Sometime during graduate school, they suggest that you pretest an idea before going whole hog. I usually do, but somehow in the rush of things, I went whole hog. And as I noted in an earlier post, I’ve kind of dug myself into a hole.
That’s not exactly true. The amount of time it took me to collect this data would have been almost exactly the same if I had started out with a 2% sample, so I went ahead and got it all. And there, on the right, is a scatterplot of the average score of a little over 30,000 Slashdot posts, compared with an “affiliation score,” that is, whether folks posted among friends.
Yes, I know it looks like something, but trust me, it’s not. There is a negligible (0.025, p<0.0001, or effectively 0) correlation between posting score and affiliation. Yes, I know it looks like I could do something like a nonlinear regression, but it's an illusion. That pretty curve is described by a few outliers in among tens of thousands of individual posts. So now I'm faced with a decision. Do I take another run at this, or do I ditch it and try to resurrect earlier work (mine the dissertation) that would make sense in this context. I've gone back to the original call, and dug up my proposal, and I’ve found little to go on so far. I have a lot of data, but now I have to see if it can talk to me in any interesting ways.