The effect of crowdfunding outcomes on subsequent funding decisions

(1)

1

The effect of crowdfunding outcomes on subsequent funding

decisions

Master Thesis – August 2015

MSc Business Economics, Finance track

Jaap Woltjes - 10676899

(2)

2 Statement of Originality

This document is written by Student Jaap Woltjes 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)

3

Crowdfunding is becoming an increasingly important phenomenon to raise financing for entrepreneurial ideas. It is defined as an open request over the internet for financial support in the form of a donation, often in exchange for a reward, service or a future product (Kleemann et al., 2008; Belleflame et al., 2014). On crowdfunding platforms, such as Kickstarter, RocketHub and IndieGoGo, initiators (people who request funds) and funders (people who pledge money, i.e. fund projects) are linked. These platforms charge fees for channeling donations.

There are hundreds of platforms across the world and they are experiencing an exponential growth in popularity (Gerber et al., 2012). They fund a diverse range of projects, such as product design, software, fashion and film. In 2012, the U.S. government recognizes crowdfunding as a key to economic growth. Despite the rapid growth, few scholars have examined crowdfunding (Gerber et al., 2015).

While crowdfunding is still relatively small compared to alternative sources of financing, the 800 million invested into crowdfunding suggests the market is expected to grow further (Crowdsunite, 2013). There are multiple types of crowdfunding. This thesis focusses on reward-based crowdfunding. Reward-based crowdfunding provides entrepreneurs with financing in exchange for a reward. The academic debate has mainly focused around this type of crowdfunding. Within reward-based crowdfunding the majority of the funding is raised according the all-or-nothing model (Cuming et al., 2015). This means a project only receives funding when the funding raised is at or above the funding goal. In this research, we focus on reward-based all-or-nothing crowdfunding.

There is some research in the sparse crowdfunding literature on the motivation of people to fund single projects (Gerber et al., 2012; Brettschneider et al., 2014; Gerber et al., 2015). And other research has implicitly considered subsequent decisions made by funders as independent (Mollick et al., 2014). Belleflamme et al. (2014) has modelled the decision to fund in a project as a rational choice to maximize utility per decision, without taking the effect of previous investments into account. This effect might be relevant: the literature on prior outcomes and subsequent decision making suggests that individuals do not always follow rational beliefs in their decision making (Campbell, 2006). Decision-making research focuses on incentivized competition-based situations and reward-based all-or-nothing crowdfunding cannot be considered a competition for two reasons. First, because there is no real financial loss in case the funding goal is not reached. Second, if the all-or-nothing principle is applied, there are no individual

(5)

5 winners or losers: they all win, or they all lose. Nevertheless one could argue funders do experience a negative funding outcome as a loss, because their personally elected project is cancelled.

This raises the question whether funders make individual decisions for every single project, or that they take previous funding outcomes into account when funding projects. This leads to the following research question:

Do cancelled previous all-or-nothing reward-based crowdfunding outcomes affect subsequent decisions to stop or continue funding projects?

Besides the main question the following sub question is examined: Does education influence the decision to stop or continue pledging?

The answers on these research questions could be relevant for crowdfunding platforms for two reasons. First, it is relevant for their admission policy. Should they allow all projects, or refuse projects that are unlikely to succeed? This affects the number of negative outcomes, which is the topic of this thesis. Second, should platforms increase the number of funders, or the number of projects? Spreading the same amount of funding across more projects is likely to increase negative funding outcomes.

In this thesis an empirical study was done to measure whether there is an impact of project outcome on the subsequent decision to continue or stop pledging in projects and if there are influences of education. Data were retrieved from the largest crowdfunding platform Kickstarter and personal data were

ingeniously retrieved from Linkedin. The dataset consists out of four parts: a list of funders on Kickstarter (name, location, education), the projects they funded, the funding outcomes and their Linkedin profiles (if available). The Kickstarter dataset was joined to the correct Linkedin profiles. In total there are 34.577 unique matches found for the dataset of this thesis.

The research is structured as follows: first, the literature on crowdfunding and decision making is reviewed. Second, the data and methodology are described. Third, the results of the analysis are presented, and finally, the conclusions derived from the results are discussed.

(6)

6

2. Literature review

2.1 Crowdfunding

Crowdfunding is defined as “an open call, mostly through the Internet, for the provision of financial resources either in the form of donation or in exchange for the future product or some form of reward to support initiatives for specific purposes.” (Belleflamme et al., 2013). In general, the literature on

crowdfunding is sparse, because it has only recently become popular (Kuppuswamy & Bayus, 2013). There are different types of crowdfunding models, which are explained by categorizing them according to two different criteria: First, the exchange between the crowd and the crowdfunding projects. Second, in terms of the conditions required for the exchange to take place. Regarding the exchange, the crowd always provides a fixed dollar amount of funding to the project. Depending on the crowdfunding model, the crowd receives either equity, interest, a good or service, or nothing in case of a donation. In case of equity, the model is called equity-based crowdfunding. The market-size for this model is relatively small due to legislative restrictions (Simon, 2014). In case of interest the model is referred to as peer-to-peer (P2P) lending, which has market size of 5.5 billion dollars in the U.S (PwC, 2015). In case of a good or service, it is called pre-ordering or reward-based crowdfunding. This form has grown quickly. Two of the most popular examples include Kickstarter and IndieGoGo. On these platforms, donation-based

crowdfunding is included as well. Bellaflame et al. (2013) discusses the choice between crowdfunding models. This thesis focusses on reward-based crowdfunding models, because the academic debate has mainly focused around reward-based crowdfunding.

Reward-based crowdfunding can be further categorized into keep-it-all and all-or-nothing models. In case of the keep-it-all model the funder receives all the funding provided by the crowd. The all-or-nothing model requires the funding to be above a funding goal defined by the crowdfunding project initiator. The funding goal consists of a fundraising amount and a time period for fundraising. During this time period, funders can pledge to fund the project. A pledge is a commitment to fund the project when the funding goal is reached. At the end of the time-period the funding goal is compared with the actual funding raised. Then the principle of the all-or-nothing model is applied: the project only receives funding when the funding raised is at or above the funding goal. Figure 1 provides an overview of the all-or-nothing model. This thesis refers to positive funding outcomes as projects that received funding because there were enough pledges to reach the funding goal. In the other case, the project does not

(7)

7 receive funding, which is referred to as a negative funding outcome. This thesis focusses on the all-or-nothing model, since the majority of funding is raised through the all-or-all-or-nothing model (Cumming et al., 2015).

Figure 1: Overview of one project over time.

The figure displays the course of a project over time. When a projects’ commitments exceed its goal, the project receives the commitments. When the projects’ commitments are lower than the goal, the commitments are cancelled.

In the research on reward-based keep-it-all crowdfunding, three elements are distinguished: the project initiator, the project and the funder. Existing research has examined the relationship between the project initiator and the funder (Agrawal et al., 2010), the project initiator and the project (Mollick, 2014), and between the project and the funder (Gerber & Hui, 2012). This thesis focusses on the relationship between the project and the funder.

The main attributes of a project are: information about the project, the funding goal, and the rewards offered in exchange for funding. The information about the project contains at least a short text description of the project. Optionally, the project initiator can add a video about the project, and his

(8)

8 social network profile. Through this information, the funders can be informed why the funding is needed, and how they plan to spend the funding received.

Existing research has examined the relationship between these project attributes and three outcomes: delivery outcome, publicity benefits, and funding outcome. The first outcome, the delivery outcome, is examined by Mollick (2014), using a dataset of 381 Kickstarter projects. The reward-delivery outcome refers to whether the reward is delivered, and when the reward is delivered. This dataset is different from the dataset used to analyze funding success, potentially because the data collection for reward-delivery outcomes cannot be automated with software. In this dataset, 3.8% or 14 of the 381 projects were not delivered. Regarding the delivery-time, they find that large projects, i.e. projects with a high funding goal are related to delayed project delivery. The analysis for non-delivery is done using survival analysis, more specifically using a Cox proportional hazard model.

The second outcome, the publicity outcome is studied by Burtch et al. (2013). They find that the duration of the fundraising period has a positive effect on attention of the public for the crowdfunded projects. The mechanism suggested is that the higher the duration of the fundraising period, the higher the exposure to the public, the crowd. This exposure enables a larger crowd to become aware of the project. The paper uses the GMM-based Dynamic Panel Estimator method with a dataset of crowdfunding projects for journalism.

The third outcome, the funding outcome, is the focus of this thesis, because raising funding is arguably the most important goal of crowdfunding. Funding outcomes are a result of the behavior of funders. In line with this, the behavior of funders has been an important focus of crowdfunding research. Gerber & Hui (2013) examine the motivations and deterrents of funding a project though 83 semi-structured interviews of funders. They find that collection of rewards, help others, being part of a community and supporting a cause are the main motivations for funding a project, and distrust of the project initiators’ use of funds is the main objection to fund a project.

Mollick (2014) finds that the funding outcome is associated with project quality, geography and personal networks of the entrepreneur. The study is based on the analysis of 45.000 Kickstarter projects. He adopts signaling theory to explain these findings, and argues that funders interpret project attributes as signals of certainty of project quality. This is to address the potential misuse of funds, and the potential

(9)

9 that the project is promising a better outcome that they are able to deliver. Several attributes of a project are able to signal the certainty of project quality: the size of social network of the project

initiator, the preparedness of the project initiator as indicated by the presence of a video, and frequency by which the project initiator posts updates on the project page. The analysis is based on a logistic regression of funding success on these quality signals. While he finds associations, no causal interpretation is given.

The research examining the behavior of funders does not consider a relationship between the funding decisions of one funder. A relationship between multiple decisions of a single funder could also not be explained by the theory adopted in crowdfunding. Signaling theory explains the relationship between project attributes and the decision to fund projects. In general, each project has different project

attributes, which are completely independent from each other. Additionally, each project has a different project initiator, and given the geographical distance between project initiators (Mollick, 2014), they likely operate independently of each other. Therefore, one could argue that existing crowdfunding research has implicitly considered subsequent decisions made by funders as independent. This seems logical, since funders do not experience a financial loss when funding outcome is negative, i.e. the funding goal has not been reached.

On the other hand, when funding outcomes are as a win or a loss by funders, (although there is no financial loss), research on prior outcomes and subsequent decision-making might be relevant. They study the relationship between prior outcomes and subsequent decisions.

2.2 Prior outcomes and subsequent decision making

This section discusses relevant theories that might explain the relationship between prior outcomes and subsequent decision-making in general. For each discussed theory, the prediction of the theory in context of my thesis is made, and the existing empirical evidence for the theories is discussed. Finally, a hypothesis is developed, combined with its contribution to the literature.

Expectation-based reference-point theory explains how people minimize disappointment through their decision-making. Disappointment is defined as a psychological reaction caused by comparing the actual

(10)

10 outcome to one’s prior expectations (Bell, 1985). Using an experimental design, Abeler et al. (2011) empirically confirmed that people avoid disappointment by stopping to participate in a game at a point where they expect the lowest likelihood of disappointment. In context of crowdfunding, this theory would predict that a negative funding outcome would lead to stop pledging to avoid disappointment. Risk-preference theory explains how prior outcomes affect future decision-making. Thaler and Johnson (1990) find two mechanisms that affect subsequent decisions: the house-money effect and the break-even effect. The break-break-even effect describes that people are loss-averse and therefore want to decrease their loss. The house-money effect describes that people are more likely to take a risk with previously gained money, because they consider the money gained the money of others. This thesis is interested in the direct effect of funding outcome, and because a negative funding outcome is not affecting the amount of money, this means this mechanism is not applicable.

Fear of failure theory explains that people experience shame when failing. Empirical evidence show that people attempt to mentally escape this unconformable state by withdrawing (Elliot and Church, 1997). In context of funding outcomes, this theory predicts that funders stop pledging when they experience fear of failure due to a negative funding outcome.

Ego utility theory describes that people derive utility from a positive belief about themselves, and that they attempt to maintain these positive beliefs (Köszegi, 2006). In context of crowdfunding, this theory predicts that people stop funding after a negative funding outcome to avoid negative beliefs about themselves, which are inconsistent with their positive beliefs.

Mental accounting theory predicts that people create budgets in their mind, which they use for decision-making (Thaler, 1985). When there would be a mental account for crowdfunding expenses, a negative funding outcome would lead funders to continue pledging, because they have more budget left compared to when the funding outcome was positive.

From this literature overview the question arises whether you can consider subsequent decisions made by funders as independent (Mollick, 2014) or is there a relationship between subsequent decisions analogous to competition outcomes and subsequent decisions. This leads to the following hypothesis:

(11)

11 H1: Previous negative funding outcomes have a positive effect on the subsequent decision to stop pledging

The results of this hypothesis might contribute to the literature on funding behavior in crowdfunding and on the literature on the effect of competition outcomes on subsequent decision-making.

The theory and empirical research on the motivation of funders is focused on the question why funders fund independent projects (Gerber & Hui, 2014). It is reasonable to expect that funding decisions are independent when applying signaling theory and when comparing financial outcomes. The signals regarding project quality are unrelated across projects: project initiators operate independently, do not directly influence each other, and use the funding raised only for their own project. Therefore, signaling theory does not indicate a relationship between multiple projects funded by one funder (Mollick, 2014). Additionally, the funder does not have a financial gain or loss when the funding outcome is negative. In this case the funding is not provided to the project, and the financial situation of the funder remains the same. Thus this thesis contributes to the literature on funding behavior in crowdfunding.

Theory and empirical evidence on competition outcomes and subsequent decision-making shows that winning or losing has an effect on subsequent decision-making (Buser, 2015). Previous research has focused on financial outcomes of competitions (Apicella, Dreber, and Möllerström, 2014; Buser, 2015). However, in this thesis the funder does not experience a change in the financial situation in case of a negative funding outcome. Additionally, existing research has focused on competition outcomes due to effort and due to luck (Buser, 2015; Gill and Prowse, 2012). These outcomes are different from

crowdfunding outcomes in two aspects. First, funding outcomes are not the result of a competition, but of an all-or-nothing model: either all funders of a project fund the project, or none of the funders. Second, funding outcomes are not completely derived from effort or luck, but from a combination of effort and luck. Effort is made by the funder to assess the quality signals of projects to prevent funding low quality projects and mis-use of funds (Mollick, 2015). Luck is arguably needed because the funder is dependent on the decisions of other funders. Thus, this thesis contributes to the literature on

competition outcomes and subsequent decision-making by examining if their theories are generalizable to crowdfunding despite the discussed differences.

(12)

12 Existing literature on the effect of prior outcomes on subsequent decision-making has also examined whether this effect depends on gender. For example, Buser (2015) finds that gender differences affect the relationship between prior outcomes and subsequent decision-making. The finance literature suggests that educational differences lead to a difference in financial decision-making (Campbell, 2006). This raises the question whether education could affect the relationship between prior outcomes and subsequent decision-making? The next section discusses education in context of this relationship.

2.3 Education

Research on education and subsequent decision-making has focused on regular educational attainment such as education in college, or “just-in-time” education (Fernandes et al., 2014). Just-in-time education is education at the moment the decision is made, such as recommender systems and decision-support systems. This thesis focusses on effects of regular education on subsequent financial decision-making. People with high educational attainment have higher levels of financial knowledge. This is empirically confirmed by Rooij et al. (2011) using a dataset from De Nederlandsche Bank (DNB) Household Survey and by Lusardi & Mitchell (2007) using the Health and Retirement Study (HRS) in the United States. Both studies also find that higher financial knowledge leads to a longer planning horizon. This long-term horizon could lead funders to be less sensitive to positive and negative funding outcomes, which leads to the following hypothesis:

Hypothesis 2: Higher education, decreases the effect of previous negative funding outcomes on a subsequent decision to stop pledging relative to lower educated people.

The results of this hypothesis might contribute to the literature on crowdfunding and on the literature on previous outcomes on subsequent decision-making. Previous research on crowdfunding did not consider education to be a factor in the decision-making of a funder. Additionally, the research on previous outcomes on subsequent decision-making has mainly focused on gender differences (e.g. Buser, 2015; Gill and Prowse, 2010). This thesis contributes by examining education as an alternative potential factor which may affect the relationship between previous outcomes and subsequent decisions.

(13)

13 Additionally, education might be associated with both important factors of this thesis: funding outcome and subsequent decision-making. Positive funding outcomes might be experienced more often by highly educated people, because they are more like to make funding decisions that are similar to the majority of the crowd. This is supported by the empirical research by Mollick and Nanda (2014) who show that the collective decision-making of crowds is similar to decision-making of individual highly educated experts using a dataset of Kickstarter projects and a panel of experts. Subsequent decision-making to continue funding might be associated with high education, when making the analogy with stock-market participation: Rooij et al. (2011) find that financially educated people are more likely to participate in the stock-market.

2.4 Geographic differences: U.S. and non-U.S.

Existing cross-country research shows that financial decision-making differs across countries. People in the U.S. have different risk preferences compared to other parts of the world. For example, U.S. discount future rewards and future probabilities more than e.g. China and Japan (Green & Myerson, 2002). Additionally, people in the U.S perceive less risks when buying online (Park, 2003). Therefore, one could expect that U.S. people have more negative funding outcomes relative to the average funder.

Additionally, the lower risk-aversion and higher discounting of probabilistic rewards could result in that U.S. people decide to stop funding projects less quickly.

(14)

14

3. Data

To investigate the hypotheses in the previous chapter, data are collected from the largest crowdfunding platform Kickstarter and personal data were ingeniously retrieved from Linkedin. In this chapter the data collection process and the collected data are described. This includes how the huge amount of data - 6 million funding outcomes and ~ 2 million Linkedin profiles - are collected and joined together.

3.1 Data collection process

The data collection process is automated. This section describes the process using three steps. First it is explained how the data collection process could be done manually using only a browser and some place to store data like Excel. Second, the techniques used to automate these steps are mentioned. Third, the problems that need to be overcome are discussed on a high level. For each problem, the solution used to overcome the problem is briefly discussed.

The data collection is done for Kickstarter and LinkedIn. The data collection for Kickstarter can be done manually in three steps. First, you would open a browser, visit profile of a funder on Kickstarter, and note their details and assign them an ID. A screenshot of a Kickstarter funder page can be seen in Appendix I. Second, the project page for each of their investments is opened, and the project details are noted and an ID is assigned. A screenshot of one project in the dataset - the Pebble Watch – is added to Appendix I. Third, the investments are noted as combinations of funder ID and project ID.

The Linkedin data collection can be done in two steps. First, search on Linkedin for the name of the funder. The results of the search results for one funder can be seen in Appendix I. Second, the search results are reviewed for Linkedin profiles that match the location of the funder on Linkedin. When only one Linkedin profile matches the name and location of the Kickstarter funder exactly, a match is identified. In this case, all the education mentioned on the Linkedin profile is noted as the education of the funder.

This process of visiting websites, selecting and storing information has been automated using software (Allan, 2013). For this relatively simple process, there were two problems that had to be solved. First, when website visits are done sequentially by a computer, the millions of website visits would take multiple days. To solve this, parallel execution is needed, which is realized using the queue manager Beanstalkd and multiple processes using Supervisord. Second, because of the high number of website

(15)

15 visits, the website blocks the traffic. To solve this, the website visits need to appear to originate from many different computers. To realize this, two solutions are needed simultaneously. First, the IP should be different, which is realized by requesting websites through the TOR network. Second, the website visits need to appear to be made from computers with different operating systems, browsers and screen resolutions. To realize this, a http header randomizer is used.

3.2 The data

The dataset consists out of four parts. To provide an understanding of the total dataset, the four sub-datasets are described consecutively. After this, it is described how the four subsub-datasets are joined. The first dataset is a list of funders on Kickstarter. Each funder has the following attributes: their full name and their location. This total number of funders is 3.038.524, but only 9% of these funders reported a location. The funders who did not report a location are removed from the dataset, because the location is needed to match the funders from Kickstarter to the corresponding LinkedIn profiles during the joining of datasets. At this point there are 333.798 funders left in this dataset.

The second dataset is a list with the projects where funders invested in. There are 87.260 projects. While funders are not filtered at this stage, the projects are filtered during the joining process because some projects do not have funders that are within the first dataset. The projects contain the project name, the number of funders that pledged to the project, the category, the duration of the project, the goal of the project and the amount raised by the project.

The third dataset is a list of pledges, which is synonymous to a list linking funders and projects. The reason this is separate dataset, is because of the many-to-many relationship; one funder can pledge for multiple projects, and one project can have multiple funders. The number of pledges in the dataset is 5.865.246.

The fourth dataset is the LinkedIn profiles of the funders that are available by searching for the first- and last name of funders on LinkedIn. This resulted in 1.878.018 Linkedin profiles.

The Kickstarter dataset needs to be joined to the correct LinkedIn profiles. This is done based on the following three criteria: the first- and last name has to match exactly, and the location of the funder has to be contained in the location of the LinkedIn profile, and there should only be one match for the funder. There are 106.941 Linkedin_profiles that match these criteria, of which 34.577 unique matches.

(16)

16 Unique matches are necessary; in the case there are two Linkedin profiles that have the same name and location, it is still uncertain which Linkedin profile should be matched. An overview of the sub datasets and the joined dataset can be found in table 1.

(17)

17 Table 1. An overview of the sub datasets and their contents.

Table

name No. of rows Attribute name

Attribute description Linked by No of rows after linking Ki ckst art er P ro jec t ta b le 3.038.524 Hyperlink to project 179.771 Category of project Art Comics Dance Design Fashion Film & Video Food Games Music Photography Publishing Technology Theatre Duration of collection period No of days End date of collection

period

Date Funding goal set by

project No of $ Funding raised No of $ Textual description of project Text # of comments Number Project id Number X1 fu n d er t ab le 87.260

User name Text

funder id Number X2

Location of user

Outside U.S: city and Country; Inside U.S: city and state (text)

Full name Text X3

P le _dg 5.865.246 _ese _si funder id Number X2 Project id Number X1 Li n ke d in 1.878.018

Linkedin profile id Number

Full Name Text X3

Education

University or college or any school name Location of residence Text

(18)

18 So the total number unique pledge and funder combinations is 197.771. The total number of funders was 34.755 who had an average of 5.2 pledges. The frequency of the number of pledges per funder is shown in figure 2. In this figure it is shown that ± 43% of the pledges only invested one time. The average number of pledges per funder is 5.2. The maximum is 809 for a single funder.

Figure 2: The frequency distribution of the total number of pledges per funder to their number of pledges.

0 10 20 30 40 Pe rce n t 0 2 4 6 8 10 Total pledges

(19)

19

3.2.1 Data top education compared to non-top education

There are two subsets of data: the subset of funders with top education and the subset of funders without top education. The two subsets are compared and contrasted using figure 3 and table 2. Table 2 shows that there are significantly more people without top education than people with top education. Only 3.6% of all funders has received top education. While the mean number of pledges differs only marginally between the two subsets, the people without top education have more variation in the number of pledges; the standard deviation of funders without top education is 13.2, while the standard deviation of 7.7 for funders without top education is nearly half. This difference in variation of number of pledges is also evident in the maximum number of pledges. The maximum number of pledges for funders without top education is 809, while the maximum for funders with top education is 80.

Table 2: Summary statistics

This table looks at two subsamples of funders. In the first row are the summary statistics for the first subset - funders without top education. In the second row are the summary statistics for the second subset - funders without top education. Top education is defined as having an Ivy League University from the U.S. or a Golden Triangle University from the U.K. as education mentioned on their LinkedIn Profile. There are several patterns in

the summary statistics. There are significantly more funders without top education. While the mean number of pledges are comparable, the number of pledges for people without a top university deviates significantly more, as can be seen from the higher standard deviation and higher maximum for funders without top education.

Total number of pledges per funder N mean S.D. min max

Subset: funders without top education 33324 5.199886 13.21528 1 809 Subset: funders with top education 1253 5.179569 7.758876 1 80

(20)

20 While table 2 showed the aggregate statistics, figure 3 shows a histogram for the two subsets. This different view of the two subsets shows the distributions are not equal. While the percentage funders who pledged once is less than 40% of the total subset of funders with top education, this is around 43% for funders without top education. To summarize, the two subsets are compared and contrasted, and some differences have been highlighted.

Figure 3: Total number of pledges per funder compared between funders with and without top education

This figure looks at the total number of pledges per pledges for two subsets. The left graph shows the total number of pledges per funder for people with top education. The right graph shows pledges per funder for people without top education. There are some differences between the two subsets that become appearant by contrasting the subsets in these graphs. While funders who pledged only one time represent less than 40% of the total subsample with top education, they represent more than 40% of the subsample without top education.

0 10 20 30 40 Pe rce n t 0 2 4 6 8 10

Total number of pledges per pledger

0 10 20 30 40 Pe rce n t 0 2 4 6 8 10

(21)

21

3.2.2 Data U.S. compared to Non-U.S.

There are two subsets of data: the subset of funding outcomes of U.S. funders and the subset of funding outcomes outside the U.S.. The datasets are compared and contrasted using table 3 and figure 4. Table 3 below shows the summary statistics for the two subsets. When comparing the subsets using the table 3, there are several notable differences. There are significantly more U.S. funders compared to funders from outside the U.S. 78% of the funders is from the U.S. These U.S. funders have a slightly higher average number of investments. Additionally, the standard deviations are comparable. This is in line with the comparable maximum number of pledges per funder.

Table 3: Summary statistics

This table looks at two subsamples of funders. In the first row are the summary statistics for the first subset - funders outside the United States. In the second row are the summary statistics for the second subset - funders

within the United States. This is the location from the LinkedIn Profile and Kickstarter profile page. There are several patterns in the summary statistics. There are significantly more funders within the U.S. compared to the rest of the world. The U.S. funders have a higher average number of pledges, and a higher maximum number of

pledges.

Total number of pledges per funder N mean S.D. min max

Subset: Non-U.S. funders 7594 4.950092 13.30341 1 667

Subset: U.S. funders 26983 5.269244 12.98666 1 809

(22)

22 Figure 4 compares the two subsets through two histograms. The left histogram displays the total number of pledges per funder from outside the U.S. The right histogram displays the total number of pledges per funder from inside the U.S. The clearest difference is the percentage of funders of the total subset that pledged one time. The percentage of funders that invested only one time is almost 50% for the subset with funders from outside the U.S. The percentage of funders from outside the U.S. is around 42%. For other bins in the histogram the differences are less high. Overall, there are some differences between the datasets, and the next section discusses the methodology to analyze the datasets in more detail.

Figure 4: Total number of pledges per funder compared between funders within and outside the U.S.

This figure shows the total number of pledges per funder for two subsets. The left graph shows the total number of pledges per funder for people outside the U.S. and the right graph in the U.S. There are some differences between the two subsets that become appearant by contrasting the subsets in these graphs. While funders who pledged only one time represent almost 50% of the total subsample outside the U.S., they represent around 42% of the subsample in the U.S.

0 10 20 30 40 50 P er c en t 0 2 4 6 8 10

Total number of pledges per funder 0

10 20 30 40 P e rc e n t 0 2 4 6 8 10

(23)

23

4. Methodology

This section discusses the methodology in context of the two hypotheses from the literature review. For both hypotheses, the following elements are discussed: the type of regression, the econometric model, potential endogeneity issues, how the coefficients are determined and interpreted, and potential limitations to the interpretation of coefficients.

The first hypothesis is: Previous negative funding outcomes have a positive effect on the subsequent decision to stop pledging. To test this causal relationship, an OLS regression is used. Since the independent variable, funding outcomes, can be about one or multiple previous outcomes, this

relationship is tested for one previous outcome, two previous outcomes, and three previous outcomes. Therefore, the independent variable is defined as the ratio of positive outcomes to total outcomes.

To address potential endogeneity issues, the OLS regression needs to control for potential endogenous variables and potential simultaneous causality. Based on the literature review, this thesis controls for two factors: education and living in the United States. As addressed in the literature review, education could be positively associated with the ratio of positive to total funding outcomes, and negatively associated with the decision to stop pledging. Therefore, this thesis expects a negative coefficient for education. Living in the U.S. could be associated with negative funding outcomes, and negatively

associated with the decision to stop pledging. Simultaneous causality seems unlikely, since the individual decisions to stop funding is unlikely to affect whether a crowd is going to fund a project. This leads to the following regression:

𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛𝑇𝑜𝑆𝑡𝑜𝑝 = 𝛽0+ 𝛽1

∑𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒 𝑜𝑢𝑡𝑐𝑜𝑚𝑒𝑠

∑𝑡𝑜𝑡𝑎𝑙 𝑜𝑢𝑡𝑐𝑜𝑚𝑒𝑠 +𝛽2𝑒𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛 + 𝛽3𝐼𝑛𝑈𝑆

The factors in the regression have the following meaning. The first factor is the ratio between positive outcomes and total outcomes. These outcomes are past outcomes of the funder. Based on the number of outcomes, this thesis distinguishes between three groups of funders. The first group consists of funders that have had one past outcome, which is referred to as the first round. This outcome can be either negative or positive. Given the formula for the ratio, the value of the ratio can be either 0₁ or 1₁ , i.e.

(24)

24 0 or 1. The second group consists of funders that have had two past outcomes, which is referred to as the second round. Each of the two outcomes can be either positive or negative. Therefore, the value of the ratio can be either 0

2, 1 2 or 2 2, i.e. 0, 1

2 or 1. The third group consists of funders that have had three past outcomes, which is referred to as the third round. Therefore, the value of the ratio can be either 0₃, 1₃, 2₃ or 3

3, i.e. 0, 1

3,

2

3 or 1. By doing the regression for each group of funders, differences between funders

between the first, second and third round can be found. This thesis analyzes three rounds, to address the question: are the effects between funders with one past outcome different from funders with more past outcomes. It seems reasonable to expect that when the findings for the three groups of funders are the same, the conclusions are generalizable to funders with one and multiple past outcomes.

𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛 is a binary variable which is 1 when the funder has received top education from an Ivy League University in the U.S. or a Golden Triangle university and 0 otherwise. Ivy League universities are: Brown University, Columbia University, Cornell University, Dartmouth College, Harvard University, Princeton University, University of Pennsylvania and Yale University. Golden Triangle universities are: University of Cambridge, the University of Oxford, Imperial College London, the London School of Economics, King's College London and University College London. The factor 𝐼𝑛𝑈𝑆is 1 when the funder lives in the U.S. and 0 otherwise.

𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛𝑇𝑜𝑆𝑡𝑜𝑝 is a binary variable, which is 1 if the funder stops funding projects, 0 otherwise. This binary is calculated differently for each of the three rounds. In the first round, the funder has

experienced outcomes of two pledges. Therefore, 𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛𝑇𝑜𝑆𝑡𝑜𝑝 is 1 when there is no second pledge done by the funder, and 0 otherwise. In the second round, the funder has already experienced outcomes of two pledges. Therefore, 𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛𝑇𝑜𝑆𝑡𝑜𝑝 is 1 of the funder has made a third pledge, and 0 otherwise. Subsequently, in the third round, 𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛𝑇𝑜𝑆𝑡𝑜𝑝 is 1 if the funder has made a fourth pledge, and 0 otherwise.

When 𝛽1 is negative and significant, this indicates that a decrease in the funding success ratio leads to a higher probability that funders decide to stop funding projects, which is evidence that supports the first hypothesis. Limitations to the interpretation are discussed later in this section.

(25)

25 There are potential limitations to the interpretation of 𝛽1 as a causal coefficient due to limitations of the dataset. The sample was not randomly collected, and treatment of positive and negative funding

outcomes was not randomly assigned. While there are two dummy control variables in the model, there might still be some attributes of funders that are correlated with both the decision to continue or stop funding, and the previous funding outcomes. Due to data constraints, the control variables are binary variables, which makes it likely that the error term is correlated with the regressors.

Therefore the conditional mean zero assumption might not hold.

The second hypothesis is: Higher education, decreases the effect of previous negative funding outcomes on a subsequent decision to stop pledging relative to lower educated people. To test this hypothesis, an interaction term of education and funding outcome ratio is added. The coefficient for this interaction term is expect to be positive, because, as described in the literature review, education is expected to decrease the effect of previous negative outcomes on the decision to stop or continue funding. This leads to the following econometric model:

𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛𝑇𝑜𝑆𝑡𝑜𝑝 = 𝛽0+ 𝛽1

∑𝑡𝑜𝑡𝑎𝑙 𝑜𝑢𝑡𝑐𝑜𝑚𝑒𝑠 +𝛽2𝑒𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛 + 𝛽3𝐼𝑛𝑈𝑆 + 𝛽4(𝑒𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛 ∗

∑𝑡𝑜𝑡𝑎𝑙 𝑜𝑢𝑡𝑐𝑜𝑚𝑒𝑠 )

When 𝛽4 is positive, this supports the second hypothesis, because it indicates a decrease in the effect between previous negative outcomes and the subsequent decision to stop or continue funding. By adding the interaction term with coefficient 𝛽4, the meaning of 𝛽1 changes as well. In this model, 𝛽1 indicates the effect outcome ratio for people without top education. 𝛽4 indicates the change in this effect due to education. The data for the two factors in the interaction term are the same as in the first econometric model. For this econometric model, there are also possible endogeneity issues. The same discussion as at the first hypothesis, regarding the limitations of the dataset, holds for the second hypothesis.

(26)

26

5. Results

This section discusses the results in context of the hypotheses discussed in the literature review. Table 4 shows the results from the OLS regressions of the previous funding outcomes on the decision to stop or continue pledging.

For the first hypothesis, the regressions in column (1), (3) and (5) are relevant. They show the regressions for the first, second and third round without interaction. The coefficient for the funding outcome ratio indicates that a negative funding outcome leads to a 12-15% increase in the probability that a funder stops funding. There are no significant differences for people who funded one project compared to people who funded two or three projects. This is an indication that the first hypothesis is true; despite the fact that people do not lose money from a negative funding outcome, it does affect subsequent decision-making. This evidence does not contrast with existing empirical evidence in crowdfunding, since this is the first empirical analysis on the relationship between funding outcomes and subsequent

decision-making. Furthermore, it is an indication that the empirical findings of competition-outcomes on subsequent decision-making are generalizable to crowdfunding (Buser, 2015). However, the plausibility of the mechanisms to explain the results might differ between crowdfunding and competitions, since funders do not compete; in the all-or-nothing crowdfunding model either all funders have a positive outcome, or they all have a negative outcome. The plausibility of the mechanisms is discussed in the conclusion and discussion section.

The results of this thesis could have Implications for crowdfunding platforms. Crowdfunding platforms link funders and projects, and to grow, they have to balance between number of funders and number of projects. When the number of projects increases, while the total funding remains constant, the funding is spread thinner between the projects. The findings of this thesis suggest that this could lead to a negative spiral: a thinner spread of funding leads to more negative funding outcomes, which leads to more funders deciding to stop funding, and so on.

For the second hypothesis, columns (2), (4) and (6) are relevant. They show the regressions for the first, second, and third round with an interaction factor between education and the funding outcome ratio. The coefficient is positive for the first round, and negative for the second and third round, and in none of the rounds the coefficient is significant. This is an indication that the second hypothesis is not true;

(27)

27 people with top education are not more sensitive to funding outcomes when deciding to stop or

continue funding projects. Previous empirical research indicates that higher education is associated with a longer planning horizon. Contrary to what this thesis expected, the results show that the longer planning horizon of higher educated people does not lead to a difference in the effect between previous funding outcomes on the subsequent decision to stop or continue funding crowdfunding projects.

(28)

28 Table 4

The decision to stop or continue funding as a function of outcomes of previous funding outcomes

This table shows the results of an OLS regression of the funding outcome ratio on the decision to stop pledging. The dependent variable is the same for column (1) to (6) and is defined as a binary variable that is 1 when the funder stops and 0 when the funder continues funding projects. The independent variable funding outcome ratio is calculated as the ratio of positive funding outcomes over the total number of funder outcomes. The independent variable education is defined as a binary variable which is 1 when the person attended an Ivy League university in the United States or a Golden Triangle university in the United Kingdom, and 0 otherwise. The

last factor in the OLS model is an interaction term between the binary variable education and the funding outcome ratio.

ROUND 1 ROUND 2 ROUND 3

Decision to stop funding Decision to stop funding Decision to stop funding

VARIABLES (1) (2) (3) (4) (5) (6)

Funding outcome ratio -0.121*** -0.120*** -0.151*** -0.153*** -0.145*** -0.146*** (0.00852) (0.00868) (0.0143) (0.0146) (0.0190) (0.0194)

education -0.0388*** -0.00965 -0.0367** -0.0814 -0.0101 -0.0321

(0.0136) (0.0434) (0.0153) (0.0682) (0.0170) (0.0883)

In US -0.0703*** -0.0703*** -0.0243*** -0.0243*** 0.0116 0.0116

(0.00640) (0.00640) (0.00780) (0.00780) (0.00847) (0.00847)

Education * Funding outcome ratio -0.0327 0.0496 0.0244

(0.0457) (0.0729) (0.0949)

Constant 0.551*** 0.550*** 0.437*** 0.439*** 0.352*** 0.353***

(0.00965) (0.00977) (0.0151) (0.0153) (0.0194) (0.0197)

Observations 34,577 34,577 21,193 21,193 15,233 15,233

R-squared 0.010 0.010 0.006 0.006 0.005 0.005

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

(29)

29

6. Discussion & conclusion

The main question was: do previous funding outcomes affect the decision to continue or stop funding projects? On the one hand, no effect was expected, since funders do not have any financial loss, and the projects and project initiators are operating completely independent of each other. All funders lose in case of a cancelled project. On the other hand, evidence on competitions indicate that previous outcomes do have an effect on future decision-making. To answer this question, this thesis collected data on crowdfunding platform Kickstarter and the social network LinkedIn, and performed an OLS regression. The results indicate that negative funding outcomes do increase the probability to stop funding projects with 12-15%.

The question is: how can this finding be explained? It might indicate that funders experience a funding outcome as a win or loss. Therefore the plausibility of the mechanisms from the competition outcomes on subsequent decision-making are discussed.

Expectation-based reference point theory predicts that a negative funding outcome leads to the decision to stop funding to avoid disappointment (Abeler et al., 2011). This is a plausible explanation: although there is no loss relative to the status-quo, there could be a loss compared to the funders’ expectations. Although the mechanisms of making up for prior losses (break-even effect) and taking more risk with gained money (house-money effect) could explain relationships between competition outcomes and subsequent behavior in context of competition outcomes (Thaler and Johnson, 1990), these mechanisms are not able to explain the findings of this thesis, since the results indicate the experience of a loss, although there is no real financial loss or gain involved.

Fear of failure theory predicts that people are more likely to stop crowdfunding after a negative funding outcome to avoid shame when failing (Elliot and Church, 1997). This explanation has been considered likely in effort-based competitions, because the result of winning and losing can be attributed to the participant (Buser, 2015). However, it seems an unlikely explanation for the findings of this thesis because the funder is largely dependent on other funders to obtain a positive funding outcome.

Ego utility theory suggests that people derive utility from positive beliefs of themselves, even when they are not realistic. Consequently, people avoid decisions that confront them with evidence that these positive beliefs are not true. Despite the fact that the findings of this thesis are consistent with this

(30)

30 theory, it seems an unlikely explanation because negative funding outcomes seem more likely to affect beliefs about other funders than about beliefs of themselves.

Mental accounting theory assumes that people create budgets in their mind, and rely on these budgets when making financial decisions (Thaler, 1985). In the literature review the possibility of a mental budget for crowdfunding was discussed. The existence of such a mental budget is in contrast with the findings of this thesis: people are more likely to stop crowdfunding after a negative funding outcome, while a negative funding outcome leaves more money in the crowdfunding budget than a positive funding outcome.

To summarize, you might conclude that the expectation-based reference-point theory gives the most plausible explanation for the results. Funders are disappointed because their personally elected project is cancelled.

Existing crowdfunding research did not take into account that funding outcomes could be related to the decision-making of funders. Existing theory implicitly assumes that funding outcomes are independent (Mollick, 2014; Belleflame, 2013). The findings of this thesis suggest that they should reconsider their assumption that investments in a crowdfunding projects are independent rational choices.

The second question was: does top education decrease the effect of previous negative funding outcomes on the subsequent decision to stop funding projects? A decrease in effect is expected because existing literature shows that higher educated people have longer planning horizons (Lusardi & Mitchell, 2007), which could lead to framing of funding outcomes in larger context, and therefore decrease their effect on short-term decision-making. Contrary to this expectation, this thesis finds that there is no significant difference in the effect of previous outcomes on subsequent decision-making.

Limitations to the current research originate mainly from the limitations of the dataset. One problem in the dataset is measuring the decision to stop funding projects. The dataset is based on one crowdfunding platform (Kickstarter). When a funder has moved to another crowdfunding platform (e.g. IndiGoGo), this is incorrectly marked as a decision to stop. The alternative story that people switch crowdfunding platform after a negative funding outcome cannot be eliminated. Another limitation is the research design; most research on previous outcomes and subsequent decision-making has been done using experiments (e.g. Gill and Prowse, 2010), which allows for more reliable estimation of causal

(31)

31 two variables were the only endogenous variables, the results would still likely be biased because our data only allowed for two binary control variables.

The results might have implications for crowdfunding platforms balancing their efforts between two options: getting more projects or getting more funders. The first option, getting more projects on the platform while keeping the number of funders constant, could lead to more negative funding outcomes, because the funding is spread more thinly across the projects. Our results suggest this leads to an increased probability that funders stop funding. Subsequently, this could potentially lead to increase in negative funding outcomes. This could lead to a negative spiral: the thinner spread of funding outcomes could lead to more negative funding outcomes, which might lead to more funders to stop funding. The second option, increasing the number of funders while keeping the number of projects constant, could lead to more successfully funded projects. Since crowdfunding platforms profit from successfully funded projects, our results might imply that a higher emphasis on getting more funders on the platform is more profitable for crowdfunding platforms.

Potential future research could address the limitation of this study. The data could not indicate if funders stopped participating in crowdfunding, or moved to another crowdfunding platform. Additionally, the findings are applicable to the reward-based all-or-nothing model. Future research could address the question whether our findings differ for funders in other crowdfunding models.

(32)

32

7. References

Abeler, J., Falk, A., Goette, L., & Huffman, D. (2011). Reference points and effort provision. The American

Economic Review, 470-492.

Agrawal, A., Catalini, C., & Goldfarb, A. (2011). The Geography of Crowdfunding. SSRN Electronic Journal, 1–57. doi:10.2139/ssrn.1692661

Allan (2013). Django Kickstarter Scraper. Accessed from https://github.com/neight-allen/django-kickstarter-scraper#db-dump.

Apicella, C. L., Dreber, A., & Mollerstrom, J. (2014). Salivary testosterone change following monetary wins and losses predicts future financial risk-taking.Psychoneuroendocrinology, 39, 58-64.

Beattie, J., & Baron, J. (1990). Psychological determinants of decision attitude. Science magazine.

Bell, D. E. (1985). Disappointment in decision making under uncertainty. Operations research, 33(1), 1-27. Belleflamme, P., Lambert, T., & Schwienbacher, A. (2014). Crowdfunding: Tapping the right

crowd. Journal of Business Venturing, 29(5), 585-609.

Burtch, G., Ghose, A., & Wattal, S. (2013). An empirical examination of the antecedents and

consequences of contribution patterns in crowd-funded markets. Information Systems Research, 24, 499– 519. doi:10.1287/isre.1120.0468

Buser, T. (2015). The Impact of Losing in a Competition on the Willingness to seek Further Challenges. Campbell, J. Y. (2006). Household finance.The Journal of Finance,61(4), 1553-1604.

Crowdsourcing.org. (2012). Crowdfunding Industry Report. Retrieved from http://www.crowdfunding.nl/wp-content/uploads/2012/05/92834651-Massolution-abridged-Crowd-Funding-Industry-Report1.pdf

Cumming, D. J., Leboeuf, G., & Schwienbacher, A. (2014, June). Crowdfunding Models: Keep-it-All vs. All-or-Nothing. In Paris December 2014 Finance Meeting EUROFIDAI-AFFI Paper.

Du, W., Green, L., & Myerson, J. (2002). Cross-cultural comparisons of discounting delayed and probabilistic rewards. Psychological Record, 52(4), 479-492.

Elliot, A. J., & Church, M. A. (1997). A hierarchical model of approach and avoidance achievement motivation. Journal of personality and social psychology, 72(1), 218.

Fernandes, D., Lynch Jr, J. G., & Netemeyer, R. G. (2014). Financial literacy, financial education, and downstream financial behaviors. Management Science,60(8), 1861-1883.

Galai, D., & Sade, O. (2006). The “Ostrich Effect” and the Relationship between the Liquidity and the Yields of Financial Assets*. The Journal of Business. doi:10.1086/505250

Gerber, E. M., & Hui, J. (2013). Crowdfunding: Motivations and deterrents for participation. ACM

Transactions on Computer-Human Interaction (TOCHI),20(6), 34.

Gerber, E. M., Hui, J. S., & Kuo, P. Y. (2012, February). Crowdfunding: Why people are motivated to post and fund projects on crowdfunding platforms. InProceedings of the International Workshop on Design,

Influence, and Social Technologies: Techniques, Impacts and Ethics.

Honohan, P. (2008). Cross-country variation in household access to financial services. Journal of Banking

(33)

33

Karlsson, N., Loewenstein, G., & Seppi, D. (2009). The ostrich effect: Selective attention to information. Journal of Risk and Uncertainty, 38, 95–115. doi:10.1007/s11166-009-9060-6

Kleemann, F., Voß, G. G., & Rieder, K. (2008). Un (der) paid innovators: The commercial Utilization of Consumer Work through Crowdsourcing. Science Technology and Innovation Studies, vol. 4, pp. 5-26, 2008.

Köszegi, B. (2006). Ego utility, overconfidence, and task choice. Journal of the European Economic

Association, 4(4), 673-707.

Kuppuswamy, V., & Bayus, B. (2013). Crowdfunding creative ideas: the dynamics of projects backers in Kickstarter. SSRN Electronic Journal. Retrieved from

http://funginstitute.berkeley.edu/sites/default/files/V.Kuppuswamy_Crowdfunding - UCBerkeley.pdf Lin, M., & Viswanathan, S. (2013). Home Bias in Online Investments: An Empirical Study of an Online Crowd Funding Market. Available at SSRN. doi:10.2139/ssrn.2219546

Lusardi, A., & Mitchell, O. S. (2007). Baby boomer retirement security: The roles of planning, financial literacy, and housing wealth. Journal of monetary Economics, 54(1), 205-224.

Mollick, E. (2014). The dynamics of crowdfunding: An exploratory study. Journal of Business Venturing, 29(1), 1–16. doi:10.1016/j.jbusvent.2013.06.005

Mollick, E. R., & Nanda, R. (2014). Wisdom or madness? Comparing crowds with expert evaluation in funding the arts. Comparing Crowds with Expert Evaluation in Funding the Arts (June 20, 2014). Harvard

Business School Entrepreneurial Management Working Paper, (14-116).

Schwienbacher, A., & Larralde, B. (2010). Crowdfunding of small entrepreneurial ventures. SSRN Electronic Journal, 2010. Retrieved from

http://lillebørs.dk/wp-content/uploads/edd/2013/06/Schweinbacher-Larralde-2010.pdf

Simon, R. (2014, April 14). Frustration Rises Over Crowdfunding Rules.Wall Street Journal.

Shafir, E., & Thaler, R. H. (2006). Invest now, drink later, spend never: On the mental accounting of delayed consumption. Journal of economic psychology,27(5), 694-712.

Park, C., & Jun, J. K. (2003). A cross-cultural comparison of Internet buying behavior: Effects of Internet usage, perceived risks, and innovativeness.International Marketing Review, 20(5), 534-553.

PwC. (2015, February 1). Peer pressure - how peer-to-peer lending platforms are transforming the industry. PwC.

Thaler, R. (1985). Mental accounting and consumer choice. Marketing science,4(3), 199-214.

Thaler, R. H., & Johnson, E. J. (1990). Gambling with the house money and trying to break even: The effects of prior outcomes on risky choice.Management science, 36(6), 643-660.

Van Rooij, M., Lusardi, A., & Alessie, R. (2011). Financial literacy and stock market participation. Journal

(34)

34

Appendix I

(35)

(36)

The effect of crowdfunding outcomes on subsequent funding decisions