Sunday, September 14, 2008

More on Methodology

Questions have come up as to exactly what our Methodology is, and how we differ from other prediction websites out there? That's simple. Our projections are, at their base, a geometric projection. This is done by taking the poll numbers and applying it to the last election in a way so that if a party is polling at 10% in a province they polled 5% in last election, then all their numbers are doubled. Therefore a riding where they took 5,000 votes, we project they will take 10,000 this time.

What makes our methods different is how we arrive at those polls. Rather then use a straight 5-poll average, we average out all the polls in the past 7 days. We give each poll a "Weight", the default is 9. If the same polling firm has put out many polls in the past 7 days, the "Weight" of their older polls is less then 9 (7 for the second poll, 5 for the third, 3 for the fourth, etc) Also, polls 4 days or older are given a weight below 9 as well. This gives us a full range of opinion that is affected less by one-poll changes then a 5 poll average is. Currently we have 15 polls in our "average", and half of the "weight" comes from the 5 polls taken over the past 3 days.

The second way we differ is what we do once the geometric projection has been completed. We use that projection to find seats that are close, say within 3% or less. We then look at each of those ridings individually and judge if there is any reason the numbers might be wrong. Did a star run here last time who is not running this time? Is the opposite true? Has the incumbent resigned? so on and so forth. When combined with the numbers this gives us a powerful picture of what is going on electorally across the country.

The differences between our methodology and that of others comes out most clear in Quebec. Currently we are projecting the highest numbers for both the Tories and NDP, and lowest for the Liberals and the Bloc. This comes from our Quebec numbers. There is often doubt and people want to discount the numbers when they tell you something that seems incredible (like the NDP winning Hull) but we trust our numbers and methodology in this case. It was my using of these numbers to project 10 Tories in Quebec (and going against every prediction website in the country in doing so) that motivated me to run my own site this time. We have strong confidence in our Quebec numbers.

Numbers stable

The addition of 5 new polls and dropping of 2 old ones has a net change of only two seats on the numbers. The "Election Norm" appears to have set in, and this is where the parties will be fighting from for the next month.