Monday, September 22, 2014

Modeling the election result

As I detailed earlier, I have built a seat model for New Brunswick.  You input polling data and based on the results of the last 5 elections and other factors, it predicts the results in New Brunswick's new 49 ridings.

On the eve of the election, there are two fresh polls to consider.  A poll from local firm Corporate Research Associates taken Sept. 15-18 and a poll by ubiquitous robo-poller Forum Research taken on Sept. 21.

There are pros and cons to both of these polls.

CRA has an excellent track record in New Brunswick and came within two points of calling the 2010 election.  It also has decades of experience polling New Brunswick and there is as much art as science in finding a truly representative sample, especially in a relatively small province with distinct regional variations.  The disadvantage is that the data is older and developments like the CTV Leaders' Debate were missed.

Forum has the advantage of recentness and a huge sample size.  In fact, I do not think I have ever seen an opinion poll done in New Brunswick with a simple this large.  However, Forum's methodology is one which has incredibly low response rates.  And because it was done just on one day, they may have missed a variety of voter types.  Forum's track record is spotty, particularly when it polls smaller jurisdictions where getting a random sample is trickier.

CRA says it is an almost certain Liberal win with Liberals at 45, PCs at 36, NDP at 11 and Greens at 6.  Forum agrees for the smaller parties, but has the PCs and Liberals tied at 40 points each.

Forum has their own seat model which says a 40-40 tie would yield a 26-23 PC majority.  My model agrees with that assessment.

CRA's poll paints a completely different picture: a two-to-one Liberal majority of 33-16.

Who's right?  We'll know in about 24 hours.

There are a few ways to crunch these numbers, and I'll be curious to see how Eric Grenier deals with it over at ThreeHundredEight.com.

In my gut, I trust CRA's knowledge of the region and track record above that of Forum.  So I would give CRA double the relative weight of Forum.  However, because Forum polled more than 4 times as many people, they would still get the advantage.

If you combine the polls and weight Forum's sample at 50%, you get a combined sample of 1014: 333 from CRA and 681 from Forum.  That gives a 43-37 Liberal win in the popular vote and a 30-19 win in the seat count.

That will be the official nbpolitico projection for the election.  My gut prediction was posted earlier and is 38-8-2-1.  It assumes a few strong local campaigns for the third parties that pollsters will have missed, and that New Brunswickers will have been influenced by the nine point lead in the CRA poll and given the Liberals a late boost through bandwagon effect.  It is hard to say what effect the Forum poll will have on voter perceptions; will it being released at midnight tonight it probably won't make the papers and many media outlets have a policy of not reporting polls on election day.

No comments: