Wednesday, September 22, 2004

Delphi on Colorado: FasTracks.

As promised I will begin to unveil how I come up with my projections, starting first with FasTracks.

If you click here you will be taken to a graph that shows the original graph that I did using just the Guide the Ride poll data from 1997 and the final election result.

Why even look at a referendum that failed 7 years ago. Well, in business you access things, such as how much you should pay for a company, or what the fair price of an IPO stock is by looking at comparables. FasTracks is, in most ways, a repackaged Guide the Ride. Therefore, Guide the Ride is the most apt example we have of a FasTracks like referendum.

A word on reading the graphs. At the top is the actual regression equation. In all cases C2 refers to the percentage vote in favor of Guide the Ride and C1 refers to the number of days before the election. The equation that I produced explains about 90% of variation in the expected vote percentage based simply upon how far the referendum is away from election day. It also predicted that Guide the Ride would fail with 46.67% of the vote.

As we have begun to look at FasTracks, one thing is certain, it, like other referendums before it, will lose support before election day. The question is, will the decrease in support level off before election day, or will it increase? And will it matter if the support for FasTracks is in the 60s. The regression that I had built using the same quadratic equation as the Guide the Ride equation proved unsatisfactory to me. If you click here you can see why. I can only explain about three quarters of the variation in the model, and the points are not very well defined around the regression line.

This lead to the most recent incarnation of the FasTracks model, which is a cubic equation. As such it fits the data more exactly, explaining almost 80% of the variation in the model.

Some questions remain unsolved. Should I weight the election result in '97 more heavily than other poll results. I probably should since an election should have a much higher statistical significance than a poll. However, I haven't weighted the 42% vote in 1997 more heavily than any other poll. The reason is that I have heard much criticism over 1997 being an off year election, and that more pro FasTracks people will show up in an presidential election year. While that isn't what the Guide the Ride supporters were saying in 1997, I can respect that today. So I figure that by not weighting the vote in '97, I am in fact, correcting for the increased voter turnout in this presidential election year.

So now the Mile High Delphi prediction is that FasTracks will lose with 47% of the vote. However, three or four more polls in the 60s could swing the model around to predicting a narrow victory.

Hope you enjoy the graphs, tomorrow I will put up our graphs for the Senate race and for one or both of the US House Districts.

No comments:

Post a Comment