Friday, April 27, 2012

How To Read Polls (Like a Partisan)

Following polls, closely, is, I think, a little like "Fantasy Football" for political junkies. You choose your team (or try to convince people you don't have one) and then look with fascination at each poll that rolls out. Trying to understand polls is complex--I'm not going to bore you with it here (although there are some good articles here, here, and here). No--I'm going to tell you how to read the polls as a partisan1.

You are a Democrat and See a Poll You Don't Like
If you are a partisan Democrat then you have a bit of a problem: you may say that the main-stream media is against you--but why isn't anyone buying it? They aren't buying it because it's flimsy. The media may, as a whole, be dedicated to seeking the truth ratings--but that "most reporters are Republicans?" I don't think too many people are going to swallow that. 

If you are faced with a poll you don't like you should do this:

  1. Is it Rasmussen or Fox? If YES, you're good. This graph is old (from 2000) but who cares: Raise your right hand and say "Rasmussen over-samples Republicans and they have a lousy track record" (link from Nate Silver). But if you want something more recent, check this out: Rasmussen Polls Were Biased and Inaccurate (also Nate Silver)

  2. What if it's someone else? That's easy: It's still Rasmussen's fault! If you can show that the poll number is an aggregate? Well, they're factoring in a barrage of Rasmussen robo-calls to knock down the more carefully done polls
Rasmussen’s polls, being automated “robocalls”, constitute a huge fraction of the aggregate polling data for any period of time. No outfit comes anywhere near producing the number of polls he does. So anyone doing an aggregate of all polls over a, say, one week period will find much more data from Rasmussen than anyone else.
You are a Republican and See a Poll You Don't Like
 This is also easy: Unless the poll is Rasmussen you do the following:
  1. It doesn't count because it's the liberal media bias! CNN? Pfft. Mainstream. MSNBC!? HAH--Lame-stream! And so on.
  2. Internals or it didn't happen. If the poll doesn't publish its internals then you dismiss it abruptly. It's trash.
  3. Check the over-sampling of Democrats. If you ask Democrats if Obama is doing a good job of course they're going to say "Yes" (Or Heil, or something, right? Amirite?). So when you look at the internals, check and see if it's a perfect D/R/I split.
  4. Likely voters vs. voters vs. adults. If the poll says it samples "likely voters" then it's the best--if their model is good. If it says it samples "voters" it's crap. It it samples "adults?" Uhhh ... why'd you bother. If' it's a web poll? urinate on it.
  5. If it's Rasmussen? People are stupid.
You Are Not A Partisan
If you are not a partisan, how do you handle polls? Easy: They, uh, don't matter this far out.
Drawn from FiveThirtyEight’s polling database, which includes thousands of surveys, here is the average of the 10 most recent national polls as of late April in each election year going back to 1972 (for the 1980 election and before, fewer than 10 general election polls conducted in the first four months of the year were available).

The leader in national polls at the end of April in the past two elections has gone on to win. Before 2004, however, the April leader lost the popular vote more often than not.

1 This is based on the material I have seen in the comments and content of various blogs.

No comments:

Post a Comment