The Vancouver mayor’s race, similar to last year, will be equivalently interesting this year. Last election (2005), on the NPA side, Christy Clark lost the nomination to Sam Sullivan, and during the general election Sam Sullivan prevailed over (COPE candidate) Jim Green by a very slim margin.
This year, Sullivan’s nomination is being challenged by Peter Ladner, another city councilor. Both candidates have been signing up people for the upcoming nomination and have been releasing election propaganda showing why they are the better candidate.
I will look at one survey released by the Peter Ladner camp – A poll claiming that he is the favoured candidate by a margin of nearly 2:1. Doing some stats work, I was curious to see how credible this claim was. From a 500 person survey (assuming randomized distribution), 102 of them identified themselves as NPA supporters.
Out of those 102 NPA supporters, 58 said they would support Ladner for the nomination, 30 would support Sullivan and 14 were undecided.
Is this statistically credible? For a confidence interval of 95%, your error bars would be around 9.7% using a normally distributed sample based upon a 50% expectation. For a 60% expectation the error bars go down to 9.5%. Assuming the worst-case error margin, the 9.7%, we have the following:
Ladner 57% [95% confidence range: 47-67%]
Sullivan 29% [95% confidence range: 29-39%]
So the error bars aren’t overlapping. If you apply a 99% confidence interval, you raise the bars to 12.7% in each direction:
Ladner 57% [99% confidence range: 44-70%]
Sullivan 29% [99% confidence range: 16-42%]
So Ladner is looking pretty good, assuming random statistical methodology. Of course, this survey is likely not to be random – for example, the sampling is likely biased toward people that actually answer questions truthfully, or even know who either candidate are, etc. Also, the demographic impact of the inability to call mobile phone lines for sampling can be corrected for if you know those inputs.
Finally, the math is wrong – there are three inputs, mainly the “unknown” vote. If you remove those 14 undecided voters and scale the rest of the rest of the 88 votes, you’ll have overlapping error boundaries on the 99% confidence interval. Ladner should feel OK about this survey, but he can hardly feel like he clinched the nomination without doing a more thorough survey.
Fortunately for Ladner, Sullivan’s survey of 300 doesn’t indicate too much other than that half the voters are “aware” of Sullivan, and the biggest piece of statistical quackery I have seen is on page 9, where 30% of decided voters support Sullivan vs. 24% for Ladner. Of course what they don’t mention is that this is 51 vs. 36 of a sample space of at most 160 people. This is the definition of “statistically insignificant”.
On the Vision Vancouver side, in my books Gregor Robertson has it wrapped up – this survey provides some statistically insignificant evidence that this is the case (out of the 80 ID’ed Vision Vancouver voters, 41 of them said they would support Robertson). Unless if COPE nominates a mayoral candidate, it would take an incompetent campaign run by Vision Vancouver to lose the upcoming election. The NPA’s best (and likely only) chance is to get COPE to run a candidate for mayor.
The easy conclusion is that it’s likely in the best interests of Vision Vancouver and COPE to cut a deal where each slate would run a certain number of people for council in exchange for COPE not running a candidate for mayor.