# Anatomy of a crowdfund: after the dust settled

Last Monday, November 26^{th}, I and my team successfully concluded our 52-day grassroots campaign to raise $25,000 for a Small Science project in the area of drug and addiction research. As far as I’m aware, we set a new world record for a basic research project in the science crowdfunding space, and it was all thanks to 385 micro-philanthropists from around the world – not to mention a whole lotta hustlin’ on our part.

As we transition from campaign mode (Crowdsourcing Discovery) to research mode (Crowd4Discovery) in the weeks ahead, I want to take stock of exactly how far we’ve come in the form of a retrospective study. As I recently tweeted, the first scholarly publication to result from Crowd4Discovery will be an analysis of crowdfunding metrics. Basically, all of my weekly campaign dispatches will be condensed into a single citable document.

This exercise achieves two aims. First, I want to create an official “how-to” guide that demystifies the crowdfunding process for aspiring crowdfunders in the sciences and other creative endeavors. Second, I want to preview the open process by which we will assemble future scholarly publications borne out of the actual research our team will conduct in the next few months.

Let’s start by taking a look at the distribution of donors across the eight donation levels spanning $10 to $1,000:

In percentage terms, over half of our donors contributed at the $25 level. Why was this our most popular donation? Well, we offered a pretty nifty tangible reward: a 3D printed methamphetamine model. But it also turns out that the median donation for many successful campaigns is $25, according to stats disseminated by crowdfunding site blogs.

However, small-dollar donors are only one part of a winning albeit lopsided coalition:

For example, $5-$49 donors (in blue) constitute 63% of all donors but they fund only 21% of our $25,000 goal. On the other hand, $200-$1,000 (in yellow) donors make up only 7% of all donors but contribute 35% of all donations in dollar terms.

So what did a typical day look like? As shown in this histogram, on most days we had fewer than 5 donors:

Along the bottom axis is the number of donors in a day. So according to this histogram we had five 1-donor days; ten 2-donor days; eight 3-donor days; and so on. There were only a dozen days when we had anywhere near double-digit donor “turnout.” And the outlier all the way to the right is the last day of our campaign.

To get a better appreciation for the really active days, take a look at these plots, where each orange circle is a day in the life of our campaign:

Again, the outlier is the last day of our campaign, which was unlike any day preceding it. Excluding the final 48 hours, the average daily haul was $367. Including every day boosts that average to $482, just above the magic number $481, which is what you get when you divide our campaign goal by the length of our campaign.

Digging a little deeper into the demographics of our donors, it turns out that 70% of our supporters are men, at least 25% are scientists, and roughly half are strangers or 2+ degrees removed. But can we say anything more specific about the composition of our donor pool?

In the aforementioned weekly campaign dispatches, I’ve been visualizing my Facebook network as a graph, with friends represented as nodes (circles) and friendships represented as edges (lines). The size of each circle is proportional to its connectedness, or degree, in the network. Let’s look at the distribution of donors (in yellow) across my Facebook network:

As I noted before, the distribution of donors doesn’t appear to follow any obvious spatial pattern, at least from this bird’s eye perspective. For example, it doesn’t appear to be the case that my most highly connected friends donated larger sums to our campaign. However, something interesting happens when you break up my Facebook network into friendship clusters, which correspond to different life phases or shared interests:

Donors are in yellow, and the fractions indicate the number of donors in each cluster. Several patterns emerged, at least to my eye. First, while the total donation turnout from my entire Facebook network is 17%, individual clusters donated either more or less or the average. Second, there appear to be at least two qualitatively distinct cluster topologies, or configurations in plain speak. On the one hand, the center-left cluster is a tight-knit core group of friends with more loosely connected individuals radiating outward. An extreme version of this topology is the bottom-left cluster. On the other hand, the center-right cluster is anchored by a central hub to which several smaller, non-mutual spokes of friends attach. In graph theory speak, what’s different about these two kinds of clusters is the average degree.

Does the average degree of donors differ from the average degree of their cluster as a whole? In other words, do yellow nodes tend to be larger or smaller in some clusters versus others? In some instances, the answer seems to be yes. Take a look at the two clusters below:

The average degree and total size of these two clusters is comparable. Yet the donors belonging to the right cluster have an average degree that is nearly twice the average degree of the entire cluster, meaning our message appealed more to core friends rather than to peripheral friends. On the other hand, the donors belonging to the left cluster have an average degree that is approximately equal to the average degree of the entire cluster, meaning our message appealed to core friends and peripheral friends the same. So content and connection both matter, depending on where friends reside in the network.

Now there are two caveats I’d be remiss not to mention. First, each friend in my Facebook network doesn’t see my posts equally given Newsfeed preferences and Facebook’s secret algorithm for presenting and prioritizing content. Second, some fraction of people in general are Facebook lurkers, which means they miss a lot of shared content.

I’ll keep plugging away at this Facebook dataset, and soon I’ll begin analyzing the donor distribution in my Twitter network. I’d love to get feedback from network aficionados out there!

For a summary of the final 24 hours, please go here.

For a summary of week 6, please go here.

For a summary of week 5, please go here.

For a summary of week 4, please go here.

For a summary of week 3, please go here.

For a summary of week 2, please go here.

For a summary of week 1, please go here.

And for a summary of the first 96 hours, please go here.