Tuesday, April 08, 2008

Ezra Klein thinks highly of the Ray Fair forecasting model for presidential elections.

I don't

I think it is incorrect to refer to Ray Fair models for Presidential elections in the singular. The fact is that the model has been rejiggered -- the specification changed.

Now I haven't kept up, and it appears that Fair no longer changes his model after every election "1 Nov 2006 ... No specification changes have been made to the equation since the update in November 1994, following the 1992 election"

This means that the model has been used to forecast 3 data points out of sample.

It is always possible to make a model after the fact that fits the past. Remember the key insight of Yogi Bera "predictions are difficult especially about the future."

There is no reason to be interested in the performance of the current model in fitting results before 1994 when it was re-specified to make sure it fit results before 1994.

Also, very importantly, the Fair model includes subjective variables. For example, he seems to have decided that the USA was at peace in 2004 ! Inventing the explanatory variables makes it very easy to fit the past.

This means that Fair's current model has made 3 predictions. One was a bit off. "by Thomas E. Mann, September 10, 2004

During the Q&A session at the Roundtable on Forecasting the 2004 Presidential Election, a distinguished member of the audience, Thomas E. Mann, W. Averell Harriman Chair and Senior Fellow at The Brookings Institution, criticized the forecasts for November made with the quantitative models presented therein, as well as by Ray Fair (all of which, and others besides, are factored into the Pollyvote).

Asked to reconstruct his remarks for Polly, Dr. Mann had this to say via e-mail:

"I made three points at the APSA panel. (1) Ray Fair's model has very little economics in it. He generates a 55.57 % Bush share of the 2-party vote simply on the basis of an incumbent Republican running for reelection after his party has been in office only one consecutive term. His three economic variables produce less than 2 points additional vote for Bush.

"Dr. Mann had this to say via e-mail:

'I made three points at the APSA panel. (1) Ray Fair's model has very little economics in it. He generates a 55.57 % Bush share of the 2-party vote simply on the basis of an incumbent Republican running for reelection after his party has been in office only one consecutive term. His three economic variables produce less than 2 points additional vote for Bush.' "

No comments: