Page 144 - 臺大管理論叢第32卷第2期

The More, the Merrier? The Bystander Effect on Crowdfunding Platforms

performance (C3), entertainment (C4), writing and publication (C5), society and culture
(C6), technology application (C7), food and drink (C8), and travel (C9). Different types
of projects may have different impacts on the daily amount of fundraising. While product
design and technology application projects usually deliver tangible products, society and

culture projects typically propose donations. For instance, a crowdfunding project creator,
Milk House in the food and drink category, delivers milk to backers. Such a reward-based
project not only attracts investors who have the same belief with the firm but also attracts

investors who are interested in high-quality fresh milk. Another example is Ponzi Game
in the product design category. Project creators plan to raise funds for developing board
games; funders will receive a set of the board game in return. Campaigns in the society
and culture category typically support different social causes. For example, one project
seeks funds from individuals to support the Anti-nuclear Movement in Taiwan. Another

crowdfunding campaign, a social movement called “Sunflower,” also successfully raised
funds to support their belief in reforming the government. Table 1 summarizes the key
variables in our research.

3.3 Statistical Methods
We use regression models with robust standard errors to test our hypotheses. We use
robust standard errors to consider the correlation among error terms over time within the
same project. In the regression models, we also added year, month-of-year, and day-of-

week fixed effects to control for unobservable time factors that may influence the pledge
amount over time. Specifically, for the year fixed effect, our sample spans across 2014 and
2015, so we add one dummy variable Year 2014, which equals one if an observation is 2014;

observations in year 2015 are viewed as the benchmark group. For the month-of-year
fixed effect, as our sample spans across December, January, February, and March, we in-
clude three dummy variables Month , Month , and Month , each of which equals one
Dec
Feb
Jan
if an observation is in one of these months. For the day-of-week fixed effect, we follow a
similar procedure. We include six dummy variables Day Mon , Day , Day , Day , Day ,
Tue
Fri
Thu
Wed
Day , each of which equals one if an observation is in one of these days. Category fixed
Sat
effects are added to control for unobservable category-level heterogeneity. In particular, as
we have 9 project categories in our sample, we add 8 project category dummy variables (C1
to C8) in our regression model, with observations in the travel category (C9) as our refer-

136