[MNJohnnie]

Of course this is based on the assumption that the polling is accurate and not as grotesquely inaccurate as the Wisconsin recall polls. I suspect 2012 will be like 1980. A year where the pollsters are left red faced on Election eve trying to explain why their projections are so far off from reality.

[Political Junkie Too]

Two things addressing this concern: 1) I've been limiting myself to only Rasmussen polls, and 2) I've been showing the plot on a weekly basis for all to see how the polls have changed over time.

In 2010 when I was doing this for the Key House Races, it was interesting to see how the composite polls from the "experts" diverged until a few weeks before the election, when they suddenly all came together. Some people attribute it to the undecideds remaining on the fence until the end. I attribute it to the "experts" astroturfing one side for as long as possible in the hope of swaying undecided voters to join the supposed masses, and then scrambling at the end to match reality.

We shall see.

-PJ

[Political Junkie Too]

One more thing...

I used to test the sensitivity to bias by runing a variant where I presumed a +1% bias towards Democrats, so I subracted 1% from the Democrat result and added 1% to the Republican result. If I do that to the current Rasmussen presidential poll, Romney goes from an EV of 257.37 votes to 276.80 votes, or from a 28.95% chance of winning to 67.08% chance of winning.

Being less severe, if I use a +0.5% adjustment, Romney gets an EV of 267.82, or a 48.51% chance of winning.

If I go for broke and test that I don't know what the bias is, but I assume it's a triangle between 0% and 2.5%, with a likely at 1%, Romeny's EV goes to 279.37, with a 70.21% chance of winning.

In the Senate, let's do the same thing.

Adjusting for a 1% bias towards Democrats and a 1% bias against Republicans, the EV goes from 50.37 seats to 52.55 seats, or from a 44.5% chance of winning control to a 96.56% chance of winning control.

Being less severe at a 0.5% bias, the EV is 50.93 seats, or a 64.56% chance of winning control.

With uncertain bias as before, the EV is 52.74 seats, or a 93.26% chance of winning control.

I can play with the range of bias to test the sensitivity of the results, trying a spread with a lower high-side and different likely. When I tried this with McCain in 2008, the +/- 1% bias-adjusted result showed McCain leading the race. After that election, I stopped using the bias adjustement and just went with straight Rasmussen numbers.

-PJ -PJ