May 11 2013: Technical adjustments when a party isn't running a full slate of candidates

Two of the four main parties in BC aren't running candidates in the 85 ridings. In fact, the Green party only has 61 candidates while the BC Conservatives have 56 (or 60 if you include the 4 candidates who are on the ballot without the Conservatives identification, due to an error in registration with BC Election). For projections, this creates essentially two problems.

1) Polls (usually) don't account for that and therefore overestimate the support for these parties.

Except for last night's Angus-Reid poll who made full use of its online panel to offer respondents only the options available in their riding (i.e: if you live in a riding without a Green candidate, AR would not have offered this option to you), most (if not all) polls simply ask the same generic question to everyone. So there could be a small but significant numbers of respondents who declare wanting to vote for the Green party for instance, even though there isn't a candidate in this riding. After all, it's quite reasonable to assume that most respondents don't actually know if there is a candidate for each party in the riding and will only find out on election day.
 When that's the case, the support for such a party is clearly overestimated. The effect could be particularly important with the Green and Conservatives parties since they aren't even close to having 85 candidates.

On the other hand, if the polling firm accounts for that (as did Angus-Reid), then the support for the Green party is unbiased. And if you look at the AR poll, you indeed find the Green and BC Cons below their usual levels in other polls. But in this case, it actually underestimates the actual level of support in ridings where there is indeed a Green candidate. This post from BC Iconoclast explains it clearly.

In average though, since most polls don't account for this, expect the Green and Conservative supports to be overstimated. I don't currently correct for this in my projections but will make sure to do so for the final ones.

2) The provincial swing is affected.

My model, as most of these type of models, use the provincial swing to make projections in every riding. While I do account for regional and other effects, I have to make additional adjustments for the Green and BC Cons. Here is why: the Green party got 8.2% of the votes in 2009, with 85 candidates. There are now 24 ridings where there isn't a candidate anymore. In these ridings, the average level of support received by the Green candidates was 7%. So you see that the BC Green party is "efficient" and has decided to run candidates in ridings where it has a better chance this time around (specifically, in ridings where there is a Green candidate in 2013, the average result for this party in 2009 was 8.4%). Still, these 24 ridings with an average of 7% account for a good part of the overall 8.2% level of support. In fact, without these ridings, the Green would only have received around 6% of the votes province-wide. So when the polls (assuming we corrected for the effect explained in part 1) show the Green party at (say) 10%, it means that in the ridings where there is indeed a candidate, the provincial swing isn't 2 points (10-8), but 4 points (10-6). If you don,t account or correct for this, then you clerly underestimate the Green candidates in your projections.

To take a similar example as BC Iconoclast, I currently have the Green party at 18% on Vancouver Island. So quite similar to most polls (which shows that my regional coefficients to transpose the provincial swing actually work). But this 18% is an average out of the 14 ridings of the island. In reality, the BC Green party only has candidates in 11 ridings. So if I average over these 11 ridings only, I find the Green standing at 23%.

In conclusion, remember that most polls will have the Green and Conservatives supports wrong. But count on me for making sure to account for that in my projections (as well as in the simulator that you can use on this site)

Share this

Related Posts

Previous
Next Post »