random Selection)" and the right cluster is labeled "Low Final Selection Randomness (Careful Selection)." Within each cluster, a white bar represents "Low initial selection randomness (out of 10 movie trailers)" and a gray bar represents "High initial selection randomness (out of 100 movie trailers)," as indicated by a legend at the bottom. Error bars show +/- 1 SE. For the High Final Selection Randomness cluster: Low initial randomness = 68.28, High initial randomness = 78.10. An asterisk bracket connects the two bars, indicating a significant difference (p = .009 in the text). For the Low Final Selection Randomness cluster: Low initial randomness = 57.69, High initial randomness = 56.56. No significance bracket appears, as the two bars are nearly identical and the difference is nonsignificant. The visual pattern shows that increasing both initial and final randomness produces the highest enjoyment, whereas when the final selection is described as carefully chosen by a marketer, the degree of initial randomness does not matter. The figure values (68.28, 78.10, 57.69, 56.56) differ slightly from the means reported in the Results text (68.77, 78.41, 57.19, 56.03), which likely reflects adjusted vs. raw means or rounding in figure production. The interaction is significant (p = .045).*
*Note: Figure described by AI (Claude Opus) from published image. Values cross-referenced against paper text.*
*Notes*: Error bars = +/− 1 SE. ^†^*p* \< .10. \**p* \< .05. \*\**p* \< .01. Unbracketed comparisons are not significantly different from one another.
**FIGURE 4: STUDY 4 RESULTS**
*Bar chart with the y-axis labeled "Satisfaction" in bold (scale 1 to 7, gridlines at each integer) and two clusters of two bars each on the x-axis. The left cluster is labeled "Diagnostic Information" and the right cluster is labeled "Nondiagnostic Information." Within each cluster, a white bar represents "Personal choice" and a gray bar represents "Serendipity," as indicated by a legend at the bottom. Error bars show +/- 1 SE. For the Diagnostic Information cluster: Personal Choice = 5.07, Serendipity = 4.25. An asterisk bracket connects the two bars, indicating a significant difference (p < .001 in the text). The Personal Choice bar is visibly taller, indicating that when consumers had diagnostic information they preferred choosing for themselves. For the Nondiagnostic Information cluster: Personal Choice = 4.80, Serendipity = 5.57. An asterisk bracket connects the two bars, indicating a significant difference (p = .001 in the text). Here the pattern reverses: the Serendipity bar is taller, indicating that when consumers lacked diagnostic information they were more satisfied with serendipitous encounters. The crossover pattern is visually striking, with the bars forming an X-shaped interaction. All four figure values match the means reported in the Results text exactly: Nondiagnostic Serendipity M = 5.57 (SD = 1.29), Nondiagnostic Personal Choice M = 4.80 (SD = 1.82), Diagnostic Personal Choice M = 5.07 (SD = 1.61), Diagnostic Serendipity M = 4.25 (SD = 1.77). The overall interaction was significant (F(1, 389) = 23.01, p < .001).*
*Note: Figure described by AI (Claude Opus) from published image. Values cross-referenced against paper text.*
*Notes*: Error bars = +/− 1 SE. ^†^*p* \< .10. \**p* \< .05. \*\**p* \< .01.
[^1]: As indicated in the Web Appendix procedures, we also measured meaningfulness in Studies 2, 3, and Supplemental Studies C-1 and D-1. This measure follows the same pattern as the main dependent variables, and was more central in a previous version of this paper. For this reason, the results for the subsequent studies are only discussed in the Web Appendix.
# Web Appendix
WEB APPENDIX
**Serendipity: Chance Encounters in the Marketplace Enhance Consumer Satisfaction**
# Table of Contents {#table-of-contents .TOC-Heading}
[Web Appendix A: Additional Methodological Details and Analyses for Study 1 [3](#web-appendix-a-additional-methodological-details-and-analyses-for-study-1)](#web-appendix-a-additional-methodological-details-and-analyses-for-study-1)
[Effects of the Company Replicates and Simple Effects for Each Company Replicate [3](#effects-of-the-company-replicates-and-simple-effects-for-each-company-replicate)](#effects-of-the-company-replicates-and-simple-effects-for-each-company-replicate)
[Analyses of Covariates [4](#analyses-of-covariates)](#analyses-of-covariates)
[Main Results without Exclusion Criterion [6](#main-results-without-exclusion-criterion)](#main-results-without-exclusion-criterion)
[Materials [7](#materials)](#materials)
[Web Appendix B: Additional Methodological Details and Analyses for Study 2 [12](#web-appendix-b-additional-methodological-details-and-analyses-for-study-2)](#web-appendix-b-additional-methodological-details-and-analyses-for-study-2)
[Effects of the Painting Replicates [12](#effects-of-the-painting-replicates)](#effects-of-the-painting-replicates)
[Pretest Positive and Negative Paintings [12](#pretest-positive-and-negative-paintings)](#pretest-positive-and-negative-paintings)
[Meaningfulness Measure Analysis [13](#meaningfulness-measure-analysis)](#meaningfulness-measure-analysis)
[Alternative Explanations [14](#alternative-explanations)](#alternative-explanations)
[Main Results without Exclusion Criterion [15](#main-results-without-exclusion-criterion-1)](#main-results-without-exclusion-criterion-1)
[Materials [16](#materials-1)](#materials-1)
[Web Appendix C: Additional Methodological Details and Analyses for Study 3 [20](#web-appendix-c-additional-methodological-details-and-analyses-for-study-3)](#web-appendix-c-additional-methodological-details-and-analyses-for-study-3)
[Effects of the Movie Trailer Replicates [20](#effects-of-the-movie-trailer-replicates)](#effects-of-the-movie-trailer-replicates)
[Mediation by Feelings of Serendipity Interest in the Platform [20](#mediation-by-feelings-of-serendipity-interest-in-the-platform)](#mediation-by-feelings-of-serendipity-interest-in-the-platform)
[Meaningfulness Measure Analysis [21](#meaningfulness-measure-analysis-1)](#meaningfulness-measure-analysis-1)
[Alternative Explanations [22](#alternative-explanations-1)](#alternative-explanations-1)
[Main Results without Exclusion Criterion [23](#main-results-without-exclusion-criterion-2)](#main-results-without-exclusion-criterion-2)
[Materials [24](#materials-2)](#materials-2)
[Web Appendix C-1: Additional Study on the Role of Chance in Serendipity [32](#web-appendix-c-1-additional-study-on-the-role-of-chance-in-serendipity)](#web-appendix-c-1-additional-study-on-the-role-of-chance-in-serendipity)
[Method [32](#method)](#method)
[Results [35](#results)](#results)
[Discussion [37](#discussion)](#discussion)
[Materials [37](#materials-3)](#materials-3)
[Web Appendix D: Additional Methodological Details and Analyses for Study 4 [43](#web-appendix-d-additional-methodological-details-and-analyses-for-study-4)](#web-appendix-d-additional-methodological-details-and-analyses-for-study-4)
[Analysis of Perceived Knowledge to Make a Choice [43](#analysis-of-perceived-knowledge-to-make-a-choice)](#analysis-of-perceived-knowledge-to-make-a-choice)
[Effects of the Song Replicates [44](#effects-of-the-song-replicates)](#effects-of-the-song-replicates)
[Alternative Explanations [46](#alternative-explanations-2)](#alternative-explanations-2)
[Mediation Analyses Interest in the Platform [46](#mediation-analyses-interest-in-the-platform)](#mediation-analyses-interest-in-the-platform)
[Mediation Analyses Willingness to Recommend [47](#mediation-analyses-willingness-to-recommend)](#mediation-analyses-willingness-to-recommend)
[Mediation Analyses Willingness to Pay [48](#mediation-analyses-willingness-to-pay)](#mediation-analyses-willingness-to-pay)
[Main Results without Exclusion Criterion [48](#main-results-without-exclusion-criterion-3)](#main-results-without-exclusion-criterion-3)
[Materials [49](#materials-4)](#materials-4)
[Web Appendix D-1: Additional Study on the Role of Information [57](#web-appendix-d-1-additional-study-on-the-role-of-information)](#web-appendix-d-1-additional-study-on-the-role-of-information)
[Method [57](#method-1)](#method-1)
[Results [59](#results-1)](#results-1)
[Discussion [64](#discussion-1)](#discussion-1)
[Main Results without Exclusion Criterion [64](#main-results-without-exclusion-criterion-4)](#main-results-without-exclusion-criterion-4)
[Materials [66](#materials-5)](#materials-5)
# Web Appendix A: Additional Methodological Details and Analyses for Study 1
## Effects of the Company Replicates and Simple Effects for Each Company Replicate
In this section, we report details of the replicate interactions and main effects, as well as simple effects of condition within each company replicate. There was no interaction between condition and the company replicates for the satisfaction (F(3, 803) = .12, *p* = .946), meaningfulness (F(3, 803) = .02, *p* = .997), willingness to recommend (F(3, 803) = .19, *p* = .902), and willingness to extend the subscription (F(3, 803) = .18, *p* = .911). We also note that there was no interaction between condition and the company replicates for the feelings of serendipity mediator (F(3, 803) = .76, *p* = .519) and the expectations alternative explanation measure (F(3, 803) = .17, *p* = .916).
Because we did not make any a priori predictions about replicate main effects, we report the main effects, means and SDs (Table WA-1), and means that differ from each other using Sidak correction to control for family-wise error rate. We also report the simple effects per company replicate in Table WA-1. Overall, the results suggest that the effect of condition (serendipity vs. personal choice) on the measured consumer outcomes is consistent across companies. After the table, we report the post-hoc tests with Sidak corrections for replicate main effects.
++
| **Table WA-1: Means, SDs, and Simple Effects of Condition per Company Replicate** |
+:============:+:===============:+:=============:+:======================:+:=============:+:======================:+:==============:+:======================:+:===================:+:======================:+
| **Company** | **Condition** | **Satisfaction** | **Meaningfulness** | **Willingness to Recommend** | **Willingness to Extend Subscription** |
| | +++++++++
| | | **Mean (SD)** | **Simple Main Effect** | **Mean (SD)** | **Simple Main Effect** | **Mean (SD)** | **Simple Main Effect** | **Mean (SD)** | **Simple Main Effect** |
+++++++++++
| Birchbox | Serendipity | 5.94 (.96) | *p* = .013 | 5.10 (1.30) | *p* = .029 | 5.88 (1.26) | *p* \< .001 | 5.44 (1.52) | *p* = .020 |
| +++ ++ ++ ++ |
| | Personal Choice | 5.48 (1.51) | | 4.67 (1.44) | | 5.12 (1.71) | | 4.87 (1.88) | |
+++++++++++
| Stitchfix | Serendipity | 5.83 (1.37) | *p* = .003 | 4.90 (1.49) | *p* = .053 | 5.80 (1.45) | *p* = .006 | 5.22 (1.74) | *p* = .025 |
| +++ ++ ++ ++ |
| | Personal Choice | 5.29 (1.60) | | 4.51 (1.54) | | 5.21 (1.76) | | 4.66 (2.02) | |
+++++++++++
| The Tie Bar | Serendipity | 6.09 (1.08) | *p* = .038 | 5.32 (1.31) | *p* = .041 | 5.73 (1.34) | *p* = .011 | 5.46 (1.65) | *p* = .016 |
| +++ ++ ++ ++ |
| | Personal Choice | 5.71 (1.53) | | 4.92 (1.41) | | 5.19 (1.65) | | 4.86 (1.70) | |
+++++++++++
| FabFitFun | Serendipity | 6.17 (.92) | *p* = .010 | 5.17 (1.23) | *p* = .052 | 6.09 (1.06) | *p* = .005 | 5.69 (1.49) | *p* = .001 |
| +++ ++ ++ ++ |
| | Personal Choice | 5.70 (1.31) | | 4.79 (1.52) | | 5.49 (1.79) | | 4.91 (1.96) | |
+++++++++++
For the satisfaction measure, there was a main effect of the company replicate (F(3, 803) = 3.55, *p* = .014). Post-hoc tests with Sidak correction revealed that people were less satisfied with Stitchfix (M = 5.56, SD = 1.51) than with The Tie Bar (M = 5.90, SD = 1.34; *p* = .061) and FabFitFun (M = 5.93, SD = 1.16; *p* = .028). For the meaningfulness measure, there was a main effect of the company replicate (F(3, 803) = 3.09, *p* = .026). Post-hoc tests with Sidak correction revealed that people reported less meaningfulness from the Stitchfix consumption experience (M = 4.70, SD = 1.52) than The Tie Bar (M = 5.12, SD = 1.37; *p* = .021).
There was no main effect of the company replicate for willingness to recommend (F(3, 803) = 1.99, *p* = .114) and willingness to extend the subscription (F(3, 803) = 1.48, *p* = .218) measures. We also note that there was no main effect of the company replicate for the feelings of serendipity mediator (F(3, 803) = 2.53, *p* = .056) and the expectations alternative explanation measure (F(3, 803) = .75, *p* = .523).
## Analyses of Covariates
In Study 1, participants responded to a series of control items designed to test the influence of alternative factors on the results. None of the covariates changed the results. Table WA-2 reports details for the covariates. Tables WA-3 reports the analyses of the dependent variables with covariates.
++
| **TABLE WA-2 DETAILS OF COVARIATES** |
+:=====================:+:===============:+:===================:+:================:+:===================:+:===================:+:======================:+:=============================:+:===================================:+
| **Company** | **Condition** | **Product Cost** | **Product Type** | **Product Quality** | **No. of Products** | **Time from Purchase** | **Attitudes Towards Company** | **Satisfaction w/Purchase Process** |
| | | | | | | | | |
| | | **M (SD)** | **M (SD)** | **M (SD)** | **M (SD)** | **M (SD)** | **M (SD)** | **M (SD)** |
++++++++++
| Birchbox | Serendipity | \$29.13 (\$33.14) | 6.16 (2.16) | 7.19 (1.33) | 5.82 (2.30) | 24.84 (39.64) | 7.78 (1.25) | 5.76 |
| | | | | | | | | |
| | | | | | | | | (1.13) |
| +++++++++
| | Personal choice | \$33.26 (\$49.27) | 5.95 (2.08) | 6.82 (1.80) | 5.27 (1.65) | 34.67 (61.72) | 7.26 (1.63) | 5.53 |
| | | | | | | | | |
| | | | | | | | | (1.40) |
++++++++++
| Stitchfix | Serendipity | \$108.16 (\$88.69) | 5.89 (1.85) | 7.01 (1.37) | 4.18 (1.62) | 27.71 (44.26) | 7.59 (1.46) | 5.59 |
| | | | | | | | | |
| | | | | | | | | (1.21) |
| +++++++++
| | Personal choice | \$103.16 (\$90.63) | 5.87 (1.83) | 6.78 (1.68) | 4.12 (1.58) | 24.52 (26.31) | 7.19 (1.75) | 5.15 |
| | | | | | | | | |
| | | | | | | | | (1.44) |
++++++++++
| The Tie Bar | Serendipity | \$118.59 (\$134.41) | 5.50 (2.24) | 7.55 (1.25) | 3.94 (2.11) | 32.33 (36.92) | 7.82 (1.34) | 5.68 |
| | | | | | | | | |
| | | | | | | | | (1.21) |
| +++++++++
| | Personal choice | \$92.57 (\$87.43) | 5.77 (5.50) | 7.10 (1.61) | 4.32 (5.72) | 29.31 (32.21) | 7.50 (1.50) | 5.52 |
| | | | | | | | | |
| | | | | | | | | (1.21) |
++++++++++
| FabFitFun | Serendipity | \$51.81 (\$37.95) | 6.19 (1.94) | 7.20 (1.52) | 6.98 (3.47) | 34.62 (43.08) | 7.85 (1.42) | 5.74 |
| | | | | | | | | |
| | | | | | | | | (1.10) |
| +++++++++
| | Personal choice | \$55.94 (\$57.25) | 6.17 (2.00) | 6.90 (1.82) | 7.40 (4.83) | 32.51 (29.80) | 7.52 (1.59) | 5.49 |
| | | | | | | | | |
| | | | | | | | | (1.39) |
++++++++++
| Total | Serendipity | \$76.26 (\$91.39) | 6.00 (2.01) | 7.24 (1.38) | 5.25 (2.76) | 29.82 (41.12) | 7.76 (1.37) | 5.70 |
| | | | | | | | | |
| | | | | | | | | (1.16) |
| +++++++++
| | Personal choice | \$71.31 (\$78.31) | 5.88 (2.05) | 6.90 (1.73) | 5.30 (4.14) | 30.25 (39.87) | 7.37 (1.62) | 5.42 |
| | | | | | | | | |
| | | | | | | | | (1.37) |
++++++++++
| Condition Main Effect | | *p* = .304 | *p* = .370 | *p* = .002 | *p* = .837 | *p* = .894 | *p* \< .001 | *p* = .002 |
++++++++++
| Replicate Main Effect | | *p* \< .001 | *p* = .042 | *p* = .044 | *p* \< .001 | *p* = .316 | *p* = .170 | *p* = .116 |
++++++++++
| Interaction Effect | | *p* = .179 | *p* = .896 | *p* = .899 | *p* = .409 | *p* = .296 | *p* = .902 | *p* = .715 |
++++++++++
Table WA-3 lists the effects of the 2 (condition) × 4 (company) ANOVAs on each of the dependent variables when each of the covariates is included in the model. As can be seen from the condition p-values, the results remain unchanged even when the covariates are accounted for. In addition, the covariates do not change the lack of condition by company interaction, suggesting that the pattern remains the same for all companies investigated in the study.
++
| **TABLE WA-3 ANALYSES INCLUDING COVARIATES** |
+:==================================:+:================:+:================:+:================:+:===================:+:===================:+:======================:+:=============================:+:===================================:+
| **DV** | **Effect on DV** | **Product Cost** | **Product Type** | **Product Quality** | **No. of Products** | **Time from Purchase** | **Attitudes Towards Company** | **Satisfaction w/Purchase Process** |
++++++++++
| Satisfaction | Condition | *p* \< .001 | *p* \< .001 | *p* \< .001 | *p* \< .001 | *p* \< .001 | *p* = .001 | *p* \< .001 |
| +++++++++
| | Replicate | *p =* .003 | *p =* .014 | *p =* .058 | *p =* .022 | *p =* .012 | *p =* .132 | *p =* .022 |
| +++++++++
| | Interaction | *p =* .900 | *p =* .932 | *p =* .475 | *p =* .942 | *p =* .944 | *p =* .870 | *p =* .988 |
++++++++++
| Meaningfulness | Condition | *p* \< .001 | *p* \< .001 | *p* \< .001 | *p* \< .001 | *p* \< .001 | *p* = .017 | *p* = .007 |
| +++++++++
| | Replicate | *p =* .011 | *p =* .020 | *p =* .362 | *p =* .029 | *p =* .020 | *p =* .132 | *p =* .056 |
| +++++++++
| | Interaction | *p =* .989 | *p =* .999 | *p =* .966 | *p =* .998 | *p =* 1.00 | *p =* .987 | *p =* .739 |
++++++++++
| Willingness to Recommend | Condition | *p* \< .001 | *p* \< .001 | *p* \< .001 | *p* \< .001 | *p* \< .001 | *p* \< .001 | *p* \< .001 |
| +++++++++
| | Replicate | *p =* .017 | *p =* .182 | *p =* .001 | *p =* .236 | *p =* .096 | *p =* .040 | *p =* .006 |
| +++++++++
| | Interaction | *p =* .816 | *p =* .913 | *p =* .761 | *p =* .916 | *p =* .937 | *p =* .974 | *p =* .501 |
++++++++++
| Willingness to Extend Subscription | Condition | *p* \< .001 | *p* \< .001 | *p* \< .001 | *p* \< .001 | *p* \< .001 | *p* \< .001 | *p* \< .001 |
| +++++++++
| | Replicate | *p =* .067 | *p =* .273 | *p =* .157 | *p =* .301 | *p =* .141 | *p =* .632 | *p =* .579 |
| +++++++++
| | Interaction | *p =* .868 | *p =* .895 | *p =* .795 | *p =* .907 | *p =* .877 | *p =* .684 | *p =* .560 |
++++++++++
We also note that the effect of condition remains significant when we include all covariates at the same time in the model for the satisfaction (*p* = .001), meaningfulness (*p* = .030), willingness to recommend (*p* \< .001), and willingness to extend subscription (*p* = .001). Finally, the indirect effect of serendipity in the mediation analysis remained significant even when we included all covariates in the model for satisfaction (β = .09, SE = .03, 95% CI: \[.04 to .16\]), meaningfulness (β = .12, SE = .04, 95% CI: \[.05 to .20\]), willingness to recommend (β = .10, SE = .03, 95% CI: \[.04 to .16\]), and willingness to extend subscription (β = .09, SE = .03, 95% CI: \[.03 to .15\]). These results provide evidence for the unique influence of serendipity, which occurs even when important factors such as cost and quality are accounted for.
## Main Results without Exclusion Criterion
The Table below presents all the key analyses from the main text, with and without the exclusion criterion. When conducting such analyses, we also checked the pattern of results (descriptive statistics), which did not change.
++
| **Study 1** |
+:============================================:+:=========================================:+:=========================================:+:=================:+
| **Analysis** | **Excluding \>3SD P's** | **Not Excluding \>3SD P's** | **Any change?** |
| | | | |
| | **(N = 811)** | **(N = 829)** | |
+++++
| Satisfaction | F(1, 803) = 25.58, *p* \< .001 | F(1, 821) = 25.83, *p* \< .001 | No |
+++++
| Willingness to Recommend | F(1, 803) = 34.30, *p* \< .001 | F(1, 821) = 34.66, *p* \< .001 | No |
+++++
| Willingness to Extend Subscription | F(1, 803) = 25.82, *p* \< .001 | F(1, 821) = 25.86, *p* \< .001 | No |
+++++
| Meaningfulness | F(1, 803) = 16.48, *p* \< .001 | F(1, 821) = 16.72, *p* \< .001 | No |
+++++
| Feelings of Serendipity | F(1, 803) = 25.89, *p* \< .001 | F(1, 821) = 25.51, *p* \< .001 | No |
+++++
| Satisfaction Mediation | β = .32, SE = .07, 95% CI: \[.19 to .47\] | β = .32, SE = .07, 95% CI: \[.19 to .45\] | No |
+++++
| Willingness to Recommend Mediation | β = .35, SE = .07, 95% CI: \[.21 to .50\] | β = .36, SE = .08, 95% CI: \[.21 to .51\] | No |
+++++
| Willingness to Extend Subscription Mediation | β = .37, SE = .08, 95% CI: \[.22 to .52\] | β = .35, SE = .07, 95% CI: \[.21 to .50\] | No |
+++++
| Meaningfulness Mediation | β = .36, SE = .07, 95% CI: \[.22 to .50\] | β = .35, SE = .07, 95% CI: \[.21 to .49\] | No |
+++++
## Materials
Design: 2 (personal choice vs. serendipity) × 4 (company replicate) between-subjects design.
**\[Serendipity Condition\]**
This survey is about your consumption experiences from specific companies.
Please examine the companies listed below and indicate a company that recently (within the past month) selected products for you and sent them to you as a box you received in the mail.
If you have used multiple companies, indicate the store which delivered the most recent package to you.
Birchbox
Stitchfix
The Tie Bar
FabFitFun
I did not receive a box at any of these stores over the past month.
Hello and welcome.
We are interested in how people process moments of their life.
When you are ready to begin this task, please click below.
Earlier, you indicated that \[company name piped text\] recently selected products for you and you received your purchase in the mail. We would like to know about this consumption experience more in detail.
Please think about the time when you recently received a package from \[company name piped text\]. Take a minute to remember what it felt like to receive the products and then describe the products and how you felt when you opened the box.
You will be able to proceed after 30 seconds, but you can write as much as you want.
\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_
\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_
\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_
**\[Personal Choice Condition\]**
This survey is about your consumption experiences from specific companies.
Please examine the companies listed below and indicate a company where you recently (within the past month) selected products from and received your purchase in the mail.
If you have used multiple companies, indicate the store which delivered the most recent package to you.
Birchbox
Stitchfix
The Tie Bar
FabFitFun
I did not select products at any of these stores over the past month.
Hello and welcome.
We are interested in how people process moments of their life.
When you are ready to begin this task, please click below.
Earlier, you indicated that you recently selected products from \[company name piped text\] and received your purchase in the mail. We would like to know about this consumption experience more in detail.
Please think about the time when you recently received a package from \[company name piped text\]. Take a minute to remember what it felt like to receive the products and then describe the products and how you felt when you opened the box.
You will be able to proceed after 30 seconds, but you can write as much as you want.
\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_
\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_
\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_
How satisfied were you with the products you received? (1 = "not at all," and 7 = "very satisfied")
Now, we would like to ask you a few questions related to your consumption experience with \[company name piped text\]. Considering this consumption experience, please answer the questions below.
Please indicate the extent to which you agree with the statements below.
I feel that the products I received from the company were a good surprise. (1 = "strongly disagree," and 7 = "strongly agree")
I feel lucky to have come across these products. (1 = "strongly disagree," and 7 = "strongly agree")
I feel that these products were an unexpected discovery. (1 = "strongly disagree," and 7 = "strongly agree")
I feel that there was some element of chance involved in having received these exact products. (1 = "strongly disagree," and 7 = "strongly agree")
Now, we have a few more questions related to your consumption experience with \[company name piped text\]. Considering this consumption experience, please answer the questions below.
The experience with the products I received was meaningful. (1 = "strongly disagree," and 7 = "strongly agree")
The experience with the products was more meaningful than regular consumption experiences. (1 = "strongly disagree," and 7 = "strongly agree")
I felt that the fact that I got these products was \"meant to be.\" (1 = "strongly disagree," and 7 = "strongly agree")
These products are meaningful to me (1 = "strongly disagree," and 7 = "strongly agree")
How likely are you to purchase an additional 6-month subscription from \[company name piped text\]? (1 = "not likely at all," and 7 = "very likely")
How likely are you to recommend \[company name piped text\] subscription service to a friend? (1 = "not likely at all," and 7 = "very likely")
How high were your expectations about the products before you got them? (1 = "very low," and 7 = "very high")
How satisfied did you expect to be with the products before you got it them? (1 = "not at all," and 7 = "very satisfied")
How much did you pay for the product? \$ \_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_
What are your perceptions of the product? This product is ...
(1 = "utilitarian (i.e., for functional benefits)," and 9 = "hedonic (i.e., for pleasure)")
(1 = "low quality," and 9 = "high quality")
How many products did you get in your recent package? \_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_
How long did it take for your package to arrive? (1 = "not much (quick delivery)," and 7 = "A lot (slow delivery")
How long ago you did you buy the product? \_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_ days ago
In general, what are your perceptions of the company where you bought the product?\
\
the company is . (1 = "very negative," and 9 = "very positive")
How satisfied are you with the purchase process? (1 = "not at all satisfied," and 7 = "very satisfied")
# Web Appendix B: Additional Methodological Details and Analyses for Study 2
## Effects of the Painting Replicates
*Enjoyment.* A 3 (condition) × 2 (valence) × 2 (painting replicate) ANOVA did not reveal a three-way interaction (F(2, 435) = .563, *p* = .570) nor two-way interactions between the painting replicate and condition (F(2, 435) = .183, *p* = .833) or valence (F(1, 435) = .483, *p* = .487). There was a main effect of painting replicate (F(1, 435) = 11.90, *p* = .001), such that participants enjoyed the set of paintings titled "On the Border" (M = 60.66, SD = 27.37) more than the set of paintings titled "Moving Around" (M = 50.85, SD = 33.09). Because this main effect does not change the interpretation of the results, we collapsed across painting replicates for the enjoyment analyses.
*Feelings of Serendipity.* A 3 (condition) × 2 (valence) × 2 (painting replicate) ANOVA did not reveal a three-way interaction (F(2, 435) = .642, *p* = . 527), nor two-way interactions between the painting replicate and condition (F(2, 435) = .200, *p* = .819) or valence (F(1, 435) = .088, *p* = .766), nor a main effect of painting replicate (F(1, 435) = 2.35, *p* = .126). Thus, we collapsed across replicates for the subsequent analyses.
## Pretest Positive and Negative Paintings
We conducted a pretest to ensure that the negative paintings were indeed seen as more negative. A separate sample (N = 119) was randomly assigned to seeing one of the four paintings described in Study 2's procedure. After seeing the paintings, participants responded to three items using 7-point scales: "What are your perceptions of the painting?" (1 = "unattractive," and 7 = "attractive"; 1 = "unappealing," and 7 = "appealing"; and 1 = "negative", and 7 = "positive") (α = .96). A 2 (valence) × 2 (painting replicate) ANOVA only revealed a main effect of valence (F(1, 115) = 23.28, *p* \< .001), such that participants perceived seeing the positive paintings (M = 5.44, SD = 1.41) as more positive than the negative paintings (M = 3.86, SD = 2.08). There was no painting replicate by valence interaction or a replicate main effect (all F's \< 1).
## Meaningfulness Measure Analysis
A 3 (condition) × 2 (valence) × 2 (painting replicate) ANOVA on meaningfulness did not reveal a three-way interaction (F(2, 435) = 1.09, *p* = .337) nor two-way interactions between the painting replicate and condition (F(2, 435) = .116, *p* = .890) or valence (F(1, 435) = 2.11, *p* = .147). There was a marginally significant effect of painting replicate (F(1, 435) = 3.03, *p* = .082), consistent with the results observed for the enjoyment measure, such that participants perceived the set of paintings titled "On the Border" (M = 4.07, SD = 1.80) as more meaningful than the set of paintings titled "Moving Around" (M = 3.78, SD = 1.90). We collapsed across painting replicates for the meaningfulness analysis.
A 3 (condition) × 2 (valence) ANOVA revealed an effect of valence (F(1, 441) = 5.84, p = .016), such that participants perceived seing a positive painting as more meaningful (M = 4.13, SD = 1.77) than a negative painting (M = 3.71, SD = 1.93). The effect of condition was not significant (F(2, 441) = 2.30, *p* = .101). There was a marginally significant condition by valence interaction (F(2, 441) = 2.78, p = .063). When participants saw a positive painting, there was an effect of condition (F(2, 441) = 5.14, *p* = .006). Participants reported greater perceptions of meaningfulness in the serendipity (M = 4.60, SD = 1.70) than in the baseline condition (M = 3.64, SD = 1.71; F(1, 441) = 10.27, *p* = .001). The personal choice condition (M~ ~= 4.14, SD = 1.77) did not differ from the serendipity (F(1, 441) = 2.39, *p *= .123) and the baseline (F(1, 441) = 2.72, *p *= .100) conditions. When participants saw a negative painting, there was no effect of condition (F \< 1).
We conducted a bootstrapping analysis for moderated mediation using the three conditions (baseline, personal choice, serendipity) as multicategorical independent variables, valence as the moderator, feelings of serendipity as the mediator, and meaningfulness as the dependent variable (Hayes 2018; PROCESS model 8). When participants saw a positive painting, the pathway to meaningfulness through feelings of serendipity was significant when comparing the serendipity to the personal choice (β = .53, SE = .19, 95% CI: \[.15 to .89\]) and baseline (β = .78, SE = .21, 95% CI: \[.38 to 1.19\]) conditions. When participants saw a negative painting, the pathway to meaningfulness through feelings of serendipity was not significant when comparing serendipity both with the personal choice (β = .04, SE = .22, 95% CI: \[-.46 to .41\]) and baseline (β = .20, SE = .21, 95% CI: \[-.22 to .63\]) conditions.
## Alternative Explanations
We collapsed across replicates to analyze alternative explanations. There was only a significant main effect of valence on regret (F(2, 441) = 13.74, *p* \< .001), such that participant felt more regretful about the negative paintings (M = 3.70, SD = 2.13) than the positive paintings (M = 2.96, SD = 2.14). In addition, there was a marginally significant main effect of condition on attachment to the alternative option (F(2, 441) = 2.95, *p* = .053), such that participants in the serendipity condition felt greater attachment to the alternative option (M = 3.46, SD = 2.11) compared to the baseline condition (M = 2.94, SD = 1.99; F(1, 441) = 4.72, *p* = .030). There was no difference between the personal choice and the serendipity (F \< 1) conditions. Participants in the personal choice condition felt marginally greater attachment (M = 3.40, SD = 1.95) to the alternative option compared to baseline conditions (F(1, 441) = 3.22, *p* = .074). There was no significant main effects nor an interaction on scrutinizing (*p*'s \> .11). Finally, we combined the questions about stress and frustration to form a negative-feelings index (r = .84). There were no significant main effects nor an interaction on negative feelings (*p*'s \> .14). Thus, while some differences were found, none of them can explain the pattern of results on enjoyment.
## Main Results without Exclusion Criterion
The Table below presents all the key analyses form the main text, with and without the exclusion criterion.
++
| **Study 2** |
+:========================:+:=====================================================================================================================================================================================================================================:+:=====================================================================================================================================================================================================================================:+:=========================================================================================================================================:+
| **Analysis** | **Excluding \>3SD P's** | **Not Excluding \>3SD P's** | **Any change?** |
| | | | |
| | **(N = 447)** | **(N = 462)** | |
+++++
| Enjoyment | Interaction was significant (F(2, 441) = 3.92, *p* = .021; Positive Valence Simple Main effects: Serendipity vs. Baseline (F(1, 441) = 4.17, *p* = .042), Serendipity vs. Personal Choice (F(1, 441) = 4.32, *p* = .038). | Interaction was significant (F(2, 456) = 3.77, *p* = .024; Positive Valence Simple Main effects: Serendipity vs. Baseline (F(1, 456) = 4.35, *p* = .038), Serendipity vs. Personal Choice (F(1, 456) = 4.24, *p* = .040). | No |
+++++
| Serendipity | Interaction marginally significant (F(2, 441) = 2.39, *p* = .093; Positive Valence Simple Main effects: Serendipity vs. Baseline (F(1, 441) = 13.80, *p* \< .001), Serendipity vs. Personal Choice (F(1, 441) = 6.31, *p* = .012). | Interaction was not significant (F(2, 456) = 2.19, *p* = .113; Positive Valence Simple Main effects: Serendipity vs. Baseline (F(1, 456) = 13.16, *p* \< .001), Serendipity vs. Personal Choice (F(1, 456) = 6.28, *p* = .013). | The interaction goes from marginally significant to not significant if we do not screen participants. The simple effects remain the same. |
+++++
| Meaningfulness | Interaction was marginally significant (F(2, 441) = 2.78, *p* = .063; Positive Valence Simple Main effects: Serendipity vs. Baseline (F(1, 441) = 10.27, *p* = .001), Serendipity vs. Personal Choice (F(1, 441) = 2.39, *p* = .123). | Interaction was marginally significant (F(2, 456) = 2.79, *p* = .062; Positive Valence Simple Main effects: Serendipity vs. Baseline (F(1, 456) = 10.28, *p* = .001), Serendipity vs. Personal Choice (F(1, 456) = 2.47, *p* = .116). | No |
+++++
| Enjoyment Mediation | Positive painting: pathway was significant when comparing the serendipity to the personal choice (β = 8.85, SE = 3.11, 95% CI: \[2.69 to 14.86\]) and baseline (β = 13.05, SE = 3.35, 95% CI: \[6.46 to 19.79\]) conditions | Positive painting: pathway was significant when comparing the serendipity to the personal choice (β = 8.66, SE = 3.03, 95% CI: \[2.60 to 14.61\]) and baseline (β = 12.42, SE = 3.23, 95% CI: \[6.13 to 18.71\]) conditions | No |
+++++
| Meaningfulness Mediation | Positive painting: pathway was significant when comparing the serendipity to the personal choice (β = .53, SE = .17, 95% CI: \[.16 to .90\]) and baseline (β = .78, SE = .21, 95% CI: \[.38 to 1.21\]) conditions | Positive painting: pathway was significant when comparing the serendipity to the personal choice (β = .52, SE = .19, 95% CI: \[.16 to .89\]) and baseline (β = .75, SE = .20, 95% CI: \[.36 to 1.14\]) conditions | No |
+++++
## Materials
Design: 3 (condition: baseline, personal choice, serendipity) × 2 (valence: positive vs. negative) × 2 (painting replicate) between-subjects design
**Instructions**
In this study, we are interested in how people respond in different situations.
Click below to start the study.
**\[Baseline Condition\]**
Imagine you enter an art gallery. You will see and rate a painting on the next page.
Please see and rate this painting.
*Participants viewed one of two paintings* (randomly presented)*.*
Once you view the painting, click below to rate this painting.
**\[Personal Choice Condition\]**
Imagine you enter an art gallery. Two of the paintings the gallery features appear below. Please select one of these two paintings to view.
**Moving Around** by Gerald Chodak **On the Border** by Gerald Chodak
*[Painting thumbnail]* *[Painting thumbnail]*
*Note: Painting images not included in machine-readable version. Participants viewed paintings by Gerald Chodak ("Moving Around" and "On the Border") in either positive or negative valence versions. A pretest confirmed that positive paintings were rated as more positive than negative paintings (see Web Appendix B). See published paper for the visual stimuli.*
**Click below to proceed.**
Please see this painting.
*Participants viewed the painting they chose.*
Once you view the painting, click below to proceed.
**\[Serendipity** **Condition\]**
Imagine you enter an art gallery. Two of the paintings the gallery features appear below.
**Moving Around** by Gerald Chodak **On the Border** by Gerald Chodak
Click below to proceed.
Imagine that you walk down a hallway in the art gallery and turn a corner. Just as you turn the corner, you happen to find this painting on the wall.
*Participants viewed one of two paintings* (randomly presented)*.*
Once you view the painting, click below to proceed.
**\[Positive Valence Condition\]**
**Moving Around** by Gerald Chodak **On the Border** by Gerald Chodak
See paintings at (Moving Around) and (On the Border)
**\[Negative Valence Condition\]**
**Moving Around** by Gerald Chodak **On the Border** by Gerald Chodak
See paintings at (moving around) or (on the border)
How much did you enjoy the painting? (0 = "I hated it," and 100 = "I loved it")
Please indicate your agreement with the following statements:
The way I came across this painting made it seem more meaningful. (1 = "strongly disagree," and 7 = "strongly agree")
Getting to see the painting I just saw was a good surprise. (1 = "strongly disagree," and 7 = "strongly agree")
I came across this painting by luck. (1 = "strongly disagree," and 7 = "strongly agree")
This painting was an unexpected discovery. (1 = "strongly disagree," and 7 = "strongly agree")
To what extent did you feel attached to the other option? (1 = "not at all attached," and 7 = "very much attached")
To what extent did you feel regretful about the painting you saw? (1 = "not at all regretful," and 7 = "very much regretful")
How much did you scrutinize the painting? (1 = "not at all," and 7 = "very much")
How stressed were you with the painting selection process? (1 = "not at all stressful," and 7 = "very much stressful")
How frustrated were you with the painting selection process? (1 = "not at all frustrated," and 7 = "very much frustrated")
# Web Appendix C: Additional Methodological Details and Analyses for Study 3
## Effects of the Movie Trailer Replicates
*Enjoyment*. A 2 (initial selection randomness) × 2 (final selection randomness) × 5 (movie trailer replicate) ANOVA did not reveal a three-way interaction (F(4, 369) = 1.52, *p* = .196), nor two-way interactions between the movie trailer replicate and initial (F(4, 369) = .20, *p* = .940) or final (F(4, 369) = .35, *p* = .841) selection randomness, nor a main effect of the movie trailer replicate (F(4, 369) = .629, *p* = .642). Thus, we collapsed across trailers for the enjoyment analyses.
*Interest in the Platform.* A 2 (initial selection randomness) × 2 (final selection randomness) × 5 (movie trailer replicate) logistic regression did not reveal any effect involving the movie trailer replicates (all *Wald* statistics \< 1) for both interest in the platform measures. Thus, we collapsed across replicates for all the analyses involving interest in the platform.
*Feelings of Serendipity.* A 2 (initial selection randomness) × 2 (final selection randomness) × 5 (movie trailer replicate) ANOVA did not reveal a three-way interaction (F(4, 369) = 1.08, *p* = .368), nor two-way interactions between the movie trailer replicate and initial (F(4, 369) = .611, *p* = .655) or final (F(4, 369) = .380, *p* = .823) selection randomness, nor a main effect of the movie trailer replicate (F(4, 369) = .194, *p* = .942). Thus, we collapsed across trailers for the feelings of serendipity analyses.
## Mediation by Feelings of Serendipity Interest in the Platform
We conducted a bootstrapping moderated mediation analysis using the degree of randomness in the initial selection as the independent variable, degree of randomness in the final selection as the moderator, serendipity as the mediator, and interest in the platform as the dependent variable (Process Model 8; Hayes 2018). For the first interest in the platform measure (yes vs. no), the index of moderated mediation was significant (Index = .27; 95% CI: \[.01 to .64\]). When there was a high degree of randomness in the final selection, the pathway to interest through feelings of serendipity was significant (β = .27, SE = .11, 95% CI: \[.11 to .51\]). When there was a low degree of randomness in the final selection, the pathway to interest through feelings of serendipity was not significant (β = .00, SE = .11, 95% CI: \[-.24 to .21\]). For the second interest in the platform measure (presence of a valid e-mail address), the index of moderated mediation was significant (Index = .22; 95% CI: \[.02 to .55\]). When there was a high degree of randomness in the final selection, the pathway to interest through feelings of serendipity was significant (β = .22, SE = .10, 95% CI: \[.08 to .45\]). When there was a low degree of randomness in the final selection, the pathway to interest through feelings of serendipity was not significant (β = .00, SE = .09, 95% CI: \[-.19 to 19\]).
## Meaningfulness Measure Analysis
A 2 (initial selection randomness) × 2 (final selection randomness) × 5 (movie trailer replicate) ANOVA revealed a marginal three-way interaction (F(4, 369) = 2.28, *p* = .061), no two-way interactions between the movie trailer replicate and initial (F(4, 369) = .369, *p* = .831) or final (F(4, 369) = .742, p = .564) selection randomness, and no main effect of the movie trailer replicate (F(4, 369) = 1.05, *p* = .382). Because the three-way interaction was marginal and did not reveal any consistent pattern, and the other effects involving the movie trailer replicates were not significant, we collapsed across replicates for the subsequent meaningfulness analysis.
A 2 (initial selection randomness) × 2 (final selection randomness) ANOVA on the meaningfulness measure revealed an effect of randomness in the final selection (F(1, 385) = 37.75, *p* \< .001), such that perceptions of meaningfulness were greater when the movie was randomly (M = 4.98, SD = 1.60) rather than deterministically selected (M = 3.91, SD = 1.91). There was a marginally significant effect of degree of randomness in the initial selection (F(1, 385) = 3.36, *p* = .068), such that perceptions of meaningfulness were greater when the movie trailer was selected out of a pool of 100 movies (M = 4.65, SD = 1.85) than when it was selected out of a pool of 10 movies (M = 4.30, SD = 1.80). The interaction was significant (F(1, 385) = 7.46, *p* = .007). When there was a high degree of randomness in the final selection, participants reported greater perceptions of meaningfulness in the high randomness (M = 5.39, SD = 1.39) than in the low randomness in the initial selection condition (M = 4.59, SD = 1.69; F(1, 385) = 10.95, *p* = .001). When there was a low degree of randomness in the final selection, there was no difference between the high (M = 3.83, SD = 1.96) and low (M = 3.99, SD = 1.88) initial selection randomness conditions (F \< 1).
We conducted a bootstrapping moderated mediation analysis using the degree of randomness in the initial selection as the independent variable, degree of randomness in the final selection as the moderator, serendipity as the mediator, and meaningfulness as the dependent variable using Process Model 8 (Hayes 2018). The index of moderated mediation was significant (Index = .54; 95% CI: \[.04 to 1.05\]). When there was a high degree of randomness in the final selection, the pathway to meaningfulness through feelings of serendipity was significant (β = .54, SE = .14, 95% CI: \[.27 to .82\]). When there was a low degree of randomness in the final selection, the pathway to meaningfulness through feelings of serendipity was not significant (β = .00, SE = .21, 95% CI: \[-.42 to .41\]).
## Alternative Explanations
There were no significant main effects nor an interaction on regret (*p*'s \> .38), stress (*p*'s \> .90), frustration (*p*'s \> .42), and scrutinizing (*p*'s \> .13). Thus, these alternative mechanisms cannot explain the pattern of results.
## Main Results without Exclusion Criterion
The Table below presents all the key analyses form the main text, with and without the exclusion criterion.
++
| **Study 3** |
+:===================================================:+:=====================================================================================================================================================================================================:+:====================================================================================================================================================================================================:+:=================:+
| **Analysis** | **Excluding \>3SD P's** | **Not Excluding \>3SD P's** | **Any change?** |
| | | | |
| | **(N = 389)** | **(N = 400)** | |
+++++
| Enjoyment | Main effect of randomness in the final selection (F(1, 385) = 40.08, *p* \< .001; The interaction was significant (F(1, 385) = 4.05, *p* = .045) | Main effect of randomness in the final selection (F(1, 396) = 36.44, *p* \< .001; The interaction was significant (F(1, 396) = 4.23, *p* = .040) | No |
+++++
| Interest in the Platform (Yes vs. No) | Interaction: ß = 1.28, *Wald* = 5.13, *p* = .024; High (vs. low) randomness in the initial selection when movie was randomly chosen: (P~high~ = 34.3% vs. P~low~ = 15.2%): *Wald* = 10.05, *p* = .002 | Interaction: ß = 1.32, *Wald* = 5.61, *p* = .018; High (vs. low) randomness in the initial selection when movie was randomly chosen: (P~high~ = 34.3% vs. P~low~ = 15.7%): *Wald* = 9.71, *p* = .002 | No |
+++++
| Interest in the Platform (Presence of Valid E-Mail) | Interaction: ß = 1.05, *Wald* = 2.59, *p* = .108; High (vs. low) randomness in the initial selection when movie was randomly chosen: (P~high~ = 21.2% vs. P~low~ = 11.4%): *Wald* = 3.60, *p* = .058 | Interaction: ß = 1.04, *Wald* = 2.70, *p* = .101; High (vs. low) randomness in the initial selection when movie was randomly chosen: (P~high~ = 20.6% vs. P~low~ = 12.0%): *Wald* = 2.83, *p* = .093 | No |
+++++
| Serendipity | Main effect of randomness in the final selection (F(1, 385) = 64.90, *p* \< .001; The interaction was significant (F(1, 385) = 4.74, *p* = .030) | Main effect of randomness in the final selection (F(1, 396) = 63.95, *p* \< .001; The interaction was significant (F(1, 396) = 4.34, *p* = .038 | No |
+++++
| Enjoyment Mediation | Path when movie was randomly chosen: β = 8.23, SE = 2.17, 95% CI: \[4.07 to 12.57\] | Path when movie was randomly chosen: β = 8.47, SE = 3.93, 95% CI: \[4.27 to 12.86\] | No |
+++++
| Interest (Yes vs. No) Mediation | Path when movie was randomly chosen: β = .26, SE = .11, 95% CI: \[.10 to .51\] | Path when movie was randomly chosen: β = .25, SE = .10, 95% CI: \[.09 to .49\] | No |
+++++
| Interest (Valid E-Mail) Mediation | Path when movie was randomly chosen: β = .22, SE = .10, 95% CI: \[.08 to .45\] | Path when movie was randomly chosen: β = .19, SE = .09, 95% CI: \[.05 to .41\] | No |
+++++
## Materials
Design: (degree of randomness in the initial selection: high vs. low) × 2 (degree of randomness in the final selection: high vs. low) between-subjects design.
**MOVIE TRAILER ZONE - HELPING YOU DECIDE**\
Dear participant,
We are investigating consumers\' responses to \"Movie Trailer Zone\", a recently launched platform that allows members to receive curated movie trailer recommendations to help them to decide which movie to see. This platform creates a profile for its users and recommends movie trailers. The purpose is to help members to find trailers fitting their preferences, ultimately helping users decide which movie to see.
In this study, you will perform a task that resembles how this platform is used. Since you are a new user, the platform you first learn your preferences. Next, it will recommend a movie trailer. You will watch the movie trailer and answer a few questions.
Please click the arrow button to proceed.
**WELCOME TO MOVIE TRAILER ZONE!**
Movie Trailer Zone is a platform designed to help you make better movie choices. We give curated movie trailer recommendations based on your preferences. We have a wide variety of trailers available.
When you create a Movie Trailer Zone account, you answer a few questions designed to get a grasp of your preferences. We use a subset (out of all movie trailers we have available) to learn your preferences. Then, we make recommendations of movie trailers.
In order to proceed, please create a username. Then, please click the arrow button.
\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_
**\*\[piped username from the previous page\], choose 5 movies you like.**\
\
It will help us find trailers for the movies we think you might like!
After you select five movies, please click the arrow button at the bottom of the page.
Participants saw a 10 x 5 matrix with several movie covers. The names of the movies are below.
Conan the Barbarian, Mission Impossible: Fallout, Fast Color, Final Cut, Creed II, The Osiris Child, The Colony, Judgment Day, Detroit, Antz, How to Train your Dragon: The Hidden World, Chicken Run, Good Will Hunting, Rodeo & Juliet, The Good Guy, My Best Friend's Wedding, All Square, Fun with Dick and Jane, Fox Fire, High Rise, Crypto, Frozen Ground, What Men Want, Fighting with my Family, The Commuter, Arbitrage, Bridget Jones, Dog Days, Wonder, Block Rock, The Prince, Into the Ashes, I Kill Giants, Along with the Gods, The Sisters Brothers, Hero, 50/50, Changeland, Running with the Devil, Lincoln Lawyer, Spinning Man, Big Wedding, Leap Year, And So It Goes, We Die Young, The Backtrace, Daylight's End, Damsel, Under The Tree, The Treasure.
Thank you for letting us know about your preferences. We are building your profile based on your movie selections. Please wait. The page will advance automatically in a few seconds. It might take 10 seconds or a little bit longer depending on your movie selections.
Loading spinner icon (loading spinner gif URL: )
Thank you for letting us know about your preferences. Your profile is ready. Now, we need to examine your preferences in order to make a recommendation. On the next page, while we analyze your preferences, you will answer some general questions that will not be used in developing recommendations. Then, you will experience the platform \"as is\".
Please click the arrow button to proceed to the next page.
**MOVIE TRAILER ZONE**
Analyzing your preferences\...
(Please answer the questions below so we can further fetch your profile)
How often do you go to the movie theater? (1 = Not Very Often, 7 = Very Often)
How often do you watch movies at home? (1 = Not Very Often, 7 = Very Often)
How often do you rent movies via streaming services? (1 = Not Very Often, 7 = Very Often)
Please indicate the extent to which you like or dislike the movie genres below (1 = Dislike a Great Deal, 4 = Neither Like, Nor Dislike, 7 = Like a Great Deal)
Action
Adventure
Animation
Biography
Comedy
Crime
Documentary
Drama
Fantasy
History
Horror
Musical
Mystery
Romance
Science Fiction
Thriller
War
Western
**MOVIE TRAILER ZONE**
\*Loading spinner icon\*
Preparing your recommendation\...
\*The page advanced automatically after 10 seconds (marketer salience: salient vs. non-salient) × 2 (degree of randomness: high vs. low)
**MOVIE TRAILER RECOMMENDATION**
**\[Low Randomness in the Initial Selection, High Randomness in the Final Selection\]**
Based on your profile, we have curated a selection of 10 movie trailers that match your preferences. We have examined your preferences based on these trailers and have randomly selected one movie trailer (out of these 10) that you may enjoy.
**\[High Randomness in the Initial Selection, High Randomness in the Final Selection\]**
Based on your profile, we have a curated selection of 100 movie trailers that match your preferences. We have examined your preferences based on these trailers and have randomly selected one movie trailer (out of these 100) that you may enjoy.
\[**Low Randomness in the Initial Selection, Low Randomness in the Final Selection\]**
Based on your profile, we have curated a selection of 10 movie trailers that match your preferences. We have examined your preferences based on these trailers and have carefully selected one movie trailer (out of these 10) that you may enjoy.
\[**High Randomness in the Initial Selection, Low Randomness in the Final Selection\]**
Based on your profile, we have a curated selection of 100 movie trailers that match your preferences. We have examined your preferences based on these trailers and have carefully selected one movie trailer (out of these 100) that you may enjoy.
**\[All Conditions\]**
Please watch it below. You will be able to proceed once you finish watching it.
\*The movie trailer was embedded into the Qualtrics survey
\*Participants saw one of the following trailers (randomized):
Chronicle
Last Stand
Lawless
Priceless
Wildlife
How much did you enjoy the movie trailer (0 = "I hated it," and 100 = "I loved it")
Do you want to sign up to receive more information about the streaming platform? If so, please leave your e-mail below. We will only send you one email with more information about the platform. (Yes vs. No; The yes option had a text entry box)
Please indicate your agreement with the following statements
The way I came across this movie trailer made it seem more meaningful (1 = "strongly disagree," and 7 = "strongly agree").
Getting to watch this movie trailer was a good surprise (1 = "strongly disagree," and 7 = "strongly agree").
The movie trailer was an unexpected discovery (1 = "strongly disagree," and 7 = "strongly agree").
I came across this movie trailer by luck (1 = "strongly disagree," and 7 = "strongly agree").
Based on how the service works, there was a low chance that I would be watching the specific movie trailer that was selected for me (1 = "strongly disagree," and 7 = "strongly agree").
The movie trailer was selected through a random process (1 = "strongly disagree," and 7 = "strongly agree").
To what extent did you feel regretful about the movie trailer you watched? (1 = "not at all," and 7 = "very much")
How stressed were you with the movie trailer selection process? (1 = "not at all," and 7 = "very much")
How frustrated were you with the movie trailer selection process? (1 = "not at all," and 7 = "very much")
While watching the movie trailer, how much did you scrutinize the movie trailer? (1 = "not at all," and 7 = "very much")
While watching the movie trailer, how many thoughts did you have in your mind? (1 = "not at all," and 7 = "very much")
# Web Appendix C-1: Additional Study on the Role of Chance in Serendipity
We ran an additional study examining the role of chance in serendipitous encounters, an idea we explored in Study 3 (main text). This study investigates the idea that certain choices can increase feelings of serendipity. This occurs as long as these choices involve unknown components that lead to an attribution to chance. We examine food consumption and enjoyment in three conditions: (1) a condition in which participants knew beforehand they would be eating pretzels (no surprise condition), (2) a condition in which participants did not know they would be eating pretzels (surprise condition), and (3) a condition in which participants made an initial choice between two experiences, without knowing the specifics of what they involved, leading them to unexpectedly eat pretzels (serendipity condition). This third condition represents marketplace situations where consumers make an initial decision (e.g., to visit a certain location, website, or radio station) that leads to experiences that are attributed to chance (i.e., "I had no idea what was going to happen and, as a result of that choice, I am now experiencing this positive surprise"). This condition also allows us to show an increase in serendipity relative to a condition in which there is surprise but not the same attribution to chance (condition 2). Given that only the third condition had all necessary components for serendipity to occur (a combination of positiveness, unexpectedness, and attribution to chance), we predicted that participants in this condition would enjoy the pretzels more than participants in the other two conditions.
## Method
*Participants and Design.* One hundred and thirty-eight undergraduates completed the study in exchange for course credit. Because the procedure required participants to receive and taste pretzels, we eliminated fourteen participants who did not eat pretzels, resulting in a final sample of 124 participants (39.5% men; 18 to 45 years, M = 19.89, SD = 2.64). No participants were excluded based on the criterion outlined in Study 1. This study had a three-cell (no surprise, surprise, serendipity) between-subjects design.
*Procedure.* Upon arriving at the behavioral lab, participants were seated in individual cubicles with computers and were asked to wait for the research assistant's instructions. They were told that we were interested in understanding consumer preferences. Because the manipulation involved verbal instructions given to all students at the beginning of the study session, the conditions were randomly administered per session (one condition per session over two mornings and two afternoons. In the no surprise condition, the research assistant told participants: "As part of today's session, you will participate in a study that involves tasting pretzels. If you have any allergy or other reasons for not tasting the pretzels, just let me know. If not, please taste the pretzels and follow the survey instructions accordingly." Participants in the surprise and serendipity conditions did not receive such instructions, just the general lab instructions thanking them for their participation. Thus, they did not know they would be tasting pretzels as part of the study. All participants then started the electronic survey.
In the no surprise condition, participants read that they would participate in a consumer preference study that involved tasting pretzels. The electronic survey repeated the information that had been verbally stated by the research assistant to ensure this information was salient. In the surprise condition, participants read that they would participate in a consumer preference study, without any mention that they would have the opportunity to taste pretzels later. In the serendipity condition, participants read that they would participate in a consumer preference study, that there were two studies available ("Marketing Five Senses Study" and "Consumer Feelings Study"), and were asked to choose one of them. Making this choice should increase serendipity relative to the surprise condition because, without knowing what each study involved, participants should infer that chance had it that their choice led them to eat pretzels. Independent of what they chose, all participants completed the same study next.
After reading the instructions, participants answered filler questions about general eating habits (see the detailed procedure and materials below). These filler questions were part of the purported consumer preferences study. These questions were displayed below a "Consumer Preferences Study" title. In addition to the title, participants in the serendipity condition read a subtitle that piped the name of the study they had chosen on the previous page. Once participants finished the filler questions, they were told to raise their hands so the research assistant would approach their cubicle to give them a code required to proceed to the next page. After raising their hands, participants were approached by the research assistant and received six Utz Mini Twist Pretzels placed inside a Styrofoam bowl. The bowl was covered by a napkin and a paper card that read: "Here are some pretzels for you! Please type the code "pretzels" to continue with the study." Next, participants were told to eat their pretzels for the next minute or so, but that they could take their time. As in the no surprise condition, we asked participants to let us know in case there were any reasons they could not eat pretzels (e.g., an allergy), so we could eliminate participants who did not eat any pretzels.
After tasting the pretzels, participants were asked: How much did you enjoy the pretzels? (0 = "I hated it," and 100 = "I loved it"). To measure feelings of serendipity, we asked: "Getting to taste these pretzels was a good surprise," "The pretzels were an unexpected discovery," "I came across these pretzels by luck," and "when I started doing the study, there was a low chance that I would be eating these pretzels (1 = "strongly disagree," and 7 = "strongly agree"). We combined these measures to create a serendipity index (α = .55). This level of reliability is acceptable, but not high, likely because of variation in inter-item correlation across conditions manipulating different components of serendipity.
We also added two measures to understand whether participants could have enjoyed the pretzels more in the serendipity condition because they felt they had more of a choice over what they were doing (i.e., a positive effect of having a choice). We asked the extent to which it was their choice to eat the pretzels and how responsible they felt for the fact that they ended up eating pretzels (1 = "not at all," and 7 = "very much"). There was no effect of condition on perceptions of having chosen to eat the pretzels (F(2, 121) = 1.43, *p* = .244) or responsibility (F(2, 121) = .535, *p* = .587), which shows that the effects described below were not due to a positive effect of having a choice. When participants finished the survey, the research assistant approached each participant's desk, removed the Styrofoam bowl, and confirmed whether the participant had eaten any pretzels (each bowl had six full, unbroken pretzels). We removed the participants who did not eat any pretzel.
## Results
*Enjoyment*. An ANOVA revealed an effect of condition (F(2, 121) = 11.03, *p* \< .001). Participants reported higher enjoyment of the pretzels in the serendipity (M = 72.59, SD = 20.54) than in the surprise (M = 62.03, SD = 23.94; F(1, 121) = 4.35, *p* = .039) and no surprise (M = 49.43, SD = 24.19; F(1, 121) = 22.02, *p* \< .001) conditions. In addition, participants reported higher enjoyment of the pretzels in the surprise than in the no surprise condition (F(1, 121) = 6.05, *p* = .015).
*Meaningfulness.* An ANOVA revealed an effect of condition (F(2, 121) = 10.85, *p* \< .001). Participants reported higher meaningfulness in the serendipity condition (M = 5.05, SD = 1.84) than in the surprise (M = 3.58, SD = 1.88; F(1, 121) = 12.24, *p* \< .001) and no surprise (M = 3.26, SD = 1.95; F(1, 121) = 21.52, *p* \< .001) conditions. There was no difference between the no surprise and the surprise conditions (F(1, 121) = .56, *p* = .456), which indicates that surprise by itself does not make an experience more meaningful.
*Mediation by Feelings of Serendipity*. An ANOVA revealed an effect of condition (F(2, 121) = 16.57, *p* \< .001). Participants reported greater feelings of serendipity in the serendipity (M = 5.20, SD = 1.01) than in the surprise (M = 4.60, SD = .95; F(1, 121) = 6.06, *p* = .015) and no surprise (M = 3.83, SD = 1.30; F(1, 121) = 33.04, *p* \< .001) conditions. Participants reported greater feelings of serendipity in the surprise than in the no surprise condition (F(1, 121) = 9.08, *p* = .003). This indicates that the more components of serendipity are present, the more feelings of serendipity the experience generates.
We conducted a bootstrapping mediation analysis with the three conditions as multicategorical independent variables, enjoyment as the dependent variable, and feelings of serendipity as the mediator (Hayes 2018; PROCESS model 4). When comparing the serendipity to the surprise condition, the pathway to enjoyment through feelings of serendipity was significant (β = 5.30, SE = 2.11, 95% CI: \[1.35 to 9.71\]). When comparing the serendipity to the no surprise condition, the pathway to enjoyment through feelings of serendipity was also significant (β = 12.05, SE = 3.50, 95% CI: \[5.98 to 19.83\]).
We conducted a similar bootstrapping mediation analysis with the three conditions as multicategorical independent variables, meaningfulness as the dependent variable, and serendipity as the mediator (Hayes 2018; PROCESS model 4). When comparing the serendipity to the surprise condition, the pathway to meaningfulness through serendipity was significant (β = .61, SE = .22, 95% CI: \[.17 to 1.02\]). When comparing the serendipity to the no surprise condition, the pathway to meaningfulness through serendipity was also significant (β = 1.39, SE = .27, 95% CI: \[.87 to 1.96\]). These findings are consistent with our predictions about the unique influence of serendipity on perceptions of meaningfulness.
## Discussion
This study provides evidence for our conceptualization in the context of a food consumption experience. The findings support the idea that feelings of serendipity occur when experiences are positive, unexpected, and attributed to chance. This is evidence that serendipity makes experiences more enjoyable not simply because these experiences involve a positive surprise, but because the surprise is accompanied by attributions to chance.
## Materials
Design: three-cell (no surprise, surprise, serendipity) between-subjects design.
**\[Procedures Outside the Electronic Survey\]**
Because the manipulation involved verbal instructions given to all students at the beginning of the study session, the conditions were randomly administered per session (one condition per session over two mornings and two afternoons).
Upon arriving at the behavioral lab, they seated in individual cubicles with computers and were asked to wait for the upcoming research assistant's instructions. Because the manipulation involved verbal instructions given to all students at the beginning of the study session, the conditions were randomly administered per session (one condition per session over two mornings and two afternoons).
All participants were given standard behavioral lab instructions thanking them for their time and effort, asking them to not use their cell phones and pay attention to the studies. However, as part of the no surprise condition manipulation, the research assistant told participants: "As part of today's session, you will participate in a study that involves tasting pretzels. If you have any allergy or other reasons for not tasting the pretzels, just let me know. If not, please taste the pretzels and follow the survey instructions accordingly." This condition allowed us to see what the average enjoyment of the pretzels was when participants were expecting them, without surprise or serendipity. There were no additional verbal instructions in the surprise and serendipity conditions, so they did not know they would be tasting pretzels as part of the study. After receiving these verbal instructions, participants started the electronic survey.
**\[Electronic Survey\]**
What is your computer / station number? (dropdown list question where participants could indicate their computer station number)
**\[No Surprise Condition\]**
**CONSUMER PREFERENCE STUDY**\
In this session, you will participate in a consumer preference study that involves tasting pretzels. After answering some questions, you will raise your hand and the experimenter will bring some pretzels for you to taste.\
\
Please click the arrow button to continue.
**\[Surprise Condition\]**
**CONSUMER PREFERENCE STUDY**\
In this session, you will participate in a consumer preference study.\
\
Please click the arrow button to continue.
**\[Serendipity Condition\]**
**CONSUMER PREFERENCE STUDY**\
In this session, you will participate in a consumer preference study. There are two studies available. Please choose one of them.\
\
After making your choice, please click the arrow button to continue.
Marketing Five Senses Study Consumer Feelings Study
\*Independent of what they chose, all participants completed the same study next.
**CONSUMER PREFERENCE STUDY**
**\*(participants in the Serendipity condition also saw a piped text retrieving the name of the chosen study \[Marketing Five Senses Study or Consumer Feelings Study\] on the previous page)**\
You will now answer some questions about your eating habits. Please answer the following questions by considering your eating habits for the previous 7 days. In case your eating habits during the previous 7 days have differed significantly from your habitual routine (because you may have been on holiday, you may have temporarily changed your work routine or a life event has occurred), please consider the time before this period.
How many meals do you eat a day (breakfast, lunch, dinner, snacks)?
1-2 meals a day
3-4 meals a day
5-6 meals a day
more than 6 meals a day
\...on the weekend:
1-2 meals a day
3-4 meals a day
5-6 meals a day
more than 6 meals a day
At which times do you usually eat meals?
\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_
At which times do you usually eat snacks?
\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_
Which meals do you usually partake?
breakfast
lunch
dinner
snacks
Do you get up at night to eat because you are hungry or because you can\'t sleep?
Yes
No
What type of food do you preferentially buy (you can choose more than one answer)?
pre-cooked foods
fresh foods
frozen foods
canned foods
other \_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_
\*Upon finishing the filler questions, all participants were told to raise their hands so the research assistant would approach their cubicle to give them a code required to proceed to the next page.
Thank you for your participation so far. Please raise your hand. The experimenter will approach your desk and give you the code required to proceed to the next page. Please keep your hand up and the experimenter will approach your desk very soon!
\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_ (text entry box)
\*After raising their hands, participants were approached by the research assistant and received six Utz Mini Twist Pretzels placed inside a Styrofoam bowl. The bowl was covered by a Napkin and a paper card that read: "Here are some pretzels for you! Please type the code "pretzels" to continue with the study."
++
| **Here are some pretzels for you!** |
| |
| *[Photo of paper card reading "Here are some pretzels for you!" placed above a napkin covering a Styrofoam bowl]* |
| |
| **Please type the code "pretzels" to continue with the study.** |
+===================================================================================+
Paper card (that was put above the napkin for each Styrofoam bowl)
*[Photo of six Utz Mini Twist Pretzels placed inside a Styrofoam bowl, covered by a napkin and the paper card described above. After raising their hands, participants were approached by the research assistant and received this bowl.]*
*Note: Stimulus photos not included in machine-readable version. See published paper for the visual.*
Pretzels inside the Styrofoam bowls
Please eat your pretzels for the next minute or so. You will be able to proceed to the next page after 30 seconds, but take your time.
If there are reasons you cannot eat pretzels (e.g., an allergy), please let us know in the box below. Otherwise, leave the box blank.
\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_
How much did you enjoy the pretzels? (0 = "I hated it," and 100 = "I loved it")
Please indicate your agreement with the following statements:
The way I came across these pretzels made it seem more meaningful. (1 = "strongly disagree," and 7 = "strongly agree")
Getting to taste these pretzels was a good surprise. (1 = "strongly disagree," and 7 = "strongly agree")
I came across these pretzels by luck. (1 = "strongly disagree," and 7 = "strongly agree")
The pretzels were an unexpected discovery. (1 = "strongly disagree," and 7 = "strongly agree")
When I started doing the study, there was a low chance that I would be eating these pretzels (1 = "strongly disagree," and 7 = "strongly agree").
How responsible did you feel for the fact that you ended up eating pretzels? (1 = "not at all," and 7 = "very much")
To which extent was it your choice to eat the pretzels? (1 = "not at all," and 7 = "very much")
# Web Appendix D: Additional Methodological Details and Analyses for Study 4
## Analysis of Perceived Knowledge to Make a Choice
We used four items to measure participants' perceived knowledge to make a choice. We told them to consider their experience with Brain.fm and the information that was provided to them, and asked: "From the information provided about functional music, I was knowledgeable enough to choose a song to listen to," "I learned enough about functional music from the information provided that I could decide what would be best for me," "I knew enough about functional music from the information provided to form judgments about what makes a good functional song," and "When it comes to functional music, the information provided allowed me to distinguish the good from the bad" (1 = "strongly disagree," and 7 = "strongly agree"). We combined the items to form a perceived knowledge to make a personal choice index (α = .95).
*Mediation Analysis.* In addition, a 2 (condition) × 2 (information) ANOVA on perceived knowledge to make a personal choice revealed a main effect of information, such that participants in the diagnostic condition (M = 4.77, SD = 1.61) reported greater knowledge to make a choice than those in the nondiagnostic condition (M = 3.41, SD = 1.69; F(1, 389) = 73.15, *p* \< .001). The interaction (F(1, 389) = .188, *p* = .665) and main effect of condition (F(1, 389) = .220, *p* = .639) were not significant.
We conducted a bootstrapping moderated mediation analysis using PROCESS model 15 (Hayes 2018). We used information (diagnostic vs. nondiagnostic) as the independent variable, condition as the moderator variable, and perceived knowledge to make a choice as the mediator for each of the outcomes. Thus, differently from the main text mediation analysis, the independent variable must be information, as this allows us to understand how perceived knowledge played a role in each condition (serendipity vs. personal choice). For satisfaction, the index of moderated mediation was significant (Index = -.80; 95% CI: \[-1.19 to -.47\]). In the personal choice condition, the pathway to satisfaction through perceived knowledge to make a choice was positive (β = .53, SE = .12, 95% CI: \[.31 to .80\]). In the serendipity condition, this pathway was negative (β = -.27, SE = .12, 95% CI: \[-.51 to -.06\]). These results suggest that when consumers have information that makes them think they can make their own choices, being able to choose increases satisfaction, whereas being assigned an option in a serendipitous way decreases satisfaction. A similar pattern emerged for the interest, willingness to recommend, and willingness to pay measures (see the mediation analyses sections in Web Appendix D).
## Effects of the Song Replicates
*Satisfaction.* A 2 (condition) × 2 (information) × 5 (song replicate) ANOVA did not reveal a three-way interaction (F(4, 373) = .549, *p* = .700) nor two-way interactions between song replicate and condition (F(4, 373) = 1.42, *p* = .227) or information (F(4, 373) = 1.79, *p* = .130). There was a main effect of song replicate (F(4, 373) = 4.43, p = .002). Because we did not make any a priori predictions about replicate main effects, we compare the song replicates using Post-hoc tests with Sidak correction to control for family-wise error rate. We found that participants who listened to the song "Piano Focus" (M = 5.48, SD = 1.43) were more satisfied than those who listened to "Atmospheric Focus" (M = 4.52, SD = 1.84; *p* = .001), "Electronic Focus" (M = 4.73, SD = 1.74; *p* = .037), and "Grooves Focus" (M = 4.65, SD = 1.74; *p* = .012). We also found that participants who listened to "LoFi Focus" (M = 5.29, SD = 1.48) were more satisfied than those who listened to "Atmospheric Focus" (*p* = .032). Participants were equally satisfied when then listened "LoFi Focus" or "Piano Focus" (*p* = .999).
*Interest in the Platform.* A 2 (condition) × 2 (information) × 5 (song replicate) ANOVA did not reveal a three-way interaction (F(4, 373) = .985, *p* = .415) nor two-way interactions between song replicate and condition (F(4, 373) = 1.09, *p* = .362) or information (F(4, 373) = 1.23, *p* = .300). There was a marginally significant effect of song replicate (F(4, 373) = 2.24, *p* = .064). Because we did not make any a priori predictions about replicate main effects, we compare the song replicates using Post-hoc tests with Sidak correction to control for family-wise error rate. Participants were marginally more interested in the platform when the song played (or chosen) was "Piano Focus" (M = 4.52, SD = 1.81) rather than "Atmospheric Focus" (M = 3.70, SD = 2.12; *p* = .059). Because this effect does not change the interpretation of the results, we collapsed across song replicates for the interest in the platform analysis.
*Willingness to Recommend.* A 2 (condition) × 2 (information) × 5 (song replicate) ANOVA did not reveal a three-way interaction (F(4, 373) = 1.00, *p* = .406) nor and two-way interactions between song replicate and condition (F(4, 373) = .874, *p* = .480) or information (F(4, 373) = 1.24, *p* = .295). There was a marginally significant effect of song replicate (F(4, 373) = 2.07, *p* = .084), but post-hoc tests with Sidak correction did not reveal any significant (or marginally significant) effects. Because this main effect does not change the interpretation of the results, we collapsed across song replicates for the willingness to recommend analysis.
*Willingness to Pay.* A 2 (condition) × 2 (information) × 5 (song replicate) ANOVA did not reveal a three-way interaction (F(4, 373) = .616, *p* = .651), nor two-way interactions between song replicate and condition (F(4, 373) = 2.03, *p* = .090) or information (F(4, 373) = .260, *p* = .904), nor a main effect of song replicate (F(4, 373) = .358, *p* = .839). We collapsed across song replicates for the willingness to pay analysis.
*Feelings of Serendipity.* A 2 (condition) × 2 (information) × 5 (song replicate) ANOVA on feelings of serendipity did not reveal a three-way interaction (F(4, 373) = 1.07, *p* = .372), nor two-way interactions between song replicate and condition (F(4, 373) = .56, *p* = .694) or information (F(4, 373) = 1.44, *p* = .221), nor a main effect of song replicate (F(4, 373) = .757, *p* = .554). Thus, we collapsed across replicates.
*Perceived Knowledge.* A 2 (condition) × 2 (information) × 5 (song replicate) ANOVA on perceived knowledge to make a personal choice did not reveal a three-way interaction (F(4, 373) = .587, *p* = .672), nor two-way interactions between song replicate and condition (F(4, 373) = .511, *p* = .728) or information (F(4, 373) = .990, *p* = .413), nor a main effect of song replicate (F(4, 373) = 1.18, *p* = .321). Thus, we collapsed across replicates.
## Alternative Explanations
We collapsed across replicates for the alternative explanation analyses. There was a main effect of condition on regret (F(1, 389) = 7.18, *p* = .008), such that participants felt more regretful when they made a personal choice (M = 2.25, SD = 1.74) than when the song was randomly selected to them (M = 1.83, SD = 1.32). There was no interaction (F(1, 389) = .00, *p* = .992) or main effect of information (F(1, 389) = .137, *p* = .712). Participants also scrutinized the songs marginally more in the personal choice (M = 3.88, SD = 1.94) than in the serendipity condition (M = 3.54, SD = 1.97; F(1, 389) = 3.04, *p* = .082), with no other effects (F's \< 1). Participants had more positive expectations when they made a personal choice (M = 4.43, SD = 1.46) than in the serendipity condition (M = 3.89, SD = 1.41; F(1, 389) = 13.73, *p* \< .001). While some differences were found, they cannot explain the pattern of results on any of the dependent measures, and none of the indices of moderated mediation were significant.
## Mediation Analyses Interest in the Platform
We conducted two bootstrapping mediation analyses using PROCESS model 15 (Hayes 2018) with specifications similar to those described in the main text. For the serendipity mediator, the index of moderated mediation was significant (Index = -.39; 95% CI: \[-.67 to -.16\]). When the information was nondiagnostic, the pathway to interest in the platform through feelings of serendipity was positive (β = .51, SE = .13, 95% CI: \[.27 to .78\]). When the information was diagnostic, the pathway to interest in the platform through feelings of serendipity was not significant (β = .12, SE = .07, 95% CI: \[-.01 to .27\]).
For the perceived knowledge mediator, the index of moderated mediation was significant (Index = -.57; 95% CI: \[-1.01 to -.20\]). When participants made a personal choice, the pathway to interest in the platform through perceived knowledge to make a personal choice was positive (β = .56, SE = .16, 95% CI: \[.27 to .89\]). When participants listened to a serendipitously selected song, the pathway to interest through perceived knowledge was nonsignificant (β = .01, SE = .12, 95% CI: \[-.27 to .22\]).
## Mediation Analyses Willingness to Recommend
We conducted two bootstrapping mediation analyses using PROCESS model 15 (Hayes 2018) with specifications similar to those described in the main text. For the serendipity mediator, the index of moderated mediation was significant (Index = -36; 95% CI: \[-.61 to -.15\]). When the information was nondiagnostic, the pathway to willingness to recommend in the platform through feelings of serendipity was positive (β = .54, SE = .13, 95% CI: \[.28 to .80\]). Unexpectedly, when the information was diagnostic, the pathway to willingness to recommend in the platform through feelings of serendipity was significant (β = .18, SE = .07, 95% CI: \[.05 to .32\]). While feelings of serendipity drove participants' willingness to recommend regardless of the information type, serendipity was a stronger driver for willingness to recommend in the moderate information quantity condition (pairwise contrast between indirect effects: -.36, contrast 95% CI: \[-.62 to -.15\]).
For the perceived knowledge mediator, the index of moderated mediation was significant (Index = -.52; 95% CI: \[-.92 to -.18\]). When participants made a personal choice, the pathway to willingness to recommend through perceived knowledge to make a personal choice was positive (β = .45, SE = .13, 95% CI: \[.21 to .73\]). When participants listened to a serendipitously selected song, the pathway to willingness to recommend through perceived knowledge was nonsignificant (β = -.07, SE = .12, 95% CI: \[-.31 to .15\]).
## Mediation Analyses Willingness to Pay
We conducted two bootstrapping mediation analyses using PROCESS model 15 (Hayes 2018) with specifications similar to those described in the main text. For the serendipity mediator, the index of moderated mediation was marginally significant (Index = -.28; 90% CI: \[-.60 to -.02\]). When the information was nondiagnostic, the pathway to willingness to pay through feelings of serendipity was positive (β = .41, SE = .15, 90% CI: \[.19 to .68\]). When the information was diagnostic, the pathway to willingness to pay through feelings of serendipity was not significant (β = .13, SE = .11, 90% CI: \[-.05 to .31\]).
For the perceived knowledge mediator, the index of moderated mediation was significant (Index = -.59; 95% CI: \[-1.24 to -.01\]). When participants made a personal choice, the pathway to willingness to pay through perceived knowledge to make a personal choice was positive (β = .84, SE = .24, 95% CI: \[.40 to 1.34\]). When participants listened to a serendipitously selected song, the pathway to willingness to pay through perceived knowledge was nonsignificant (β = .25, SE = .19, 95% CI: \[-.13 to .62\]).
## Main Results without Exclusion Criterion
The Table below presents all the key analyses form the main text, with and without the exclusion criterion.
++
| **Study 4** |
+:==================================:+:==================================================================================:+:==================================================================================:+:=================:+
| **Analysis** | **Excluding \>3SD P's** | **Not Excluding \>3SD P's** | **Any change?** |
| | | | |
| | **(N = 393)** | **(N = 400)** | |
+++++
| Satisfaction | The interaction was significant (F(1, 389) = 23.01, *p* \< .001) | The interaction was significant (F(1, 396) = 25.21, *p* \< .001) | No |
+++++
| Interest in the Platform | The interaction was significant (F(1, 389) = 22.56, *p* \< .001) | The interaction was significant (F(1, 396) = 24.78, *p* \< .001) | No |
+++++
| Willingness to Recommend | The interaction was significant (F(1, 389) = 34.37, *p* \< .001) | The interaction was significant (F(1, 396) = 37.66, *p* \< .001) | No |
+++++
| Willingness to Pay | The interaction was significant (F(1, 389) = 8.74, *p* = .003) | The interaction was significant (F(1, 396) = 8.36, *p* = .004) | No |
+++++
| Serendipity | The main effect of serendipity was significant (F(1, 389) = 19.64, *p* \< .001) | The main effect of serendipity was significant (F(1, 396) = 19.95, *p* \< .001) | No |
+++++
| Satisfaction Mediation | Path when information was nondiagnostic: β = .49, SE = .12, 95% CI: \[.26 to .75\] | Path when information was nondiagnostic: β = .50, SE = .12, 95% CI: \[.26 to .74\] | No |
+++++
| Interest in the Platform Mediation | Path when information was nondiagnostic: β = .51, SE = .13, 95% CI: \[.27 to .78\] | Path when information was nondiagnostic: β = .51, SE = .13, 95% CI: \[.27 to .77\] | No |
+++++
| Willingness to Recommend Mediation | Path when information was nondiagnostic: β = .54, SE = .13, 95% CI: \[.28 to .80\] | Path when information was nondiagnostic: β = .54, SE = .13, 95% CI: \[.30 to .81\] | No |
+++++
| Willingness to Pay Mediation | Path when information was nondiagnostic: β = .41, SE = .15, 90% CI: \[.19 to .68\] | Path when information was nondiagnostic: β = .41, SE = .14, 90% CI: \[.15 to .72\] | No |
+++++
## Materials
Design: 2 (condition: personal choice vs. serendipity) × 2 (information: nondiagnostic vs. diagnostic) × 5 (song replicate) between-subjects design.
**BRAIN.FM FUNCTIONAL MUSIC FOR THE BRAIN**
You will now complete a study about Brain.fm, a subscription-based platform that provides music designed for the brain.
Please put your headphones on and make sure they are working.
When you are ready to begin please click the arrow button.
We want to make sure your headphones are working properly. A sound should be playing in the background. What is this sound about? **Make sure your volume is adjusted and that the song is not muted on the browser/tab.**
Please indicate the sound that is playing on the background.
- Rain
- Birds Singing
- Piano
- Car honk
- Cricket
\*The following rain song was playing: . Participants couldn't see any information (we hid the embedded video). Participants could only proceed if they chose Rain.
Music is a potent phenomenon in human auditory perception and cognition. It can make us awake or put us to sleep. Importantly, music can also be purposefully designed to help us shape our behavior. At Brain.fm, we develop sounds that could be usefulto help us study, to push us in a workout, to get us to sleep, or many other possible functions.\
\
On the next pages, you will see a description of Brain.fm, a platform that provides music designed for the brain. Be sure to read the information very carefully - you need to learn about this platform in order to successfully complete the study.
(there was an image to increase immersion it was an animated gif extracted from Brain.fm real website - ).
**BRAIN.FM - FUNCTIONAL MUSIC TO IMPROVE FOCUS**
(gif: ).
Brain.fm is a subscription-based service that makes music to help you do what you need to do. The company uses a patented A.I. engine to create music backed by scientific research to help listeners focus and concentrate. While other music is primarily made to sound good and evoke feelings, Brain.fm works with teams of scientists and composers to engineer music specifically designed to help you enter particular mental states within minutes of use.
\*the information above was displayed side by side in a matrix. The text was at the right of the picture.
**What is functional music, anyway?**
(gif: ).
Broadly speaking, \"functional music\" is a music category where songs are used for a specific purpose. Some songs are developed to improve concentration (such as Brain.fm), but there are also platforms that offer music for relaxation and meditation. Functional music involves many characteristics, including melodies, tempo, pitch, neural phase-locking value, induced brain wave, 3D externalized sound, and brain modulation rate.
**\[Nondiagnostic Information Condition\]**
At Brain.fm, functional music is created based on three main steps:
- **Initial Composition:** First, humans compose the musical content (the Art: melodies, harmonies, chord progressions, sound design, instrument choices, etc.) We have found no substitute for the talent of brilliant musicians in laying the foundation for a new piece of music.
- **Algorithm-Based Arrangement**: Then, a patented algorithmic system (A.I.) arranges these motifs over long timescales, and adds the acoustic features which constitute our core technology (the Science: modulation to alter brain activity, 3D spatialization, salience reduction, etc.)
- **Field Experiment Testing:** Finally, compositions are tested via large-scale experiments to ensure they have the properties required to help listeners attain and sustain desired mental states over long periods of time.
**\[Diagnostic Information Condition\]**
- **Neural Phase-Locking Value (NPLV**): ranging from 4000 to 13000, NPLV refers to the extent to which populations of neurons engage in various kinds of coordinated activity. When the neurons are highly coordinated (high NPLV), people can maintain a higher level of concentration and be less distracted.
- **Induced Brain Wave:** There are five human brain waves - Delta (0.1 - 3 Hz), Theta (4 - 7 Hz), Alpha (8 - 15 Hz), Beta (16 - 30 Hz), and Gamma (31 - 100 Hz). Theta waves are for intuition and memory. Alpha waves are for visualization and creativity. Beta waves are for alertness and concentration. Gamma waves are for insight and peak focus.
- **Brain Modulation Rate:** The higher the modulation rate, the quicker it is for the effect of a certain song to kick in. Songs with low modulation rate take 15 minutes to start generating noticeable effects. Medium modulation rate songs take 10 minutes, and high modulation rate songs take 5 minutes to generate noticeable effects on brain and behavior.
Once you read and understand, click the arrow button. It will be available after some seconds.
Now that you know more about the platform, you will experience a sample of one of the platform\'s available songs.
**\[Personal Choice Condition\]**
Please choose one of the five songs available for a listening sample. You will listen to the song you choose on the next page.
**\[Serendipity Condition\]**
Please see the five songs available for a listening sample. We will randomly select one song for you to listen to on the next page.
**LoFi Focus**
\- Neural Phase-Locking Value: 6000\
- Induced Brain Wave: Gamma (80 Hz)\
- Brain Modulation Rate: Medium \
\
**Atmospheric Focus**
\- Neural Phase-Locking Value: 9600\
- Induced Brain Wave: Beta (25 Hz)\
- Brain Modulation Rate: High
**Electronic Focus**
\- Neural Phase-Locking Value: 10000\
- Induced Brain Wave: Alpha (15 Hz)\
- Brain Modulation Rate: High
**Grooves Focus**
\- Neural Phase-Locking Value: 8500\
- Induced Brain Wave: Gamma (90 Hz)\
- Brain Modulation Rate: Low
**Piano Focus**
\- Neural Phase-Locking Value: 5600\
- Induced Brain Wave: Theta (7 Hz)\
- Brain Modulation Rate: High
\*The songs participants listened to are 2-minute excerpts of functional songs used by the real Brain.fm service. The songs are akin to ambient and binaural beats songs together. More information available upon request\*
\*Participants in the personal choice condition listened to the song they chose, and participants in the serendipity condition listened to a song randomly selected within Qualtrics.
The song is playing on the background\...
** **(gif: ). ** \
**
** **Please keep listening the song. The page will auto advance after 60 seconds.
\*The song information (title and three attributes) was displayed here\*
"How satisfied are you with the song you just listened to?" (1 = "not at all satisfied," and 7 = "very satisfied")
"How satisfied are you with the song listening experience in general?" (1 = "not at all satisfied," and 7 = "very satisfied")
"How interested would you be in subscribing to Brain.fm's platform?" (1 = "not at all," and 7 = "very interested")
"How likely are you to recommend Brain.fm's subscription service to a friend?" (1 = "not likely at all," and 7 = "very likely).
Brain.fm has several subscription options, and such as other platforms (e.g., Spotify, Apple Music), the monthly plan costs between \$5 and \$15.
How much are you willing to pay for a one-month subscription to Brain.fm?
(Slider scale from 5 to 15).
Please consider your experience with Brain.fm and how the Brain.fm platform works. Please indicate the extent to which you agree with the statements below.
"Getting to experience this one song I just listened to ended up being a good surprise," (1 = "strongly disagree," and 7 = "strongly agree").
"Considering the song selection process, I feel lucky to have come across the song I listened to," (1 = "strongly disagree," and 7 = "strongly agree").
"From what it could have been, I feel that the song I listened to was an unexpected discovery," (1 = "strongly disagree," and 7 = "strongly agree").
"I feel that there was some element of chance involved in having experienced this specific (1 = "strongly disagree," and 7 = "strongly agree").
song I just listened to" (1 = "strongly disagree," and 7 = "strongly agree").
Please consider your experience with Brain.fm and the information that was provided to you. Please indicate the extent to which you agree with the statements below.
"From the information provided about functional music, I was knowledgeable enough to choose a song to listen to," (1 = "strongly disagree," and 7 = "strongly agree").
"I learned enough about functional music from the information provided that I could decide what would be best for me," (1 = "strongly disagree," and 7 = "strongly agree").
"I knew enough about functional music from the information provided to form judgments about what makes a good functional song," (1 = "strongly disagree," and 7 = "strongly agree").
"When it comes to functional music, the information provided allowed me to distinguish the good from the bad" (1 = "strongly disagree," and 7 = "strongly agree").
To what extent did you feel regretful about the song you listened to? (1 = "not at all," and 7 = "very much")
"While listening to the song, how much did you scrutinize the song?" (1 = "not at all," and 7 = "very much")
"How high were your expectations about the songs before you listened to one of them?" (1 = "very low," and 7 = "very high")
"How satisfied did you expect to like the song before you listened to it?" (1 = "not at all," and 7 = "very satisfied")
"To which extent did you make your own song choice?" (1 = "not at all," and 7 = "very much")
"How much information was provided about what attributes are necessary for a good functional song?" (1 = "not much," and 7 = "very much")
# Web Appendix D-1: Additional Study on the Role of Information
We ran an additional study examining the role of information, similar to Study 4 in the main paper. This study was different in that we manipulated the amount of information participants received about the options themselves rather than diagnostic information about the product category. We propose that having a lot of information about all the possible options a consumer may receive decreases the appeal of a serendipitous product encounter. Previous work (Botti and McGill 2006; study 3) demonstrated that when people do not have information about the choice options, they prefer to have an option chosen for them. This effect was driven by responsibility, as participants blamed themselves for the option they chose when making a choice without having enough information to differentiate the options. In the same study, people preferred to have a personal choice when they had enough information to differentiate the options. While the results were attributed to how much participants could differentiate the options in each condition, the differentiation manipulation also changed how much information about the options participants had. To test our conceptualization, we kept differentiation constant while manipulating the amount of information about the options presented to participants.
Using songs, we propose that when there is a moderate quantity of information about the options, consumers will enjoy a song more when the song was randomly chosen for them, an effect driven by serendipity. When there is a high quantity of information about the options, consumers will prefer to choose themselves, and feel responsible for the choice outcome.
## Method
*Participants and Design*. Five hundred and thirty-seven participants were recruited from Amazon's Mechanical Turk and paid a small monetary compensation. After eliminating thirteen outliers based on the criterion outlined in Study 1, the final sample size was 524 (49.5% men; 19 to 81 years, M = 39.08, SD = 12.96). This study had a 2 (condition: personal choice vs. serendipity) × 2 (attribute information quantity: moderate vs. high) × 5 (song replicate) between-subjects design.
*Procedure*. Participants were told that we were interested in how people respond in different situations and were randomly assigned to one of four conditions. In the moderate information quantity conditions, participants saw the title, artist, year of release, and genre of five songs. In the high quantity condition, participants saw the same five songs and information, but were also provided with information about the songs' uniqueness, lyrics meaningfulness, vocals, and an overall rating. Consistent with Botti and McGill (2006), in the high information quantity condition, the more diagnostic attribute (overall rating) for each of the five songs was listed as average, but the less diagnostic attributes varied from low to high (see the materials below). Notably, this low differentiation was consistent with the low differentiation in the moderate information quantity condition, allowing us to assess the effect of attribute information quantity independent of differentiation. After examining the list of songs, participants in the personal choice condition were asked to "select one of these five songs." Once participants proceeded to the next page, they heard the song they chose.
Participants in the serendipity condition went through a similar procedure, but instead of being asked to choose one of the songs, they were simply asked to see the five songs and advance to the next page. Once participants proceeded to the next page, one of the five songs, randomly chosen within Qualtrics, started playing. After listening to the song, participants were asked: "How much did you enjoy the song?" (0 = "I hated it," and 100 = "I loved it"). Participants then responded to three items measuring serendipity (α = .84) and four items measuring responsibility (α = .87). Participants also answered questions about stress and frustration, but there were no main effects or interactions on these items (F's \< 1). The complete set of items can be found in the materials presented at the end of this Web Appendix.
Finally, we conducted a pretest to ensure that the options were not perceived as more or less differentiated across information quantity conditions. A separate sample (N = 128) was randomly assigned to one of the four conditions from the main study. Afterward, participants indicated their agreement with the statement "I chose the song that I wanted to hear" (1 = "strongly disagree," and 7 = "strongly agree") and answered the question "Can you distinguish the songs from one another? (1 = "not at all," and 7 = "very much"). Participants in the personal choice condition reported that they chose the song that they wanted to a larger extent (M = 5.80, SD = 1.42) than the serendipity condition (M = 2.06, SD = 1.73; F(1, 124) = 178.72, *p* \< .001). All other effects were nonsignificant (all *p's* \> .31). The analysis of perceived differentiation revealed no main effects or an interaction (*p's* \> .14). Particularly, participants in the moderate quantity condition reported similar differentiation (M = 4.41, SD = 1.99) as those in the high quantity condition (M = 4.66, SD = 1.64; F \< 1). This suggests that the manipulations did not influence perceived differentiation among the options.
## Results
*Enjoyment.* A 2 (condition) × 2 (attribute information quantity) × 5 (song replicate) ANOVA did not reveal a three-way interaction (F(4, 504) = 1.79, *p* = .129) nor a two-way interaction between condition and song replicate (F(4, 504) = 1.27, *p* = .281). There was an interaction between attribute information quantity and song replicate (F(4, 504) = 4.55, *p* = .001) and a main effect of song replicate (F(4, 504) = 6.99, *p* \< .001). These effects occurred because some songs were rated as more enjoyable than others, and because providing more (or less) attribute information about the songs influenced enjoyment. Because these effects do not change the interpretation of results, we collapsed across song replicates for the enjoyment analysis.
A 2 (condition) × 2 (attribute information quantity) ANOVA revealed only an interaction between the condition and information quantity factors (F(1, 520) = 6.50, *p* = .011; see Figure A). The main effects of condition and information quantity were not significant (F's \< 1). Participants in the moderate information quantity condition reported higher enjoyment in the serendipity (M = 60.91, SD = 22.81) than in the personal choice condition (M = 53.27, SD = 29.05; F(1, 520) = 5.62, *p* = .018). Participants in the high information quantity condition reported a similar level of enjoyment in the the serendipity (M = 55.39, SD = 26.29) and in the personal choice conditions (M = 59.31, SD = 24.75; F(1, 520) = 1.51, *p* = .219). Within the serendipity condition, enjoyment marginally decreased when attribute information quantity was high (M~moderate~ = 60.91, SD~moderate~ = 22.81, M~high~ = 55.39, SD~high~ = 26.29; F(1, 520) = 3.74, *p* = .054).
**FIGURE WA-1: STUDY D-1 RESULTS**
*Bar chart showing enjoyment ratings (0-100 scale) across a 2 (Condition: Serendipity vs. Personal Choice) x 2 (Attribute Information Quantity: Moderate vs. High) design. When information quantity was moderate: Serendipity M = 60.91, Personal Choice M = 53.27 (p = .018). When information quantity was high: Serendipity M = 55.39, Personal Choice M = 59.31 (not significant, p = .219). The interaction was significant (p = .011), replicating the finding from Study 4 that serendipity increases enjoyment when consumers have moderate information, but the effect disappears when consumers have enough information to make informed choices themselves.*
*Note: Figure image not included in machine-readable version. See published paper for the visual.*
*Notes*: Error bars = +/− 1 SE. ^†^*p* \< .10. \**p* \< .05. \*\**p* \< .01. Unbracketed comparisons are not significantly different from one another.
*Meaningfulness.* A 2 (condition) × 2 (attribute information quantity) × 5 (song replicate) ANOVA did not reveal a three-way interaction (F(4, 504) = .964, *p* = .427), nor two-way interactions between condition and song replicate (F(4, 504) = 1.36, *p* = .246), nor a main effect of song replicate (F(4, 504) = .954, p = .433). There was an interaction between attribute information quantity and song replicate (F(4, 504) = 3.13, *p* = .015). This effect emerged because providing more (or less) attribute information about the songs influenced meaningfulness. Because this effect does not change the interpretation of results, we collapsed across song replicates for subsequent analysis.
A 2 (condition) × 2 (attribute information quantity) ANOVA revealed only a marginally significant interaction between the condition and information quantity factors on meaningfulness (F(1, 520) = 2.77, *p* = .097). The main effects of condition and information quantity were not significant (Fs \< 1). Participants in the moderate information quantity condition reported marginally higher perceptions of meaningfulness in the serendipity (M = 3.33, SD = 1.76) than in the personal choice condition (M = 2.92, SD = 1.80; F(1, 520) = 3.23, *p* = .073). Participants in the high information quantity condition did not differ in perceptions of meaningfulness between the serendipity (M = 3.19, SD = 1.83) and the personal choice conditions (M = 3.31, SD = 1.89; F(1, 520) = .361, *p* = .548). We note that, within the serendipity condition, there was no difference in feelings of serendipity across the moderate and high information conditions (M~moderate~ = 3.33, SD~moderate~ = 1.76, M~high~ = 3.19, SD~high~ = 1.83; F(1, 520) = .361, *p* = .548).
*Mediation by Feelings of Serendipity and Responsibility.* A 2 (condition) × 2 (attribute information quantity) × 5 (song replicate) ANOVA on the serendipity index did not reveal a three-way interaction (F(4, 504) = .963, p = .427) nor a two-way interaction between condition and song replicate (F(4, 504) = 1.27, p = .279). There was an interaction between attribute information quantity and song replicate (F(4, 504) = 3.91, *p* = .004). There was a marginally significant main effect of song replicate (F \< 1). Because these effects do not change the interpretation of results, we collapsed across song replicates for the subsequent analyses. A 2 (condition) × 2 (attribute information quantity) ANOVA revealed only a marginally significant interaction between the condition and information quantity factors (F(1, 520) = 3.31, *p* = .069). The effects of condition and quantity were not significant (F's \< 1). Participants in the moderate information quantity condition reported marginally greater feelings of serendipity in the serendipity (M = 4.04, SD = 1.43) than in the personal choice condition (M = 3.72, SD = 1.68; F(1, 520) = 2.72, *p* = .099). For participants in the high information quantity condition, feelings of serendipity did not change across serendipity (M = 3.86, SD = 1.59) and personal choice conditions (M = 4.03, SD = 1.48; F(1, 520) = .886, *p* = .347).
A 2 (condition) × 2 (attribute information quantity) × 5 (song replicate) ANOVA on the responsibility measure did not reveal a three-way interaction (F(4, 504) = .206, *p* = .935), nor two-way interactions between condition and song replicate (F(4, 504) = .697, *p* = .594) or attribute information quantity (F(4, 504) = 1.56, *p* = .183), nor a main effect of song replicate (F(4, 504) = 1.11, p = .353). Thus, we collapsed across replicates for the subsequent analyses. A 2 (condition) × 2 (attribute information quantity) ANOVA revealed a main effect of condition on responsibility, such that participants reported higher responsibility in the personal choice (M = 3.75, SD = 1.59) than in the serendipity condition (M = 2.44, SD = 1.61; F(1, 520) = 86.17, *p* \< .001). The effects of information quantity and the interaction were not significant (F's \< 1).
We conducted two bootstrapping analyses for moderated mediation using condition (serendipity vs. personal choice) as the independent variable, attribute information quantity (high vs. moderate) as the moderator, serendipity and responsibility as mediators (one for each analysis), and enjoyment as the dependent variable (Hayes 2018). For the serendipity mediator (PROCESS Model 8), the indirect effect was not significant for the moderate information quantity (β = 2.02, SE = 1.25, 95% CI: \[-.37 to 4.44\]) and high information quantity (β = -1.12, SE = 1.20, 95% CI: \[-3.49 to 1.24\]) conditions at the 95% confidence interval level. However, the results with 90% CIs were supportive of the notion that feelings of serendipity drive enjoyment when the information quantity is moderate (β = 2.02, SE = 1.23, 95% CI: \[.01 to 4.02\]). We used PROCESS Model 8 (i.e., a model with the moderator influencing the indirect path pre-mediator) as there was an interaction between condition and information quantity on the mediator. Thus, PROCESS Model 8 was adequate because the influence of condition on feelings of serendipity should depend on the moderator.
For the responsibility mediator (PROCESS Model 15), while responsibility drove participants' enjoyment regardless of the attribute information quantity (moderate: β = -3.15, SE = .70, 90% CI: \[-4.37 to -2.08\]; high: β = -4.51, SE = .76, 90% CI: \[-5.81 to -3.31\]), perceived responsibility was a stronger driver for enjoyment in the high information condition (pairwise contrast between indirect effects: -1.35, contrast 90% CI: \[-2.78 to -.02\]). These findings support the prediction that feelings of serendipity drive enjoyment under a moderate quantity of attribute information, but not under a high quantity of information. We used PROCESS Model 15 (i.e., a model with the moderator influencing the indirect path post-mediator) as there was a main effect of condition on the mediator. Thus, PROCESS Model 15 was adequate because the influence of condition on responsibility should depend on the amount of information.
We conducted similar bootstrapping analyses using meaningfulness as the dependent variable. For the serendipity mediator (PROCESS Model 8), the indirect effect was not significant for the moderate information quantity (β = .15, SE = .09, 95% CI: \[-.04 to .32\]) and high information quantity (β = -.08, SE = .09, 95% CI: \[-.25 to .09\]) conditions at the 95% confidence interval level. However, the results with 90% CIs were supportive of the notion that feelings of serendipity drive meaningfulness when the information quantity is moderate (β = .15, SE = .09, 95% CI: \[.01 to .29\]). For the responsibility mediator (PROCESS Model 15), we found results similar to those of enjoyment (moderate: β = -.33, SE = .06, 90% CI: \[-.42 to -.24\]; high: β = -.38, SE = .06, 90% CI: \[-.48 to -.28\]), but the contrast between indirect effects was not significant (90% CI: -.15 to .05).
## Discussion
This study provided additional support for the role of serendipity in consumer outcomes. When a song was selected for participants, the positive effect on enjoyment occurred under moderate amounts of information, but reversed when a high amount of attribute information was provided. Similar to Study 4 in the main text, the encounter still generated feelings of serendipity under a high amount of attribute information, but enjoyment decreased. This means that sometimes feelings of serendipity do not translate to positive consumer outcomes.
The results are consistent with our theory, and also with the results of Botti and McGill **(**2006), which showed that people prefer having responsibility for pleasant personal choices, which increases enjoyment. Importantly, we obtained these results under conditions of different amounts of attribute information, but a similar amount of differentiation. This indicates that the quantity of information about options may be a critical determinant of consumer preference for personal choice.
## Main Results without Exclusion Criterion
The Table below presents all the key analyses form the main text, with and without the exclusion criterion.
++
| **Study D-1** |
+:=========================================:+:========================================================================================================================:+:========================================================================================================================:+:=======================================:+
| **Analysis** | **Excluding \>3SD P's** | **Not Excluding \>3SD P's** | **Any change?** |
| | | | |
| | **(N = 524)** | **(N = 537)** | |
+++++
| Enjoyment | The interaction was significant (F(1, 520) = 6.50, *p* = .011) | The interaction was significant (F(1, 533) = 6.14, *p* = .014) | No |
+++++
| Meaningfulness | The interaction was marginally significant (F(1, 520) = 2.77, *p* = .097) | The interaction was not significant (F(1, 533) = 2.35, *p* = .126) | Yes. Interaction became non-significant |
+++++
| Serendipity | The interaction was marginally significant (F(1, 520) = 3.31, *p* = .069) | The interaction was marginally significant (F(1, 533) = 3.89, *p* = .049) | Yes. Interaction became significant |
+++++
| Responsibility | Main effect of responsibility F(1, 520) = 86.17, p \< .001 | Main effect of responsibility F(1, 533) = 82.73, p \< .001 | No |
+++++
| Enjoyment Mediation (Serendipity) | Path when information was moderate: β = 2.02, SE = 1.23, 95% CI: \[.01 to 4.02\] | Path when information was moderate: β = 2.30, SE = 1.23, 95% CI: \[.27 to 4.36\] | No |
+++++
| Enjoyment Mediation (Responsibility) | Paths - moderate: β = -3.15, SE = .70, 90% CI: \[-4.37 to -2.08\]; high: β = -4.51, SE = .76, 90% CI: \[-5.81 to -3.31\] | Paths - moderate: β = -2.99, SE = .67, 90% CI: \[-4.14 to -1.97\]; high: β = -4.53, SE = .75, 90% CI: \[-5.81 to -3.39\] | No |
+++++
| Meaningfulness Mediation (Serendipity) | Path when information was moderate: β = .15, SE = .09, 95% CI: \[.01 to .29\] | Path when information was moderate: β = .17, SE = .09, 95% CI: \[.02 to .32\] | No |
+++++
| Meaningfulness Mediation (Responsibility) | Paths - moderate: β = -.33, SE = .06, 90% CI: \[-.42 to -.24\]; high: β = -.38, SE = .06, 90% CI: \[-.48 to -.28\] | Paths - moderate: β = -.32, SE = .06, 90% CI: \[-.42 to -.23\]; high: β = -.38, SE = .06, 90% CI: \[-.48 to -.29\] | No |
+++++
## Materials
Design: 2 (condition: personal choice vs. serendipity) × 2 (attribute information quantity: moderate vs. high) between-subjects design
**Instructions**
In this study, we are interested in how people respond in different situations.
Please put your headphones on or make sure that your speakers are working. When you are ready to begin please click the arrow.
**\[Personal Choice, Moderate Information Quantity Condition\]**
Please select one of these five songs.
**The call** by All Tvvins (2016, Electropop)
**How we be** by Sinkane (2014, Alternative/Indie)
**Waves** by Yonaka (2018, Alternative/Indie)
**Lay down** by DMA\'s (2016, Alternative/Indie)
**Next stop** by Bleached (2013, Alternative/Indie)
**\[Personal Choice, High Information Quantity Condition\]**
Please select one of these five songs.
**The call** by All Tvvins (2016, Electropop) - Overall rating: average - Uniqueness: high - Lyrics meaningfulness: low - Vocals: average
**How we be** by Sinkane (2014, Alternative/Indie) - Overall rating: average - Uniqueness: low - Lyrics meaningfulness: high - Vocals: average
**Waves** by Yonaka (2018, Alternative/Indie) - Overall rating: average - Uniqueness: low - Lyrics meaningfulness: average - Vocals: high
**Lay down** by DMA\'s (2016, Alternative/Indie) - Overall rating: average - Uniqueness: average - Lyrics meaningfulness: low - Vocals: high
**Next stop** by Bleached (2013, Alternative/Indie) - Overall rating: average - Uniqueness: high - Lyrics meaningfulness: average - Vocals: low
*Participants heard the song they chose.*
**The call** by All Tvvins (2016, Electropop)
\
\
**How we be** by Sinkane (2014, Alternative/Indie)
**Waves** by Yonaka (2018, Alternative/Indie)\
**Lay down** by DMA\'s (2016, Alternative/Indie)\
**Next stop** by Bleached (2013, Alternative/Indie)
**\[Serendipity, Moderate Information Quantity Condition\]**
Please see these five songs below.
**The call** by All Tvvins (2016, Electropop)\
\
**How we be** by Sinkane (2014, Alternative/Indie)\
\
**Waves** by Yonaka (2018, Alternative/Indie)\
\
**Lay down** by DMA\'s (2016, Alternative/Indie)\
\
**Next stop** by Bleached (2013, Alternative/Indie)
**\[Serendipity, High Information Quantity Condition\]**
Please see these five songs below.
**The call** by All Tvvins (2016, Electropop)\
- Overall rating: average\
- Uniqueness: high\
- Lyrics meaningfulness: low\
- Vocals: average\
\
**How we be** by Sinkane (2014, Alternative/Indie)\
- Overall rating: average\
- Uniqueness: low\
- Lyrics meaningfulness: high\
- Vocals: average
**Waves** by Yonaka (2018, Alternative/Indie)\
- Overall rating: average\
- Uniqueness: low\
- Lyrics meaningfulness: average\
- Vocals: high\
\
**Lay down** by DMA\'s (2016, Alternative/Indie)\
- Overall rating: average\
- Uniqueness: average\
- Lyrics meaningfulness: low\
- Vocals: high
**Next stop** by Bleached (2013, Alternative/Indie)\
- Overall rating: average\
- Uniqueness: high\
- Lyrics meaningfulness: average\
- Vocals: low
*Participants heard one of the songs.*
**The call** by All Tvvins (2016, Electropop)
\
\
**How we be** by Sinkane (2014, Alternative/Indie)
**Waves** by Yonaka (2018, Alternative/Indie)\
**Lay down** by DMA\'s (2016, Alternative/Indie)\
**Next stop** by Bleached (2013, Alternative/Indie)
How much did you enjoy the song? (0 = "I hated it," and 100 = "I loved it")
Please indicate your agreement with the following statements:
The way I came across this song made it seem more meaningful. (1 = "strongly disagree," and 7 = "strongly agree")
Please indicate your agreement with the following statements:
Getting to hear this one song I just heard was a good surprise. (1 = "strongly disagree," and 7 = "strongly agree")
Please indicate your agreement with the following statements:
I came across this song by luck. (1 = "strongly disagree," and 7 = "strongly agree")
Please indicate your agreement with the following statements:
This song was an unexpected discovery. (1 = "strongly disagree," and 7 = "strongly agree")
How responsible did you feel for the song you heard? (1 = "not at all," and 7 = "very much")
How accountable did you feel for the song you heard? (1 = "not at all," and 7 = "very much")
How much did you blame yourself for the song you heard? (1 = "not at all," and 7 = "very much")
How much did you congratulate yourself for the song you heard? (1 = "not at all," and 7 = "very much")
How stressed were you with the painting selection process? (1 = "not at all stressful," 7 = "very much stressful")
How frustrated were you with the painting selection process? (1 = "not at all frustrated," 7 = "very much frustrated")