Vancouver Election Analysis – Candidate Correlation
CanadianVeggie, November 29, 2011
This will be my last election analysis post. I promise.
The Vancouver election results are particularly interesting to analyze because each voter had multiple choices to make – 1 vote for mayor, 10 council, 7 parks board, and 8 school trustees. We now know who won and how many votes each candidate got in total, but it’s not immediately obvious why. In the past week, many pundits have been musing about …
1. Why Gregor Robertson got 14,000 more votes than anyone else in Vision?
2. Why didn’t Vision’s success help COPE?
3. How did Adriane Carr win a seat?
It’s impossible to know who supported Adriane Carr or how many Vision voters didn’t vote COPE, because every ballot is secret. However, if we look at the vote percentages from the 135 polling districts, we can do a correlation analysis to try and answer some of the questions above. The high correlation between the candidates indicates that their votes were consistent across Vancouver (the same good polls and bad polls). This should be a good proxy for determining if candidates attracted support from the same voters.
Here are the scatter plots comparing Gregor Robertson’s vote totals to Raymond Louie, Ellen Woodsworth, Adriane Carr, and Elizabeth Ball.
The corresponding correlation factors are: 0.94, 0.93, 0.71, and -0.95.
Even though Woodsworth had a high correlation with the Vision vote totals, she consistently trailed the Vision candidates across the city. Why? Possibly because voters who voted for Vision and Gregor Robertson split their votes between more candidates than the NPA. Of the top 30 candidates (those getting more than 5000 vote each) 19 had a strong positive correlation with Gregor, 10 had a strong correlation with Anton (the NPA candidates), and 1 was completely random (Kelly Alm – winner of the donkey vote)
The same thing happened on parks board and school board, where voters who supported Gregor Robertson for mayor, split their votes between COPE and the Green Party, and to a lesser degree Jamie Lee Hamilton. The NPA vote largely stuck together.
So far, I’ve only compared the votes of each candidate to the mayoral candidates, but the correlation of candidates to each other is also interesting.
For Vision, Heather Deal and Andrea Reimer (0.99) and Raymond Louie and Kerry Jang (0.99) had very high correlations, while Geoff Meggs and Tony Tang (0.74) had the lowest correlation.
The NPA candidates had higher correlations with each other than the Vision candidates (most >0.98), showing the higher likelihood of NPA voters to vote the slate. The lowest correlation was between Sean Bickerton and Bill Yuen & Francis Wong (both 0.82).
Adriane Carr’s correlation is a bit more varied and interesting. Her highest correlation was with NSV candidate Elizabeth Murphy (0.88). She also had high correlations (> 0.7) with 10 other candidates (in decreasing order): Marie Kerchum, Geoff Meggs, Nicole Benson, Tim Stevenson, Andrea Reimer, Sandy Garossino, Heather Deal, Terry Martin, and Ellen Woodsworth (a mix of Vision, NSV, and COPE candidates). She had low correlations with Tony Tang (0.26) and RJ Aquino (0.34), possibly because many Vision/COPE voters choose her over the least popular Vision and COPE candidates. Her correlations with the NPA candidates were all strongly negative (the least of which being Sean Bickerton and Bill McCreery -0.41 and -0.49).
The following chart tries to visualize the correlation between the candidates. Solid lines represent correlations >0.9, dashed lines >0.8, and light dotted lines are correlations >0.6 (but only for candidates that don’t have any other strong correlations). The size of the bubbles is proportional to the votes each candidate received.
I wouldn’t draw too many conclusions from this, but it is interesting how intuitive the results look. Also, it’s shocking how different the NPA is from everyone else, and how tight their vote was. Would they have won a single seat if there weren’t so many popular candidates similar to Vision?
Link to full article and comments
