Election Forecasting For 2006

[mirror]

Slinger is one of Canada's journalistic icons, and a favoutite with many Canadians.

I usually enjoy his column.

[mirror]

mirror, if you read the document on "Democratic Space" where he explains the methodology behind his projections, it's clear that his projections are worthless.

Would you care to elaborate and also suggest whose projections are worthwhile and why, in your opinion?

[kimmy]

Slinger is one of Canada's journalistic icons, and a favoutite with many Canadians.

I usually enjoy his column.

An icon? hmm. I'd never heard of him before. I looked over a couple of his columns and thought it looked like the kind of stuff you'd find in a free weekly or a college paper.

mirror, if you read the document on "Democratic Space" where he explains the methodology behind his projections, it's clear that his projections are worthless.

Would you care to elaborate and also suggest whose projections are worthwhile and why, in your opinion?

First off, the methodology ignores issues and circumstances that might be specific to a particular riding. An example being David Kilgour in Edmonton-Beaumont. According to the methodology, the Liberals won that riding last time, and are theoretically more popular this time, so they should win that riding again. In reality, however, the Liberals held that riding because of Kilgour's longstanding personal popularity in the riding, and will not have similar success without him. I am sure that the same type of argument will apply in other ridings across the country where candidates are a major factor (what will happen in Chuck Cadman's riding? How will Gurmant Grewal's shenanigans impact in his riding? What happens if Ed Broadbent retires? What happens if Sheila Caterwaul unretires? Are other popular candidates retiring? Will star candidates be appearing in certain ridings? Are there currently hot-button issues that will be of special interest in certain ridings?) As well, ridings are changed and resized between elections, so it's a big assumption to try to apply results from one election to the next when the ridings themselves are quite changeable.

Second, he tries to project regional poll results onto specific ridings. But we know that regions are not homogenious. Rural Alberta voters are different from urban Alberta voters. Lower Mainland voters are different from Vancouver Island voters and Interior BC voters. Montreal voters are different from Quebec City voters or Chicoutami voters. So applying Quebec-wide poll results to individual ridings is on shaky ground. He uses a "sensitivity factor" to try to compensate for this. This factor reflects the idea that (for instance) a poll showing an increase in Tory support in Alberta will have a proportionately larger increase increase in Rural Alberta ridings, and that a decrease for Tory support in Alberta will have a proportionately smaller decrease in Rural Alberta ridings. That seems like a fair idea, but it assumes that the changes in poll results can still be applied evenly province-wide. That's not necessarily a good assumption. Also, it is a fact that in Canada, urban ridings have disproportionately more voters than rural ridings. I suspect that random polling will therefore always overestimate urban voters and underestimate rural voters, because rural voters have disproportionate influence per vote.

And thirdly, the way his sensitivity factor is calculated is also very suspect. It appears as though it's compiled by comparing two subsequent election results, and comparing the change in each riding relative to regional popular vote. But this strikes me as highly bogus. First off, as I mentioned earlier, there are a lot of specific factors that come into play in each riding, and trying to attribute changes in results between elections to party-specific sensitivity is a big assumption. And I think a statistician would say that two elections 3 and half years apart is not a meaningful sample and could result in gross error in trying to estimate this sensitivity factor, even if one assumes that it exists.

I am not a mathematician or a statistician or a pollster or a political scientist, but I am a highschool graduate and to me it seems like there are gaping holes in this methodology. Perhaps somebody with a stronger academic background could do a better job with this.

-k

[shoop]

I am not a mathematician or a statistician or a pollster or a political scientist, but I am a highschool graduate  and to me it seems like there are gaping holes in this methodology.  Perhaps somebody with a stronger academic background could do a better job with this.

-k

<{POST_SNAPBACK}>

great analysis kimmy. Alas wasted on a moroon like mirror. He will undoubtedly ignore your coherent analysis. And he would never provide substantiation whenever someone else asks for it.

Your third point is especially valid. What sort of results would this model have predicted in 1988 given the wildly different results in 1980 and 1984?

[mirror]

Kimmy

What with the Tories at only 26% in the polls presently, it appears to me that democratic space seat projections for the Tories is probably right on the money.

I read through his material on what his formula is for deriving his seat count, and it seemed reasonable without of course as you say examining the dynamics for every single seat. But who does that?

What was it he gave them, 71 seats, and that was based on the last Ipsos Reid polling, several weeks ago now. I think Ipsos Reid had the Tories at 28% then, so the seat projections now are probably a few less than the 71 seats that democratic space forcasted, and I would say presently the Tories are in the 60-70 seat range.

Of course Tory supporters will be disappointed with these figures which is quite understandable, it's human nature, but I think they are reasonably accurate based on the latest poling results.

[mirror]

By-the-way the latest information about the Tories being only at 26%, which sound like Bush's ratings in the US these days, came from the following link.

Link

Here are two other seat projections, based on another political seat forecaster, the UBC election forecaster

DATE            LIB   CON    BQ    NDP    OTH

06/27/05      137   76       66     28       1

06/23/05      139   67       66     35       1

[shoop]

By-the-way the latest information about the Tories being only at 26%, which sound like Bush's ratings in the US these days, came from the following link.

<{POST_SNAPBACK}>

You are using a 2 1/2 month old election forecasting model to 'prove' the CPC is at 26% in the polls? What a sad, sad person you are.

Thanks for confirming your lameness by not responding to kimmy's analysis of why the democratic space projections are a total piece of crap.