Monday, November 01, 2004
Just flush the silver handle....
A little FYI before I get started on this one: If I link an article to the rant (click on Election Polls title above), it's probably a good idea to at least look it over, if you're going to make a coherent response.
Political polling is a very interesting practice, not only because of how it is accomplished, but also because of the value we put on it. If you read the Slate article, it gives you a bit of information about how the polls weed people out, how they might potentially affect their results, and how some of them flat out don't disclose HOW they get their numbers. I'm not going to repeat what was written in that article, because the authors did a fine job the first time.
Let's look at some figures for an election where the outcome was known:
If you look at the trendlines for the 2000 election, the numbers were in the ballpark, but multiple polls indicated that Bush had a sizable lead among "likely voters". We all know how the popular vote turned out with that one.
Lesson #1 - Polls do a pretty poor job of predicting the popular vote in close elections. We're seeing this now with polls showing changing leads from day to day. Is it really that likely that people are changing their minds on a daily basis? Probably not. It's related to that "Margin of Error" (which I will address in more detail later).
Lesson #2 - Using "likely voters" as a criteria for inclusion ignores first-time voters or people who didn't vote in the last election. In some elections, this effect might be negligible. This year, new voter registrations are exploding in key states: http://www.nytimes.com/2004/09/26/politics/campaign/26vote.html
Now, this is not to say that this favors one candidate over the other, as some have reported. We can't know that because not only are none of the national polling organizations looking at this issue, but also they are deliberately excluding people who identify themselves as newly registered or first-time voters.
It seems like this year would be particularly important to look at that trend, in light of the significant impact that those voters may have on the outcome of the election, in addition to recognizing which candidate would benefit from first-time voters. If you know about a poll that examines first-time voters (my wife is one of these), let me know, and I will post more information here.
The margin of error in the polls is also particularly interesting. Most researchers and statisticians would agree that if the results of a poll are within the margin of error, that the results are uncertain. I have read that further certainty can be added to the equation by looking at trends over a period of time (ie if one candidate maintains a lead, even if it is within the margin of error, that lends validity to the lead).
Many major polls are making significant claims about who is going win the election ( see http://www.gallup.com/election2004/showdown/ for just one of many examples) based on electoral college calculations that are within the margin of error (Florida on the paged linked above) -OR- use dated polling data (see North Carolina and Nevada on the page linked above for example). I'm not sure what the agenda is here, but it's clearly a misleading way to look at an election.
For what I believe is an less biased look at the state of the election based on multiple polls, partisan and non-partisan, trend-based statistics check out the Election Scorecard at Slate.com (http://www.slate.com/id/2108751/ Updated daily as polls come in). Feel free to share other sources of information as well.