03 November 2006

Election predictions, part 1

As a student of politics I am often asked for my predictions on what's going to happen in various elections. Watching the huge returns come back for the Republicans in the midterm elections in 1994 was my first real taste of Election Night drama. I've been hooked ever since.

Before I do that I want to talk a little bit about the predictions I've made in the past as well as how I go about making my predictions (this part 1), and then specifically some of the history behind midterm elections (part 2). Finally, I'll give you my predictions for Tuesday (part 3).

In 2000, I predicted Bush would defeat Gore. I don't have my specific electoral college numbers handy, but I was off a little bit. Of course, no one expected the protracted battle of voting counting in Florida.

In 2004, I made my predictions in a contest at The Hedgehog Report. In that case, I correctly predicted Bush to beat Kerry, I nailed the Electoral College count exactly, and I was within 1% of the correct popular vote. As you will also see, I correctly predicted the eventual makeup of the U.S. Senate.

I don't have a specific formula for picking the winners, but I do think its important to look at a wide variety of polls, as many polling organizations use different methodologies. These differences are important and will reflect differences in the results they achieve. I do tend to stay away from polls generated by the campaigns or parties.

Trends are also important. Is a a candidate surging in the last days before the election? Or is he or she flat in the polls for months at a time? These are important characteristics that ought to be considered.

Polls often have three categories: candidate a, candidate b, and undecided. There is a natural base of support for any particular major party candidate based on those voters who consistently vote the party line. Let me give you example based upon the precinct where I live, State College Southeast (precinct 21).

Let's say that we had a candidate running for a town council seat which (for the sake of simplicity) consisted of this particular precinct. First, we need to look at some historical data:

For the 2005 municipal elections, there were 982 registered voters, and 403 who cast votes (41.04%). So in theory, despite their being 982 total registered voters, we really only need about 202 votes to win (again, this refers to municipal elections; midterm and presidential elections will have significantly increased turnout).

To get to 202, we first start with our most hardcore voters, which are those that vote the straight party ticket. In that election, 35 Republicans did so (8.68%). This is not insignificant. Next we look at all of the individual races within the precinct. We're looking for a base vote, basically, the lowest percentage achieved by a Republican in any reasonably competitive race. In fact, there are three races where Republicans achieved 30-31%, which we can assume is our fbase. This tells us that assuming our candidate should be able to count on at least 30%, or 121 votes (the 35 straight ticket voters are part of this number).

So our goal of 202 isn't nearly as far when we can count of 121 votes already. If we assume that the Democrats have a similar base, then when we subtract these 121 votes (x2) from our 982 registered voters, we are left with 740 votes. The statistics show us that only about 160 of these people will vote. Half plus one of those 160 votes is 81. 81 in addition to our 121 gives us 202 and the win.

I know this is oversimplified, but what I'm trying to demonstrate is that even in a precinct of nearly 1,000 registered voters, the results will often be determined by as few as 80 people. These undecided voters are the key to any election and it is often those undecided that determine who wins. So it is important to determine what sort of people make up those undecideds so that you can make an accurate guess at which way they're going to break.

A few caveats. If a candidate is consistently polling at 50%+, you can throw out all of the above analysis. Consistency above 50% in the polls (especially in polls with different methodologies) is a pretty easy method of picking a winner. Of course, this rarely happens in the closest of races. Neither Bush nor Kerry polled above 50% in 2004. On the other side, a candidate who has yet to break the 40% barrier so late in an election is doomed to fail.

When you realize that in 435 congressional races only 40-60 are usually competitive, it is because the large majority of the races have one or both of these caveats. One of the candidates is consistently polling above 50% while the challenger fails to reach 40%. These elections are an easy call on Election Day, and are referred to as "safe". The 40-60 tough calls are divided into further categories, such as likely Republican (or Democrat), leans Republican (or Democrat), or toss up. The fate of the election, as we will see, lies in these races.
Post a Comment