Referral effectiveness : an empirical study of referral sources on investment decisions in reward-based crowdfunding

(1)

MASTER’S THESIS

Referral effectiveness: An empirical study of referral sources

on investment decisions in reward-based crowdfunding

Author: Ching Yu Wong Student number: 11251395 Submission date: 31 Jan 2018 Supervisor: Dr. Andrea Weihrauch

Programme: Executive Program in Management Studies - Strategic Marketing Management Institution: University of Amsterdam / Amsterdam Business School

(2)

Statement of Originality

This document is written by Student Ching Yu Wong who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Acknowledgements

Foremost, I would like to express my gratitude to my supervisor, Dr. Andrea Weihrauch, for her professional guidance and advice for my work in this thesis. I would also like to express my deepest gratitude to my parents for their everlasting love and encouragement; my other family members and close friends here and back home for their support.

(4)

Abstract

Crowdfunding is a growing trend to finance creative projects and bring new products to markets. It generates more capital than traditional funding methods. In order to take advantage of this new phenomenon, it is worthwhile to examine funders’ perspective in supporting a crowdfunding project. The goal of this study is to identify an additional factor that influences funders' investment decisions and advances knowledge in the referral literature by investigating the effect of online referral sources in the reward-based crowdfunding. A 3 (referral sources) × 2 (product types) online experiment was conducted with 245 participants in the globe. The results indicated that present or absent of online referral sources had a similar effect on funders’ investment decisions. The study also demonstrated how positioning a product as either utilitarian or hedonic can generate substantial differences in funders’ investment decisions, specifically the utilitarian showerhead. It led to a more favorable effect on funders’ investment decisions than the hedonic showerhead.

(5)

List of Figures and Tables

Figures

Figure 1 The actors in the reward-based crowdfunding and their relationships ... 7

Figure 2 Conceptual model ... 17

Figure 3 Post-hoc test: Investment decision per treatment group ... 36

Figure 4 Two-way ANOVA interaction effect of referral source and product type on investment decision ... 38

Figure 5 Two-way ANOVA main effect of product type on investment decision ... 41

Table Table 1 Demographic Information of Participants per Treatment Group ... 32

Table 2 Means, Standard Deviations, Correlations and Reliabilities ... 33

Table 3 Investment Decision per Treatment Group ... 36

Table 4 Two-way ANOVA Results of Investment Decision ... 37

(8)

1 Introduction

1.1 Popularity of Crowdfunding

Crowdfunding was nominated as one of the breakthrough technologies by MIT Technology Review (Greenwald, 2012). It is an alternative to traditional means, such as angel investment or venture capital, for technology startups to raise money. The 2015CF Crowdfunding Industry Report published by Massolution (2015a) stated that the total funding volume increased from $2.7 billion in 2012 to $34.4 billion in 2015. Crowdfunding already surpassed angel financing in 2015 and showed a trend to surpass venture capital in 2016 for the total funding volume. In the worldwide, crowdfunding markets have been growing. All major regions were recorded over 140% growth rate in 2014: North America 145%, Asia 320%, Europe 141%, South America 167% (Massolution, 2015b).

Crowdfunding provides a solution for startups to get funding in the early phase which usually is the most difficult period. Startups usually have more risks and a higher chance of failure that financial institutions and investors find unappealing to offer funding in the initial stage of these companies. A business with good ideas and prototypes may get the initial capital through crowdfunding to prove the concept’s worth and get the business off the ground. Project owners get in touch with a large number of people (“the crowd”) who invest in the projects and receive valuable feedback and comments on crowdfunding platforms. It builds up a group of supporters and customers before the business even produces a product. Project owners not only get funding from the crowd but also win the hearts and minds of backing individuals. One of the most successful crowdfunding projects was Oculus Rift, the virtual-reality gaming headset that raised over 2 million US dollars in 2014 (Stanko & Henard, 2016). Another successful crowdfunding project was Fairphone2, the ethical smartphone that raised 9 million euros from the crowd before the company manufactured the

(9)

phones in 2013. From an economic point of view, crowdfunding is beneficial to increase financial acquisition speed and volume, fill the need of niches and enable global reach in attracting funding support (Pazowski & Czudec, 2014).

1.2 Problem Definition

A great deal of academic research has been conducted on crowdfunding from different perspectives: Ordanini, Miceli, Pizzetti, and Parasuraman (2011) studied some common motivations of funders to participate in crowdfunding. These motivations included engaging in innovative projects, interaction with other funders on platforms, and funders’ strong sense of identification with a product and/or proponent of a campaign; Gerber, Hui, and Kuo (2012) conducted interviews with project owners. The study identified major reasons for project owners to seek capital from the crowd, such as attracting public attention, and forming social networks and relationships; Mollick (2014) studied data of the crowdfunding platform, Kickstarter, to analyse the underlying success and failure factors for crowdfunding. But there is a lack of empirical studies because a majority of scholars mainly focused on examining the factors of project owners and funders participating in crowdfunding using an exploratory approach, such as interviews and case study (Gerber et al., 2012; Mollick, 2014; Ordanini et al., 2011; Pazowski & Czudec, 2014). More new insights should be provided in this emerging research field.

One aspect that has not been looked at extensively in the crowdfunding context is the role of referrals. Previous research has shown that referrals play a relevant role in consumer decision-making. Referrals are external influence affecting consumer decision-making from the information search phase to the purchase phase. Many marketing scholars have been examining various influencers on consumer behaviour since the late 1960s. Social influence was one of the external influences receiving attentions in the academic fields (Blau, 1964;

(10)

Evans, Jamal, & Foxall, 2009; Goodwin, 1987; McPherson, Smith-Lovin, & Cook, 2001; Rosen & Olshavsky, 1987). Referrals can be seemed as human-based and system-based in daily life. On one hand, there are celebrity-endorsed sportswear brands (e.g. Nike and Adidas), and friends’ recommendations of food and clothes on social media (e.g. Facebook). On the other hand, consumers readily reply on recommender systems. They use advice given by recommender systems to book hotels (e.g. Priceline.com) and flights (e.g. Skyscanner), search movies (e.g. Netflix) and look for jobs (e.g. LinkedIn) etc. Algorithmic advice has integrated into our everyday decisions and has started being relevant in financial investments, apparently, where robo-advisors have been replacing human advisors (The Economist, 2015). Consumers could like algorithms more because system-based investment advice is instant and less biased. It helps investors keep emotion out of the investment decision. Crowd-funders were characterized as tech-adapting crowd and innovators in the way they interacted with others on crowdfunding platforms (Ordanini et al., 2011). They could like computer-generated advice. It would be interesting to compare whether crowd-funders are different from more social-oriented consumers. It is relevant and particularly interesting to test the influence of two different referral sources in the context of crowdfunding.

1.3 Research Question

The purpose of this thesis is to combine two previously mentioned research streams and has two goals: 1) it identifies an additional factor that influences funders’ decisions, and 2) it advances knowledge in the referral literature by testing the effect of different referral sources in the context of crowdfunding. No previous research has yet examined funders’ behaviour towards referrals in reward-based crowdfunding. This study measures the effectiveness of referral sources on funders’ investment decisions in an experiment together with quantitative analysis. In addition, the researcher would adopt persuasion theory to

(11)

explore the moderating role of a campaign product. My primary research question is: “What is the effect of referral sources on funders’ investment decisions in reward-based crowdfunding?”

1.4 Delimitations

The research mainly focuses on funders’ investment decisions towards a reward-based crowdfunding project when referrals take place instead of project awareness and recognition. The project credibility and origin are not considered and measured in this research. The pitch video was taken out to fit the experiment. The content of a crowdfunding project was edited and shortened which was different from real full-length projects. Furthermore, the reward tiers and limited offers are not considered in this study. The crowdfunding project was modified to have only one reward tier and there was no scarcity of rewards.

1.5 Structure of Research

The remainder of the study is structured as follows. The next chapter provides a review of relevant literature related to the research topic, which includes crowdfunding and referral behaviour, to establish the conceptual model and hypotheses. Chapter three focuses on the methodology and research design. Subsequently, chapter four described the empirical result based on the collected data. A comprehensive discussion, implications, limitations and suggestions for future work are presented in chapter five. Finally, the conclusion is provided in chapter six.

(12)

2 Literature Review

This chapter discusses the relevant literature of crowdfunding and the role of referrals in consumer behaviour. This chapter is structured as follows. Initially, the key concept of crowdfunding and reward-based crowdfunding is explained and explored. The chapter continues to present the theoretical framework of referral sources guiding this empirical study. The researcher then outlines how the type of product moderates the influence between referral sources and funders’ investment decisions. An overall conceptual model which graphically illustrates the hypotheses is displayed at the end of the chapter.

2.1 Crowdfunding and Reward-based Crowdfunding

The concept of crowdfunding originated from a broader concept of crowdsourcing, which referred to obtain ideas, comments, and finances from the general public to develop goods and services in the form of open call (Bayus, 2013; Howe, 2008; Kleemann, VoB, & Rieder, 2008). Crowdfunding was a form of alternative finance for small ventures to get a project funded which they were unlikely to get from traditional means (Greenwald, 2012; Massolution, 2015a). The objective of crowdfunding is to collect small amounts of money from large audiences. The crowdfunding on the internet has often been involved in personal initiatives in the arts and music communities in the late 1990s (Agrawal, Catalini, & Goldfarb, 2014). For example, the music brand Marillion raised donations from fans to create albums in a studio. The crowdfunding model has become matured onwards. There are now crowdfunding platforms like IndieGoGo and Kickstarter where people can fund a diverse range of creative projects, such as fashion, comics, games, technology and design.

(13)

In crowdfunding, there were typically three participating actors. They are project owners (e.g. capital-seeking startups, or individuals with specific goals), funders (e.g. investors, supporters, lenders, or donors) and crowdfunding platforms as an intermediary (De Buysere, Gajda, Kleverlaan, & Marom, 2012). Project owners are those seeking funding from the crowd on the internet for the provision of resources; funders are those willing to support specific projects for purposes; crowdfunding platforms are mostly web-based to act as neutral facilitators between project owners and funders. Crowdfunding could be classified into equity-based, loan-based, donation-based, or reward-based (Agrawal et al., 2014; Hemer, 2011; Kuppuswamy & Bayus, 2013; Pazowski & Czudec, 2014) depended on the type of funders’ return (Gerber et al, 2012; Ordanini et al., 2011). In equity-based crowdfunding and loans, project owners offer funders monetary returns like interest and shares. In donation-based crowdfunding, funders usually do not want anything else in return. They normally support charitable projects for “greater good” and altruistic reasons. In reward-based crowdfunding, individuals invest projects to receive rewards. The rewards were usually tangible products or services, such as a music album, a thank-you postcard, and a meet up with the project owners (Gerber et al., 2012; Kuppuswamy & Bayus, 2013). Figure 1 illustrates the actors in the reward-based crowdfunding and their relationships.

(14)

Figure 1 The actors in the reward-based crowdfunding and their relationships

The reward-based crowdfunding platforms had two different business models, either the “all or nothing” model or “keep it all” model (Gerber et al., 2012). The “keep it all” model is that the project owners keep the funded capital even though the funding goal is not achieved. The “all or nothing” model is also called the “threshold pledge” model (Hemer, 2011):

its main characteristic is that the platform and the project initiator agree on a concrete pledging period (between two weeks and several months) and a so-called threshold, a targeted sum of money that must be reached via the contributions of the backers or crowdfunders before any financial transaction is generated. Below this threshold, there is no flow of funds (p.15).

In other words, project owners only keep the funded capital when the funding goal is reached within the agreed period, otherwise, the funds are returned to the funders. Kickstarter, being one of the most dominant reward-based crowdfunding platforms with several million community members, which uses “all or nothing” model.

(15)

The reward-based model is the most prevalent type of crowdfunding nowadays. The research is interested in studying funders’ behaviour in the reward-based crowdfunding. When funders consider a reward-based crowdfunding project, they are making investment decisions to pledge material rewards (goods and/or services), which are similar to traditional customers who make purchase decisions (Hemer, 2011; Mollick, 2014). In this situation, funders can be affected by external influences, such as culture and social factors, in making purchase decisions. Persuasion theory can, therefore, be implemented to explore funders’ investment behaviour.

2.2 Referral Behavior

In this section, the definitions and the role of referrals in consumer decision making are first presented. Then, the researcher elaborates on peer and computer-generated recommendations and posits that the two online referral forms will have different degrees of influence on a funder investment decision.

2.2.1 Defining of referrals

Senecal and Nantel (2004) proposed the following typology of referral sources: “1) Personal source providing personalized information; 2) Personal source providing non-personalized information; 3) Impersonal source providing non-personalized information; 4) Impersonal source providing non-personalized information” (p. 160). Based on the typology, personal influence (e.g. friends and family), social influence (e.g. peer groups and community) and Word-of-Mouth (WOM) communications in consumer behaviour research, can be grouped as studies exploring personal sources providing personalized or non-personalized information. WOM communications were defined as “oral, person-to-person communication between a perceived non-commercial communicator and a receiver

(16)

concerning a brand, a product, or a service offered for sale” (Arndt, 1967, p. 3). With the advent of the internet, electronic word of mouth (eWOM) further provided insights into how consumers used online information to complement their purchase decision process through the use of online rating (e.g. tripadvisor.com), social network sites and online discussion forums (East, Wright, & Vanhuele, 2008; Sparks & Browning, 2011). Ansari, Essegaier and Kohli (2000) suggested that recommender systems were impersonal sources making personalized recommendations to consumers because most recommender systems used some filtering algorithms, such as content-based filtering or collaborative filtering, to make product suggestions to consumers. On the contrary, robo-advisors were digital tools providing financial and wealth management advice based on mathematical rules or algorithms to investors (The Economist, 2015). It can be classified as an impersonal source providing non-personalized information.

2.2.2 The role of referrals in consumer decision making

Scholars have studied the use and influence of referrals on consumers, while predominantly referral sources used in consumer decision are personal sources (Arndt, 1967; Gilly, Graham, Wolfinbarger, & Yale, 1998; Price & Feick, 1984; Senecal & Nantel, 2004). For instance, Gilly et al. (1998) has shown that interpersonal WOM communication influenced consumers positively during the information search phase. In line with this study, personal sources influenced a consumer’s expectations and perceptions during the information search phase, and influenced the evaluation and selection of product choices or service providers (Price & Feick, 1984; Senecal & Nantel, 2004; Schmidt & Spreng, 1996). Personal social relations could be characterized by relationship ties from strong primary ties, such as close friends and parents, to weak secondary ties, such as relative and acquaintances (Brown & Reingen, 1987). Even though literature demonstrated that strong-tie referral

(17)

sources were perceived to be more influential than weak-tie referral sources by people in making decisions (Brown & Reingen, 1987; Rogers, 1983; Weimann, 1983), it was suggested that consumers routinely used both referral sources in a given decision situation (Duhan, Johnson, Wilcox, & Harrell, 1997). Some scholars also revealed that consumers adopted behaviour similar to those of others because of the fulfilment of human aspirations and reduction of perceived risk (Bearden & Etzel, 1982; Bearden, Netemeyer, & Teel, 1989). Much is known in consumer behaviour literature about the likeliness of consumers to use referrals in their decision-making processes. However, in the context of crowdfunding, it is barely known about how online referrals impact funders’ investment decisions.

2.2.3 Peer recommendation

Social influence has been found to have impact on consumer purchase decision because consumers took decisions with the environment around them such as peers, opinion leaders and reference groups (Evans et al., 2009; Rosen & Olshavsky, 1987). Peer recommendation influences consumer decision making because consumers are motivated to buy or use products in order to be consistent with others. Solomon, Bamossy, Askegaard and Hogg (2006) discovered two reasons for conformity: 1) people modelled their behaviour and actions of others in endeavour to show proper behaviour in a given situation because of social proof, and 2) people conformed to satisfy the expectations of others or to be accepted or liked by the group. Similarly, Sheth and Parvatiyar (1995) discussed that consumers complied with the interests of other members of social groups because consumers inherently liked to avoid conflict, attain closer relationship and resort to more cooperative behaviour. Researchers have also revealed that online consumers depended on virtual communities or other consumers’ reviews and ratings to make their own purchase decisions (Evans et al., 2009; Sparks & Browning, 2011). Individuals were inclined to affiliate with others who had similar interests

(18)

or were in similar situations (Schachter, 1959). According to Gilly et al. (1998) and v. Wangenheim and Bayon (2004), similarity was the degree to which people were alike in terms of some attributes, for instance, demography, lifestyle, and perceptual believes and values. In light of research of eWOM influence on purchase intentions, people with similar attributes are likely to have similar preference and needs. As a consequence, recommendation regarding the product information from similar source was perceived as more relevant and would exert more influence (Gilly et al., 1998). In the context of crowdfunding, studies have demonstrated that funders were motivated to fund campaigns because the behaviour was engaging and contributing to a trustful and creative community with similar interests (Gerber et al., 2012; Gerber & Hui, 2013). Based on the above discussion, the researcher believes that funders will favour recommendation by another funder to support a project, thus, the following hypothesis is formulated:

H1a: Peer recommendation has a more positive effect on a funder investment decision than

no recommendation.

2.2.4 Computer-generated recommendation

Algorithms are some defined procedures that allow computers to solve a problem. They have been used to assist human judgement and predict objective outcomes, such as clinical predications (Grove, Zald, Lebow, Snitz, & Nelson, 2000). The use of algorithms has been expanded to predict subjective tastes such as predicting which movies, music and books people would enjoy (Adomavicius & Tuzhilin, 2005; Resnick & Varian, 1997). Recommender systems, which leverage the power of algorithms, have become an important research area since the late 1990s (Ansari et al., 2000; Maes, 1999; Resnick & Varian, 1997). Ricci, Rokach, Shapira, and Kantor (2011) described recommender systems as software making recommendations for items to a person. The suggestions given by recommender

(19)

systems are intended to support consumers in decision making processes. The predictions of recommender systems are usually generated by some analytical filtering techniques. There were two widely used filtering techniques: content-based filtering algorithms and collaborative filtering algorithms (Resnick & Varian,1997; Ricci et al., 2011). The former technique recommends items similar to the ones the consumer preferred in the past. The latter predicts which products will interest a user by making correlations in purchase behaviour between people with similar tastes and preferences in databases. Recommender systems using any of these techniques are making personalized recommendations which mean different users receive diverse suggestions.

There were also recommender systems providing non-personalized recommendations (Poriya, Bhagat, Patel, & Sharma, 2014). These are much simpler to generate recommendations with various kinds of factual data, but in the absence of users’ personal preference information. The core function of a recommender system providing non-personalized recommendations was to predict that an item is worth recommending (Ricci et al., 2011). These systems are capable to predict the utility of some items, or compare the utility of some items, and based on this comparison to decide what items to recommend (Adomavicius, Sankaranarayanan, Sen, & Tuzhilin, 2005; Ricci et al., 2011). Take an example, a non-personalized recommender system that suggests the most popular book. The ground behinds this method is that a popular book, i.e. being favorited by many readers (high utility), will also be possibly liked by a reader, at least somewhat more than another randomly selected book.

Although the helpfulness of recommender systems, people were reluctant to allow an algorithm to make decisions for them (Dawes, 1979). It is argued that recommender systems are now widespread in practice, we are still in the habit of receiving recommendations from other people. According to Nielsen Global Trust in Advertising Report 2015, people trusted

(20)

other people to make good recommendations; over 80 percent of people trusted recommendations from friends and family; while over 60% trusted consumer opinions posted online. Whether deciding which restaurant to go, what film to watch or what book to read, people depended on the opinions of friends, family, and even strangers on the internet (Sinha & Swearingen, 2001; Yeomans, Shah, Mullainathan, & Kleinberg, 2017). Recommender systems are not humans. Humans have feelings and emotions, and they can express emotions after watching a movie, or tasting the food at a restaurant. Recommender systems are taught to know what we like, not why we like a thing. In line with existing crowdfunding studies, funders are more likely to make contributions because of social information, such as funding decisions of other people and be part of a community (Gerber & Hui, 2013; Kuppuswamy & Bayus, 2013). Since crowdfunding is a social phenomenon of crowd support and funding, it is expected that funders rely more on peer advice and recommendation. Reversely, funders rely less on the algorithmic advice of recommender systems. To summarize the arguments, the researcher proposes another hypothesis:

H1b: Peer recommendation has a more positive effect on a funder investment decision than

computer-generated recommendation.

2.3 Utilitarian and Hedonic Product and Campaigns

In this section, the researcher presents the definitions of products and elaborates on a persuasion theory to posit the moderating role of product type on funders’ investment decisions.

(21)

2.3.1 Defining of products

Products can be distinguished along various dimensions. Previous literature used product characteristics to make product classification, such as product form (goods versus services e.g. Parasuraman, Zeithaml, & Berry, 1985; Verhagen, Boter, & Adelaar, 2010), product evaluation feature (search goods versus experience goods, e.g. Senecal & Nantel, 2004) and product function (utilitarian goods versus hedonic goods, e.g. Bazerman, Tenbrunsel, & Wade-Benzoni, 1998; Okada, 2005). Among the classifications, product function seems the most promising because this specific typology was predominantly used as product classification in previous research (Bazerman et al., 1998; Dhar & Wertenbroch, 2000; Hirschman & Holbrook, 1982a, 1982b; Okada, 2005; Verhagen et al., 2010).

Product function classifies products into two types: a utilitarian product or a hedonic product. The difference between these two product types lies in the function (Bazerman et al., 1998; Verhagen et al., 2010). Verhagen et al. (2010) described that utilitarian products fulfilled merely instrumental functions, and product features objectively associated with the product utility; while hedonic products triggered sensory stimulation, fantasies and fun for evoking pleasure. Existing literature has revealed that the distinctive functions of hedonic and utilitarian products had essential implications for consumer online decision-making because people value utilitarian and hedonic products in a different way. For instance, Noble, Griffith and Weinberger (2005) argued that consumers searched for information and made price comparisons for utilitarian products which thus influenced their purchase behaviour. To, Liao and Lin (2007) suggested consumers evaluate hedonic products based on the affective outcomes rather than cognitive processes. Hedonic products were something novel and interesting. They triggered pleasure and enjoyment valued by consumers and stimulated the purchase intention of consumers.

(22)

2.3.2 Persuasion routes for utilitarian and hedonic products

The elaboration likelihood model (ELM) is the major theoretical model applied in consumer behaviour and marketing research. ELM is an information dual-processes model which proposes two separate routes in information processing (Petty & Cacioppo, 1986a): analytical information tends to trigger the central route of information processing, emotional information tends to trigger the peripheral route of information processing. Verhagen et al. (2010) argued that the underlying functional differences between hedonic and utilitarian products influenced the use of external sources. The information about argument quality and specifications influenced the central route, and the eWOM cue influenced the peripheral route (Cheng & Ho, 2015).

Based on the ELM model, the researcher could explain individuals favour different products under different persuasion. The central route to persuasion occurs through deliberate consideration of product information. Utilitarian products fulfill instrumental, function-based needs in which consumers process information through the central route (Hirschman & Holbrook, 1982a; Petty & Cacioppo, 1986b). During an evaluation of utilitarian products, consumers normally engage in high rational and low affective response. Thus, a non-personalized recommender system, given its computational ability and goal orientation, appears a better fit for utilitarian products to pull out rational response and triggers central route processing. Hedonic products evoke pleasure and fulfil emotional appeals in which consumers process information through the peripheral route (Batra & Ahtola, 1991; Baumeister, 2002; Bruyneel, Dewitte, Vohs, & Warlop, 2006). During an evaluation of hedonic products, consumers paid attention to emotional contents and cues enhancing excitement (Buck, Anderson, Chaudhuri, & Ray, 2004; Hirschman & Holbrook, 1982a; Mitchell & Olson, 1981). In other words, consumers normally engage in low rational and high affective response. A peer recommendation, given its personal tastes with similar

(23)

interest and value perception, appears a better match for hedonic products to pull out affective response and trigger peripheral route processing.

Despite this reasoning, no study has looked into the combined effects of product types and referral sources in the context of crowdfunding. In light of research on the consumer decision-making processes underlying product types and in consideration of congruity between referral sources and product types, it is expected that product types moderate the effect of referral sources on funders’ investment decisions. Thus, the following hypotheses are formulated:

H2: Campaign product types (hedonic product versus utilitarian product) will moderate the effect of referral sources on funders’ investment decisions.

Specifically,

H2a: Peer recommendation for a hedonic product has a more positive effect on a funder

investment decision than computer-generated recommendation for a hedonic product.

H2b: Computer-generated recommendation for a utilitarian product has a more positive

effect on a funder investment decision than peer recommendation for a utilitarian product.

2.4 Conceptual Model

Five sets of hypotheses were established in previous sections. The first two hypotheses represent the main model of this research and form the basis for other hypotheses. The main model refers to the direct relationship between referral sources and a funder investment decision. This relationship is represented by the bold arrow and boxes in figure 1.

(24)

The expected influence of campaign products has been graphically illustrated in the normal box in figure 1.

(25)

3 Data and Method

This chapter provides an overview of the data and methodology used in this research. The first part of this chapter outlines the general approach of this study. The chapter continues to describe the procedure of the experiment and the development of stimuli, followed by the flow of the survey. The chapter then presents the measurements of variables. A brief description of statistical approach to test the expected relationships discussed in the previous chapter is provided at the end.

3.1 Research Design

This research was quantitative research. An online experiment with a survey was used to conduct the research to provide a direct measurement of the hypotheses. An online study is a time-efficient way to collect a large amount of data that will be accessible for numerical analysis (Saunders, Lewis, & Thornhill, 2009). The study employed a 3 (referral source: peer recommendation/ computer-generated recommendation/ no recommendation) x 2 (product type: utilitarian/ hedonic) between-subjects design to test the proposed hypotheses. Six different treatments have been developed. Four out of six treatments were experimental groups in which a referral source was presented in the campaigns. The other two treatments in which no referral source was shown in the campaigns, served as control groups. The first factor referral source was the primary interest in the study, whereas the second factor represented the product type in a crowdfunding project. The “peer recommendation” condition corresponded to a recommendation given from another funder in the community. The “computer-generated recommendation” condition corresponded to a recommendation generated by a non-personalized recommender system. In other words, there was only one

(26)

independent variable (referral source) and one dependent variable (investment decision). The moderator variable was a campaign product. Other variables that have been taken into account as control variables were gender, age, campaign funding period, campaign funding goal, campaign price and campaign tier.

3.2 Sample

The research was conducted using non-probability convenience sampling, because the population was huge and there was no access to sampling frame. Using the convenience sampling technique was affordable and available at a given time while researchers got a fast and reliable access to participants (Etikan, Musa, & Alkassim, 2016). A minimum of 30 participants per treatment group, and in a total of 180 participants were needed to obtain a minimum reliable result in the study (Roscoe, 1975). The participants were recruited through emails and social network sites. The survey was distributed among family, friends and colleagues. In the end, a total of 331 people participated in the experiment. The usable data for analysis reduced to 245 participants after filtering out incomplete surveys, thus the criterion for Roscoe (1975) was met.

3.3 Procedure and Materials

3.3.1 Data collection overview

The data collection was done through Qualtrics which is a software tool to create and distribute online surveys. Participants could complete a portion of a survey and return later to finish the rest. Participants could also take a survey at their preferred time and location. This research used Qualtrics to run the factorial vignette survey, a way of combining traditional survey methods with experimental methods. Survey distribution started on 14th Nov 2017

(27)

and ended 3 weeks later on 4th Dec 2017. All information in the survey was written in English. In the survey, the participants were randomly exposed to the crowdfunding project in one of the six scenarios: 1) peer recommendation for a utilitarian product, 2) computer-generated recommendation for a utilitarian product, 3) peer recommendation for a hedonic product, 4) computer-generated recommendation for a hedonic product, 5) no recommendation for a utilitarian product (control scenario) and 6) no recommendation for a hedonic product (control scenario). The experiment involved manipulations of the independent variable and the moderator variable while maintaining control over other variables. Instead of two different projects for the utilitarian and hedonic products, one crowdfunding project was selected and modified by varying the contents to position the product as utilitarian or hedonic. It was because there could be effects, potentially confounding variables, inherent in two different projects selected, beyond classification as utilitarian or hedonic.

3.3.2 Selection of a reward-based crowdfunding platform

The survey presented one project in a reward-based crowdfunding platform across all six treatment groups. To enhance the external validity of research, the platform should be real and existing. The platform itself should also be internationally well-known and had a similar level of familiarity towards participants. Kickstarter, being one of the most famous and largest reward-based crowdfunding platforms, was chosen in this study. The latest information from Kickstarter (https://www.kickstarter.com/help/stats) revealed that the platform attracted over 13 million funders around the globe. With the view to enhance realism and familiarity, the Kickstarter platform was picked for this research.

(28)

3.3.3 Selection of a crowdfunding project

The survey presented one campaign product across all six treatment groups. To ensure that the participants were mainly influenced by the manipulation but no other factors, the product should fall into an existing and common product category. The familiarity of the project owner should be low so that no participants hold prior knowledge in the campaign product. The product should also be a unisex category, which means it was suitable for both genders and all age groups. To meet all criteria, the Kickstarter project from “Methven Rua” which featured a showerhead product, was picked for this research. “Methven Rua” was the first-time project owner appearing on Kickstarter, and no participants should recognize the brand.

Pre-test

In order to achieve an effective manipulation in the main study, a qualitative pre-test was carried out through a focus group. The focus group consisted of twelve people who had the same characteristics of the participants used for the main experiment. They were asked to identify an environmental-friendly showerhead or a spa-like showerhead, then to classify them to either a primary utilitarian product or a primary hedonic product. They were told a utilitarian product was seen as superior on utilitarian dimensions (e.g. useful, practical) and a hedonic product was seen as superior on hedonic dimensions (e.g. fun, pleasant). Based on the judgement of the focus group, the environmental-friendly showerhead was the closest to the utilitarian end, while the spa-like showerhead was the closest to the hedonic end on a utilitarian-hedonic continuum.

(29)

3.3.4 Materials

Stimulus materials were needed to develop in order to set up the survey. Two projects with referral recommendations were used in experimental groups, and two projects without recommendation were used in control groups. The project funding period, funding goal, reward tiers and price, as well as page layout were held constant across different treatment groups. See the appendix A for the crowdfunding project per treatment group. Below paragraphs outlined the development of two stimuli in this research.

The referral sources

The manipulation was adopted and modified using the recommendation source manipulation described by Senecal and Nantel (2004). Three treatment levels were used for the manipulation: peer recommendation, computer-generated recommendation and no recommendation. During the experiment, if participants were assigned to a referral source, they were exposed to a recommended crowdfunding project. For the “peer recommendation” treatment level, the recommendation of the campaign was described as follows.

“Rob, another funder, highly recommends this campaign to you. He comments "Hello! I find this crowdfunding campaign super interesting! It is an awesome product and you should pledge this product over the others."

For the “computer-generated recommendation” treatment level, the recommendation of the campaign was described as follows.

“Rob, the computer algorithm-based robot, highly recommends this campaign to you. It comments "Based on the computer analysis of massive data from the trend of campaign performance, popularity to campaign creator historical figures etc., you should pledge this product over the others."

(30)

Both referral sources had identical font and position (except the referral icon) on the project in order to attribute differing effects to the manipulation. Participants assigned to the “no recommendation” treatment did not read a recommendation source, i.e. no referral source was present in the campaign.

The campaign product types

The campaign products were manipulated by using one product category with different descriptions. They were chosen with the help of the pre-test mentioned in the previous section. The utilitarian product represented in the campaign was an environmental-friendly showerhead. Specially, the project was titled “Methven Rua - Power shower experience with 28% less water”. The subsequence descriptions of the product were about the water-saving function and the simple installation. The descriptions were written by the project creator. The researcher shortened and modified the information to fit the experiment. The hedonic product represented in the campaign was a spa-like showerhead. In particular, the project was titled “Methven Rua - Power shower experience bringing luxurious sensations”. The subsequence descriptions of the product were about the shower experience like a tropical rainforest waterfall bringing pleasant sensations. The researcher wrote the descriptions to fit the experiment. See Appendix A for the full descriptions of the products.

3.3.5 Flow of the survey

The survey consisted 22 questions. There were 14 Likert-scale questions, 6 multiple-choice questions, and 2 open-ended questions. All Likert-scale questions were phased in statements and participants indicated the answers on 7-point scales.

Participants first read the introduction message containing information about the experiment and the purpose of the survey. Upon agreeing to the informed consent, the participants were randomly assigned to read one of six crowdfunding projects that varied in

(31)

terms of referral sources and campaign product types. Subsequently, a peer recommendation or computer-generated recommendation was presented depending on the experimental group assigned. For control groups, no page for a referral recommendation was presented. After reading the crowdfunding project, the participants started to answer questions measuring investment decision. The first three questions were 3 items on 7-point scales on investment decision. The next question was a dichotomous question on investment decision, “If you had to make the choice of either funding, or not funding the campaign, what would you do? (Yes/No)”, followed by an optional open-ended question “Please elaborate why you choose to (not) support the campaign?” in which participants could provide opinion. Respondents then asked a 7-point scale question on whether they are influenced by the referral source from (1=) very much to (7=) not at all. The next part of the questionnaire was 10 items on 7-point scales to check whether the product manipulation served its purpose in the experiment. The final part of the survey contained six questions regarding participants’ demographic information, including Kickstarter familiarity, crowdfunding backgrounds, age, gender and place of residence. Lastly, participants were debriefed and again thanked for taking part in the survey.

3.4 Measures

The measures in this research were participants’ self-reported responses to the items on 7-point Likert scales. There were a couple of exceptions without using Likert scale, such as providing demographic information and the optional open-ended question.

3.4.1 Dependent variable

The dependent variable was a participant’s investment decision for the reward-based crowdfunding campaign. The investment decision was measured by a participant’s

(32)

willingness to pledge the product. While crowdfunding is a new field, it does not exist validated scales for a funder investment decision. Based on the discussion of relevant literature in chapter 2, when funders consider pledging a product in reward-based crowdfunding, their behaviours are similar to traditional customers who make purchase decisions. Much research was available about willingness-to-buy (Dodds, Monroe, & Grewal, 1991). The scale's reliability is high with Cronbach’s alpha (a) more than or equal to .80. Following Dodds, Monroe and Grewal (1991), three out of five items were adapted and modified for measuring the investment decision in this research. The remaining two items about the purchase price were dropped because the study did not focus on the product price. The items were measured on existing and validated 7-point Likert scales, ranging from (1=) very low to (7=) very high. The items of this study were:

1. The likelihood of funding this product is

2. The probability that I would consider funding the product is 3. My willingness to fund the product is

3.4.2 Moderator variable

The moderator variable was the campaign product type. The scales used in the study were adapted and measured following Crowley, Spangenberg, and Hughes (1992) and Voss, Spangenberg and Grohmann (2003). The scales’ reliability is high among scale scores for measuring hedonic and utilitarian dimensions. The five items on 7-point Likert scales of the hedonic dimension were:

(1=) fun ---- (7=) not fun, (1=) exciting ---- (7=) dull,

(1=) delightful ---- (7=) not delightful, (1=) thrilling ---- (7=) not thrilling,

(33)

(1=) enjoyable ---- (7=) not enjoyable

The five items on 7-point Likert scales of the utilitarian dimension were: (1=) effective ---- (7=) ineffective,

(1=) helpful ---- (7=) unhelpful,

(1=) functional ---- (7=) not functional, (1=) necessary ---- (7=) unnecessary, (1=) practical ---- (7=) impractical

3.4.3 Control variables

Demographics such as age, gender, country of residence and crowdfunding background were measured at the end of the survey. Furthermore, this research controlled the campaign funding period, funding goal, reward tiers and price because these could influence outcomes. For example, Lin, Lee and Chang (2016) found that more reward tiers, high price-goal ratio and limited offerings were positive correlated with funder’s support to the project. See appendix B for the detailed demographic questions.

3.5 Statistical Method

The Statistical Software Package for Social Sciences (SPSS) was used in this study to perform the statistical analysis. This study was interested in the direct effect and the interaction effect between variables, which is the 3 x 2 between-subjects factorial design. Therefore, two-way analysis of variance (ANOVA) was undertaken to test the proposed hypotheses. Before testing the hypotheses, several preliminary steps were done. At first, data screening, skewness, kurtosis and normality checks were performed. The researcher then computed scale means and performed manipulation check. Finally, the data was analysed to test the hypotheses. The partial eta-squared (η2_{) of the ANOVA results as described by}

(34)

Cohen (1988) was used to quantify the association between the dependent variable and the independent variable. Confidence intervals were set on a 95% interval. An alpha level of .05 was used in all statistical procedures.

(35)

4 Results

This chapter describes the results of statistical analyses. The first part focuses on the summarization of frequencies and descriptive statistics, followed by reliability and normality checks. The second parts of the chapter discuss the manipulation check, hypotheses testing and an additional analysis.

4.1 Descriptive and Frequencies Statistics

This research employed a 3 x 2 between-subjects factorial design. The survey was distributed among family, friends, colleague and social network sites. A total of 331 people participated in the experiment. After filtering out participants who did not complete the questionnaire and dropped out in the questionnaire, the usable responses contained 245 participants. The mean age of the participants was 33.45 years old (SD = 9.86), more than half of the participants were between 28 and 33 years old. A total of 95 males (38.8%), 145 females (59.2%) and 5 others (2%) participated in the survey. Almost half of the participants lived in Asia (47.3%) and almost half of the participants lived in Europe (47.3%). It is followed by America (3.3 %), Africa (1.6%) and Australia (0.4%). Regarding the crowdfunding history of the participants, 108 participants (44%) thought to fund a crowdfunding project and 74 participants (30%) have funded in a crowdfunding project at least one in the last six months.

Based on the experimental design, six treatment groups were created. There were 40-41 participants for each group. The number of participants per treatment group was randomized by the survey software, Qualtrics. To check the distribution between treatment groups for gender, age and residency, likelihood Chi-square tests were conducted. The

(36)

distribution between gender and treatment groups differed insignificantly (X2 _{= 13.73, p}

= .186). The distribution between age and treatment groups also differed insignificantly (X2 ₌

191.21, p = .462). The distribution between residency and treatment groups differed insignificantly as well (X2 _{= 30.77, p = .197). The randomization was satisfactory. Table 1}

illustrates the demographic information of participants per treatment group.

4.2 Reliability Check

A reliability analysis was performed to check internal consistency for questions that measuring the same variables. In this research, there were multiple questions measuring the same variables. These variables were the investment decision, the utilitarian dimension of the product and the hedonic dimension of the product. Cronbach’s Alpha (a) was used to check the internal consistency of these scales (Tavakol & Dennick, 2011; Gliem & Gliem, 2003). For each item of the investment decision and the product attribute, Cronbach’s Alpha (a) should be greater than or equal to .70, the “corrected item-total correlation” should be greater than .30, the difference between Cronbach’s alpha and “Cronbach’s alpha if item deleted” should be greater than .10 (Gliem & Gliem, 2003) to conclude a reasonably reliable scale.

The investment decision

Three questions were made for the evaluation of a funder’s investment decision: INV1 “The likelihood of funding this product is”, INV2 “The probability that I would consider funding the product is” and INV3 “My willingness to fund the product is”. Participants had to indicate each question on a 7-point Likert scale, from (1=) very low to (7=) very high. From the analysis, the Cronbach’s Alpha (a) was .897 for the three items of the investment decision scale which indicated a high reliability for the scale. The corrected item-total correlations were all above .30, and none of the items would substantially affect

(37)

reliability if they were deleted. In the following analysis, a new variable INV_TOT was used to the capture the average investment decision instead of three items in the investment decision scale.

The product attributes

Five questions were made for an evaluation of the hedonic dimension of the product: HED1 “Fun”, HED2 “Exciting”, HED3 “Delightful”, HED4 “Thrilling” and HED5 “Enjoyable”. Participants had to indicate each question on a 7-point Likert scale, from (1=) fun/ exciting/ delightful/ thrilling/ enjoyable to (7=) not fun/ dull/ not delightful/ not thrilling/ not enjoyable. From the analysis, the Cronbach’s Alpha (a) was .896 for the five items of the hedonic dimension scale which indicated a high reliability for the scale. The corrected item-total correlations were all above .30, and none of the items would substantially affect reliability if they were deleted. In the following analysis, a new variable HED_TOT was used to the capture the average hedonic dimension instead of five items in the hedonic dimension scale.

Another five questions were made for an evaluation of the utilitarian dimension of the product: UTI1 “Effective”, UTI2 “Helpful”, UTI3 “Functional”, UTI4 “Necessary”, UTI5 “Practical”. Participants had to indicate each question on a 7-point Likert scale, from (1=) effective/ helpful/ functional/ necessary/ practical to (7=) ineffective/ unhelpful/ not functional/ unnecessary/ impractical. From the analysis, the Cronbach’s Alpha (a) was .912 for the five items of the utilitarian dimension scale which indicated a high reliability for the scale. The corrected item-total correlations were all above .30, and none of the items would substantially affect reliability if they were deleted. In the following analysis, a new variable UTI_TOT was used to the capture the average utilitarian dimension instead of five items in

(38)

the utilitarian dimension scale. A full overview of the descriptive statistics, correlations and scale reliability is presented in the table 2.

4.3 Normality, Kurtosis and Skewness

The computed mean scale INV_TOT “average investment decision” was a continuous variable. The mean of INV_TOT equalled to 3.59 (SD = 1.49) with slightly negative skewness and slightly negative kurtosis (skewness = -.197, kurtosis = -1.04). It suggested an approximately normal distribution. This satisfied one of the necessary conditions for ANOVA (Cooper & Schindler, 2010, p.182).

(39)

Table 1 Demographic Information of Participants per Treatment Group

Peer recommendation Computer-generated recommendation No recommendation (control groups) Variable Utilitarian product

(n=41) Hedonic product (n=41) Utilitarian product (n=41) Hedonic product (n=40) Utilitarian product (n=41) Hedonic product (n=41) Age (years) M 33.75 33.02 34.68 33.17 32.75 33.31 SD 10.46 7.50 14.24 9.02 6.85 9.79 Gender n (and %) Female 29 (71) 24 (59) 21 (51) 26 (65) 25 (61) 20 (49) Male 12 (29) 14 (34) 19 (46) 14 (35) 16 (39) 20 (49) Other 0 (0) 3 (7) 1 (3) 0 (0) 0 (0) 1 (2) Residency n (and %) Africa 2 (5) 2 (5) 0 (0) 0 (0) 0 (0) 0 (0) Asia 20 (49) 18 (44) 24 (59) 18 (45) 18 (44) 18 (44) Australia/Oceania 0 (0) 0 (0) 1 (2) 0 (0) 0 (0) 0 (0) Europe 16 (39) 20 (49) 14 (34) 22 (55) 21 (51) 23 (56) North America 3 (7) 0 (0) 2 (5) 0 (0) 2 (5) 0 (0) South America 0 (0) 1 (2) 0 (0) 0 (0) 0 (0) 0 (0)

(40)

Table 2 Means, Standard Deviations, Correlations and Reliabilities

Variables

Number

of items M SD 1 2 3 4 5 6 7 8 9

1. Age 1 33.45 9.86 -

2. Gender (0 = female, 1 = male, 2 = other) 1 .43 .53 .09 -

3. Residency 1 3.04 1.08 -.09 -.02 -

4. Kickstarter familiarity (0 = no, 1 = yes) 1 .64 .48 -.199** .18** .35** -

5. Crowdfunding history 1 .52 .90 .05 .19** .24** .22** -

6. Crowdfunding wish (0 = no, 1 = yes) 1 .44 .49 -.04 .07 .32** .45** .40** -

7. Product type (1 = utilitarian, 2 = hedonic) 1 1.49 .50 -.028 .05 .04 -.04 .10 -.01 (.90/.91)

8. Referral sources (1= none, 2= peer, 3= computer) 1 1.99 .81 .03 -.024 -.06 -.07 .00 -.04 -.00 -

9. Average investment decision 3 3.59 1.49 .08 -.11 -.19** -.11 -.11 -.03 -.32** -.00 (.90)

Note: N=245. Reliability is reported along the diagonal.

(41)

4.4 Manipulation Check

In this study, the researcher manipulated the referral source and the campaign product. The first manipulation was factual: the referral source could either be absent or present (computer-generated recommendation or peer recommendation). The description of recommendation was a forced-response question. That meant all participants were mandatory to read the referral recommendation except control groups. Therefore, the researcher did not perform manipulation check on referral sources. The only manipulation check performed was the campaign product.

The campaign product

Each participant had to answer ten questions on 7-point Likert scales. The purpose of this manipulation was to obtain two respondent groups, one group for the utilitarian product and one group for the hedonic product. The mean scores of the utilitarian dimension and the hedonic dimension in both groups were compared respectively to check if the manipulation was successful.

Firstly, an independent-samples t-test was conducted to compare two products and the average utilitarian dimension UTI_TOT. The lower the UTI_TOT score, the more it is utilitarian. The result showed that utilitarian dimension was statistically lower for the utilitarian product (M = 3.16, SD = 1.20) than for the hedonic product (M = 4.33, SD = 1.35); t(243) = -7.17, p < .001. It meant the participants who saw the utilitarian product indeed more likely to agree it was a utilitarian product than those see the hedonic product. Secondly, an independent-samples t-test was conducted to compare two products and the average hedonic dimension HED_TOT. The lower the HED_TOT score, the more it is hendoic. There was not a significant difference in the hedonic dimension for the utilitarian product (M = 4.04, SD =

(42)

1.18) and hedonic campaign (M = 3.92, SD = 1.37); t(237) = .72, p = .475. It meant all participants found similar hedonic attributes in both products but directionally the hedonic product was perceived relatively more hedonic than the utilitarian product because the mean of the hedonic product (M = 3.92) is less than the mean of the utilitarian product (M = 4.04).

The researcher further carried out another approach to check the manipulation by recoding five utilitarian-dimension items to “rUTI1”, “rUTI2”, “rUTI3”, “rUTI4”, “rUTI5”. The researcher treated the utilitarian-dimension items as counter-indicative items because an agreement with the utilitarian-dimension items indicates a relatively low level of hedonic attributes. Together with five hedonic-dimension questions, a mean score PRO_TOT was computed. The lower the PRO_TOT score, the more it is hedonic. An independensamples t-test was conducted between two products and the mean score PRO_TOT. The result showed that the scores were statistically lower for the hedonic product (M = 3.80, SD = .59) than for the utilitarian product (M = 4.44, SD = .62); t(243) = 8.24, p < .001, meaning that the participants who saw the hedonic product indeed more likely to agree it was a hedonic product than those see the utilitarian product. The researcher concluded that the manipulation was effective and successful.

4.5 Hypothesis Testing

Prior testing the hypotheses, table 3 summarizes the average investment per treatment group and figure 3 illustrates how average investment differ per treatment group graphically. By looking at the means of all treatment groups, the researcher found that the average investment was evaluated slightly higher for computer-generated recommendation groups than peer recommendation groups, however, the mean values could not explain the hypotheses and further analysis was carried out.

(43)

Table 3 Investment Decision per Treatment Group

Average investment

Treatment Group n M SD

Peer recommendation / Utilitarian product 41 3.983 1.408

Peer recommendation / Hedonic product 41 3.048 1.519

Computer-generated recommendation / Utilitarian product 41 3.894 1.385 Computer-generated recommendation / Hedonic product 40 3.341 1.503

No recommendation / Utilitarian product 41 4.374 1.220

No recommendation / Hedonic product 41 2.926 1.444

(44)

Investment Decision

A two-way ANOVA was performed to test whether the investment decision towards peer recommendation was higher than the investment decision towards computer-generated recommendation, and whether there was an interaction effect between referral sources and product types on funders’ investment decisions. The test of homogeneity of variances was .479 (p > .05) which implied that the necessary condition for ANOVA (Cooper & Schindler, 2010, p.182) was satisfied. In the ANOVA, the dependent variable INV_TOT “average investment decision”, the independent variables REFERRALS “referral source” and PRODUCT “product type” were added as factors to test the interaction. Table 4 shows the results of two-way ANOVA and figure 4 graphically presents the interaction effect.

Table 4 Two-way ANOVA Results of Investment Decision

Source F Sig. Partial Eta Squared (η2₎

Referrals .200 .819 .002

Product 29.208 .000* .109

Referrals*Product 2.046 .131 .017

(45)

Figure 4 Two-way ANOVA interaction effect of referral source and product type on investment decision

H1a Peer recommendation has a more positive effect on a funder investment decision than no recommendation.

H1b Peer recommendation has a more positive effect on a funder investment decision than computer-generated recommendation.

Hypothesis 1a stipulated that funders who consult the peer recommendation are more likely to invest the campaign than funders who do not consult a recommendation source. Hypothesis 1b predicted a more favourable investment decision in the peer recommendation than the computer-generated recommendation. However, the results showed that the main

(46)

effect of referral sources was insignificant, F(2, 239) = .20, p = .82, η2_{= .002. Both}

hypotheses 1A and 1B were not supported.

H2a Peer recommendation for a hedonic product has a more positive effect on a funder investment decision than computer-generated recommendation for a hedonic product.

H2b Computer-generated recommendation for a utilitarian product has a more positive effect on a funder investment decision than peer recommendation for a utilitarian product.

Hypothesis 2a predicted that the hedonic product enhances the positive effect of peer recommendation on the investment decision. Hypothesis 2b predicted that the utilitarian product enhances the effect of computer-generated recommendation on the investment decision. However, results showed that the interaction effect (referral sources x product types) is not significant, F(2, 239) = 2.05, p = .13, η2_{= .017. Both hypotheses 2A and 2B}

were not supported.

The researcher would like to ensure that the investment decision was solely influenced by the experimental setup. The majority of the respondents should know Kickstarter. The results showed that 158 respondents (64.5%) have heard of Kickstarter and 87 respondents (35.5%) did not aware Kickstarter. Thus, a Pearson Chi-square analysis was performed. The Kickstarter familiarity of all treatment groups showed no significant difference from each other, X2_{(5) = 2.61, p = .761. Individuals' familiarity with Kickstarter}

and crowdfunding experience might impact their investment decisions. Therefore, the researcher took account of these confounding variables to further carry out ANCOVA. The dependent variable INV_TOT “average investment decision”, the independent variables PRODUCT “product type” and REFERRALS “referral source”, the covariates

(47)

HEARD_KICKSTARTER “Kickstarter familiarity” and FUND_HIS “crowdfunding history” were added as factors to test the interaction. However, the result of ANCOVA did not show much different than the result of ANOVA. The main effect of referral sources and the interaction effect of referral sources and product types remained insignificant. The researcher, therefore, did not present the result here.

4.6 Additional Analyses

The researcher discovered an interesting finding which is not part of the hypotheses. As illustrated in table 4, the result of ANOVA revealed that a significant main effect of the product type on investment decision, F(1, 239) = 29.21, p < .001, η2 _{= .109. The mean}

investment value of the utilitarian condition (M = 4.08, SD = 1.35) was higher than the mean investment value of the hedonic condition (M = 3.10, SD = 1.49), meaning that the utilitarian campaign leads to a more favourable effect on the investment than the hedonic campaign. The value of η2_{= .1 indicated the moderate to high effect size, followed the general rules of}

thumb given by Cohen (1988). Figure 5 presents a clear picture of the two-way ANOVA main effect of product type on investment decision.

(48)

Figure 5 Two-way ANOVA main effect of product type on investment decision

Given that the use of referral sources in crowdfunding is relatively new, the optional open-end question: “Please elaborate why you choose to (not) support the campaign?” served a qualitative angle to this research. In the analysis, the comments given by participants were read and counted by the researcher. Table 5 shows the amount of qualitative comments per treatment group.

Referral effectiveness : an empirical study of referral sources on investment decisions in reward-based crowdfunding

MASTER’S THESIS