Your momma still loves your #SciFund Project

Your momma still loves your #SciFund Project

Note: You have only a scant few days to sign up for #SciFund Round 2!!! Do it now if you’re interested!

So, something was still bothering me after my epic social network and science crowdfunding analysis. It’s the Facebook like thing. My initial choice of using it seemed to make sense. But the more I thought about it, the more I wondered what Facebook Likes really mean. Are they a reflection of your personal network? Your ability to reach-out? What?

I don’t think it is a clear signal of your personal network. And yet, everyone in #SciFund talked about tapping their own personal network. Family and friends were often great sports, helping out with the early phases of people’s funding.

And yet…there’s this. If you run an analysis looking, at, say, pre-goal pageviews, the signal of number of Facebook friends appears absent.

Analysis of Deviance Table (Type II tests)

Response: preGoalPageviews
                 LR Chisq Df Pr(>Chisq)
TwitterFollowers   13.158  1  0.0002864 ***
FacebookFriends     3.181  1  0.0744827 .
Facebook.likes     74.071  1  < 2.2e-16 ***
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

But then it struck my indirect-effect-loving mind, maybe your personal network doesn’t lead to pageviews. It’s smaller. Maybe it just mainlines directly into contributions!

So I did a test to look at how these different measures of social media affect number of contributors. I used the same generalized linear model as before looking at pre- and post- goal pageviews influencing number of contributors. But, this time I added Twitter followers, Facebook likes, and, most importantly, Facebook Friends. And, *boom*. There’s the signal from yo momma’.

Analysis of Deviance Table (Type II tests)

Response: Contributors
                  LR Chisq Df Pr(>Chisq)
TwitterFollowers     0.062  1   0.804128
FacebookFriends      8.752  1   0.003092 **
Facebook.likes       0.058  1   0.808924
preGoalPageviews     7.189  1   0.007334 **
postGoalPageviews  112.753  1  < 2.2e-16 ***
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Note that these other measures of ‘fanbase’ don’t affect contributors directly. It’s all indirect via pageviews. The relationship that I pulled from the coefficients is that 50 Facebook Friends = 1 Contributor. Pre-Goal pageviews stayed about the same at 0.078 contributors per view while Post-Goal views were still at 0.051 contributors per view.

So what it looks like is that my previous story – that a fanbase built by a blog leading to more Twitter followers and Likes leads to more pageviews, hence more contributors, and finally money for research is only PART right.

Friends & Family matter. They have a direct connection to contributions, bypassing the whole social-networky-fanbasey stuff. We can see that here:

Note that to have a high number of contributors a project needed to have either a) a lot of pre-goal views, b) a high number of post-goal views (more red), c) a large number of friends (larger point size), or some combination of all three. Note that I’ve split things into projects that have and havenot achieved their goal for ease of visualization.

So I decided to do something unwise to put this all together. I ran an SEM. Granted, for all of these variables (and Age, as I’m still interested in Age!) we want a minimum sample size of ~150. And we only have 49. Or, really, 28 if you scrub out folk that did not answer our Facebook or Twitter questions. I decided since this analysis was going to be seat-of-the-pants and exclude a valuable outlier (the high pageview outlier always causes the SEM likelihood algorithms to bork out on my), I’d assume not answering was the same as an answer of 0. This bumped my sample size up to 43. Um. Sure. Why not. Just for the fun of it.

I ran a model in which all of these fanbase metrics v. the personal metrics have both direct and indirect effects on contributors and a second nested model where personal metrics affect things solely through contributors whereas measures of public engagement affect things solely through pageviews. This second model fit just as well as the more saturated model (Chisq Difference = 12.909, DF=10, p=0.2288). This new model also fits the data adequately in general (Chisq = 23.267, DF=19, p=0.226).

So what’s it look like?

There’s a lot going on. The salient points are

1) Yup, Facebook Friends have a direct effect by upping your total number of contributors.
2) Twitter Followers and Facebook likes both up number of contributors indirectly, via increasing pageviews.
3) Old folk have fewer friends on Facebook.
4) We’re doing a horrible job or predicting Facebook Likes. Clearly, there’s a bigger story here that is important to understand given how much it impacts pageviews.
5) Post-Goal pageviews are still “worth” more. But, again, very few of these are above 0. I need to do some ZINB or ZIQP stuff here.

So there you have it. Both your ‘fanbase’ and friends are important for crowdfunding your hot science.

Do don’t worry, your momma loves you! And your research!