I ran my New Brunswick projection model through a test, using the 2006 election as a base. The results were not satisfactory, so I made some tweaks to give more weight to recent polls and less weight to old polls and elections. The result was much better, but the model is only as good as the polls that are put into it.
One of the problems is that in the last election almost every pollster was wrong. In all, only six polls were taken through the 2006 election campaign. Only one of the polls had the Progressive Conservatives over 46%, and the party ended up getting 47.5%. The six polls averaged 45.3% for the Tories, and the last three polls (taken between September 7 and 13) had an average of 43.7%. For the Liberals, the six-poll average was 43.7%, the last three poll average being 45%. That is well below the 47.1% the Liberals earned on election day. And then the NDP averaged 8.5% during the campaign and 9.3% in the last three polls. The party ended up getting 5.1%.
Is that a ballot box factor? Is it bad polling? Or is it that there was too much of a gap between the last poll and election day? It is impossible to know, and there isn't enough data to go on.
The last poll was taken between September 10 and 13, the election taking place on September 18. That means that the last poll incorporated polling data that was a week old, and no numbers were recorded during the last five days of the campaign. This last poll by Omnifacts Bristol ended up being not too far off, with 46% each for the Liberals and Progressive Conservatives and 7% for the NDP. That would have given 29 for the Liberals, 26 for the Progressive Conservatives, and no seats for the NDP in my projection - which is exactly what happened on election day.
That poll was correct within the margin of error, but was it just lucky? It still under-estimated the vote for the Liberals and Progressive Conservatives, and over-estimated that of the NDP. This election is going to be a very close one, and it will come down to individual votes and ridings. For that reason my model is not ideal. The election could come down to 27 seats per party, plus or minus one or two seats. That would be too precise for what my model can handle, and if we only get a half-dozen polls or less that makes it even more of a guessing game. But I will do my best with it.
In any case, with the tweaks I have made the projection has changed somewhat dramatically. I now have the Liberals forming a majority government with 30 seats and 42.6% of the vote. The Progressive Conservatives come up short with 23 seats and 40.1% of the vote, while the New Democrats elect two MLAs with 13.2% of the vote.
Shawn Graham would be returned as Premier but with a reduced caucus. David Alward picks up two seats while the NDP makes a historic breakthrough.
On the face of it, this sounds more "right" to me. I can see this happening easier than I could a Progressive Conservative minority, which is what I had yesterday. But we need some more polls!