I wish I’d had a little more panache in 2013. I’d call it something like “Crowdfunding Secrets: Revealed by Machine Learning!” or “Unlocking your Kickstarter’s Potential!”

Recovered: Anniversary and Crowdfunding Analysis

Crowdfunding Secrets: Exposed by Big Data!

I apparently missed my one-year anniversary, but the blog’s been slow recently. A lot has happened in the last half-year: I’ve graduated, I’ve moved, I’ve found employment. Unfortunately, I may not be gaming as much any more, but we’ll see how that plays out long-term.

Just because I wasn’t blogging doesn’t mean I haven’t been busy, and one of the things I worked on was an analysis of Kickstarter data. Because it was for a class, it assumes a certain vocabulary, has some weird stylistic artifacts, and has some persistent errors that weren’t severe enough to merit fixing at the time. Eventually I would like to revisit this more completely, but until then I may as well “publish” it:

I don't remember my grade, but I think it was good. What's really important is that I finished it. Image by robtowne0, via Pixabay.

2023: Summarizing

I’m not going to make you wade through my decade-old academic paper for the conclusions. Here’s a rough summary.

There are three main ways projects go: failure, success, and “runaway” success.
The strongest predictor of success under a creator’s control is a low goal. Obvious in retrospect, but nice to have the data.
The strongest predictor of success to an outside observer is the number of backers.
The strongest predictor of “runaway” success under a creator’s control is number of updates. Be careful here: this is still a correlation.
The strongest predictor of “runaway” success to an outside observer is the number of comments. This is also a much stronger predictor than number of updates.
“Percent funded” is useless as a metric (compared to, e.g. “dollars funded”). I speculate that this is because Kickstarter is used almost completely as a pre-order (ex post facto) platform and not a “funding” (ex ante) platform. Which is to say, people want stuff more than they want the project to succeed independently, so the target is largely irrelevant to their decision.

2023: The Data

The crowdfunding landscape is greatly changed in the last decade. Dan Misener’s The Kickback Machine broke sometime after my project, but I have a copy of the data he shared with me for it. I’m sharing it here, in case someone else might get some useful insight from it. It contains information about ~31,200 Kickstarter campaigns, from July 2012 to April 2013.

The Data

I’ll see if I can’t provide some notes about it:

There’s some wonky campaigns at the beginning and end with incomplete data. Probably best to exclude these.
There may also have been some duplicate entries. Worth looking for before crunching.
In reward-level-list, pledge levels with an “L” afterwards are “limited quantities.” For example, “only one pledge at this level.” The limits themselves are not recorded.
I believe kickstarter_fee is a simple percentage.
Where currency is blank, assume USD. I do not recall the calculation of pledged_in_usd, so maybe it’s best to look up historical conversion rates, for consistency.
Many of these values are not linearly independent. percent_raised is a product of goal and pledged. success is also, but project_state is not.

I believe at the time, I had been playing around with what is now OpenRefine to clean up the data, as well as the built-in tools of SIMCA-P+. SIMCA is fine, but it looks like the strongest parts of it are in R now.

Image by ptksgc, via Pixabay.