Wednesday, June 01, 2011

Early predictions for the school board election give Cannon 4-to-1 odds (Update 2)

Updated 2011-06-02 11:30 to clarify that, for the predictions, “cross-file” means “cross-file and then succeed in the primary to get both parties’ nominations.” —Tom

Updated 2011-06-02 18:13 to revise predictions after fixing a data-entry error that said Cooper had won both Democratic and Republican nominations in the primary. (He won only the first.) As a result of the change, the models give Cooper lower chances of winning a seat, with the difference being distributed to the other candidates and improving their chances. Cannon’s odds, for example, are now increased to 6-to-1 for winning a seat. —Tom


Let’s take a look at the recent primary election and see what it says about the race for the school board. This year, five seats are up for election, and six people are running for them. That means when the music stops playing, somebody is going to be left without a seat.

Of the candidates, three already have seats on the school board (Elaine Cappucci, Lawrence Lebowitz, and Edward Kubit), and three are newcomers (James E. Cannon, William Cooper, and Scott Goldman). Of the newcomers, the latter two have been endorsed by incumbents. Cannon, however, is running as a challenger to the current board’s direction. That makes him interesting.

Another thing that makes Cannon interesting is that he didn’t cross-file for an attempt to appear on both Republican and Democratic tickets. Politically savvy people have told me that this was a tactical mistake. Candidates who cross-file and then win both parties’ nominations in the primary are believed to get more votes in the general election. Passing up this bonus, then, could be costly.

How costly? Let’s find out.

Take a look at the following plot. It shows the outcomes of the 2007 and 2009 general elections for the school board. I’ve replaced votes with standardized scores that make the two elections more comparable.

Candidates who cross-filed and then won both parties’
nominations tended to do better in the general election.

Looking at these elections, the “cross-filing bonus” seems to be real. As a group, cross-filers who appeared on both tickets got about one standard deviation’s more votes than their single-ticket competitors. One standard deviation is a lot: most candidates weren’t separated from their neighboring rivals by that much.

So, given that Cannon didn’t cross-file, but also given his strong showing in the Republican primary, what can we say about his chances in the 2011 general election?

I’d say he’s got a pretty good shot. I created three simple statistical models based on the last two elections and used them to predict the outcome of the 2011 general election. All of them predict that Cappucci and Lebowitz are the most likely to win seats, that Cooper is the least likely, and that in between there’s effectively a three-way tie between Cannon, Goldman, and Kubit. All in all, I give Cannon 4-to-1 6-to-1 odds of winning a seat.

Here are the models’ predictions:

  Model M1Z                 Model M2Z                 Model M3Z

# candidate prob_elected    candidate prob_elected    candidate prob_elected
1  CAPPUCCI 99%              CAPPUCCI 99%              CAPPUCCI 99%
2  LEBOWITZ 99%              LEBOWITZ 98%              LEBOWITZ 99%
3    CANNON 86%                CANNON 94%                 KUBIT 86%
4   GOLDMAN 84%                 KUBIT 94%               GOLDMAN 86%
5     KUBIT 83%               GOLDMAN 94%                CANNON 86%
6    COOPER 49%                COOPER 22%                COOPER 44%

The percentage after each candidate’s name gives the predicted probability of that candidate winning a seat.

These predictions, as you might expect, shouldn’t be taken too seriously. They are based on models that, for obvious reasons, can’t account for chance and, for simplicity reasons, don’t try to account for anything but nomination status and performance in the primary election. And when a model predicts that something is 99% likely, we shouldn’t believe it; it’s way more than 1% likely that model itself is wrong. Perhaps most importantly, November is a long way off. There’s plenty of time for candidates to shake things up – or flame out.

Still, the models manage to explain most of the variation in the 2007 and 2009 elections, so they are likely to offer some insight into the upcoming general election. I think I’m safe in saying that Cannon’s got a pretty good shot of winning a seat.

What do you think about my models’ predictions? How do they compare to your own?

Labels: , , ,

Bookmark and Share

4 Comments:

Blogger Bill Matthews said...

I think Cannon has very good odds - if for no other reason than his predisposition to kool-aid. James has many other fine attributes AND he passes my litmus test.

Cappucci and Lebowitz also pass my litmus test - they have demonstrated the ability to listen.

This leaves Cooper and Goldman, both of whom deserve an opportunity to serve.

June 01, 2011 1:51 PM  
Anonymous John David Kendrick said...

Tom,

Would you please run your model on the last three elections ( I am referring to the last three elections regardless of whether they were a primary or general election) and report on the forecast accuracy?

June 02, 2011 2:11 AM  
Blogger Tom Moertel said...

John,

I have the data for the last two elections, so here are the predictions for those elections for each of the three models: 2007 and 2009 predictions.

All three models predicted the 2009 election fairly well. None of them predicted Walton not getting a seat in 2007, however.

Cheers,
Tom

June 02, 2011 2:25 AM  
Blogger Tom Moertel said...

John,

To clarify, the models don’t predict primary-election outcomes. They predict general-election outcomes, given filing status and the primary-election outcomes as input.

To give you a better idea of how the models work, here’s a graphical representation of the M1Z model and a graphical representation of its predictions for the 2011 general election.

Cheers,
Tom

June 02, 2011 2:39 AM  

Post a Comment

<< Home