For about the past ten years, every election cycle I would create an algorithm to predict the outcome of the gains and losses of the Dems and GOP in the House and Senate elections. It was a back-of-the-envelope experience.
I never did record the algorithm or keep any notes. Every cycle I created, de novo, a new algorithm. (This year I had in my head a highly simplified and abridged version I could compute — more or less.)
So last cycle, I decided to keep a record of the algorithm. Last night, I remembered the password I put on the document — which I had been attempting to do for months — opened the file and dusted off the algorithm.
One big problem became apparent — I kept the algorithm, but did not note which data I used to input into the algorithm or the adjustments I know I made to the data I settled on using.
(Some years, the algorithm was one seat off in either the House or Senate results. In 2008, my algorithm was 94% accurate for predicting the Dem-GOP seat distribution in the U.S. Senate, and 89% accurate for predicting the same in the U.S. House.)
Since opening the document, I have spent hours running various data sets and adjustments to the data through the algorithm, trying to remember which data set and the adjustments I used last cycle.
I woke this morning with further choices and final adjustments to the data in mind.
(These choices about which data to use and the adjustments to the data — algorithm inputs — are where math and science ends, and the art begins.)
Bottom line, I can not remember which data set or adjustments I used last cycle — so please do not use prior, unpublished results to imbue accuracy to this algorithm or to the adjusted data set I am using.
Further, because I was torn between two different adjustments to one key data point of the most powerful variable in the algorithm, I decided to use one data set for the House predictions and a wildly different data set for the Senate prediction — to see which was more accurate.
In fact, most analysts will find these results unbelievable, and the more charitable would likely say the algorithm’s predictive outcome may be possible, but not probable or likely.
Having said that, here are the results of Perrin’s Predictive Election Algorithm:
Dems lose 73 seats in the U.S. House and lose 10 seats in the U.S. Senate.
On Monday, November 1st, the day before the election, I will survey all available new data, and publish the algorithm’s final predictive results.
Sometime, when all the election results are finally settled, I will review the accuracy of my algorithm and whether I made the right choices in the data set used, and my adjustments to it.