Of 1063 survey takers, 981 reported a karma score, and 466 reported that it was zero (Yvain instructed users to put in zero if they didn’t have an account). I’m actually surprised there weren’t more zeros.
Which is why you do exploratory data analysis, and look at your distributions first.
Me, I’m surprised that people who don’t post bothered to take a long survey. You have to have an account to post, right? Do you have access to the system data, so that you could compare the distribution of karma scores for registered users versus the poll results?
Next suggestion—I’d probably make separate analyses based on LessWrongUse—never posted and never registered as one class , posted and more (interactive users) as another. And substitute rank(karmascore) over the whole population for karmascore.
Which is why you do exploratory data analysis, and look at your distributions first.
Me, I’m surprised that people who don’t post bothered to take a long survey. You have to have an account to post, right? Do you have access to the system data, so that you could compare the distribution of karma scores for registered users versus the poll results?
Next suggestion—I’d probably make separate analyses based on LessWrongUse—never posted and never registered as one class , posted and more (interactive users) as another. And substitute rank(karmascore) over the whole population for karmascore.