Sunday, September 7, 2008

Methodology

Unfortunately, what kind of polling atmosphere we will have this election is as of yet unknown. I have an ideal methodology, however weather or not I can use it will depend on how many polls come out and from whom. In short, our methodology is as follows.
All the polls from the past three days are averaged, with no one pollster being allowed to carry more then half of that average. Polls from four to seven days ago are also averaged, and given a weight of half. The success of this, of course will depend on how many polls come out every week. If two pollsters do daily rolling polls, this is super, however if we are limited to just one, then our methodology will be difficult to do. Currently, while the polls are sparse, we are using a modified methodology that weights polls from the further past more then we otherwise would.

We take these polling numbers and put them into UBC’s 2008 election forecaster. Using a special method that allows us to increase the vote of each party without swinging it from another party, we then raise or lower the vote of that party so it matches our poll average. We will also take into account the projections offered by democratic space blog, and may give this it’s own weight in our calculations.

Lastly, we then take into account each riding. We will check election prediction blog and look at the “swing” ridings, seat by seat. We then make a judgment call (a human one) based on the information available, and the margin of victory in the simulation. I feel this addition of a human element to be critical to the success of any simulation.

In the end, what we end up with is a prediction that takes into account all the elements needed to properly “see into the future”.

September 07 projection

And our projection for today.