Monday, August 30, 2010

Random sampling and you

One of the potential applications I envisioned for this site was that you out there could answer questions from the point of view of a public figure. Something like, "How would Senator X answer these questions?" We could then look at the results and do fun stuff like see how people who identify as Republican perceive certain Democrats and vice versa.

One hitch with that idea is that people would likely balk at answering the full test from somebody else's point of view. I figured the solution could come with the aid of our good friend Random Sampling. By asking users to answer a subset of the full test, I'd expect a few more responses and a better sample set.

So I tried it out. I took my latest response, which was -7.38, -5.79 (even I was surprised), and took a couple random samples. By picking five random questions from the economic set, I got values of -9.02 and -3.48, for errors of 22% and 53%, respectively, which is pretty crummy. The social set fared a little better. A random sample of eleven questions got me values of -5.67 and -5.32, for errors of 2% and 8%, which is much better.

Long story short, if I ever start doing the "how would this person answer?" stuff, I could probably get away with fewer than ten or so of each type of question and call it a day.

No comments:

Post a Comment