Riding-level results are nice to have so far as they go, but serious strategists want to know how the parties did in various parts of a riding: where they grew, where they fell back, who they switched to and whose switchers they picked up, how well they pulled their supporters out to the Advanced Polls, and so forth.

Ideally we'd like to see that mapped, but there are some key interim steps, some of which are very time-consuming. Obviously an essential one is assembling a complete set of the poll-by-poll results, which is what I'm reporting on tonight.

I now have a complete set of poll-by-poll result tables in the Pundits' Guide relational database from 1997 forward (including the 2000 Transposition), which are fully cross-referenced against all the other riding result tables, parties, elections, and candidates already in it. There are 395,371 polls and 2,174,723 individual candidate poll results, and between the size of the tables themselves and their associated database indexes (what makes the pages run reasonably quickly), this new dataset is taking up just over 700 Megabytes of space in the database.

It was assembled from the raw data provided by Elections Canada for each general election plus the transposition, and hand-typed in many cases from the PDFs they supplied for by-elections. Thanks to the Chief Electoral Officer for responding favourably to my plea for raw by-election data on their website as well, which certainly made the job easier towards the end. Thanks also to my long-suffering housemate, who read those printed by-election results out loud to me while I typed them. The whole thing was a herculean task, not least because the raw data was supplied in at least 3 different formats over that period of time, and candidates were sorted alpha by family name, rather than consistently by party. The latest Elections Canada format for poll-by-poll results, however, represents a marked improvement, and I could certainly live with it going forward. I also had to cross-check and double-check my data entry to make sure everything added up properly in every direction, in the process of which I discovered a few errors with the 1997 raw data in 4 Quebec ridings, which Elections Canada is now fixing up.

A "poll" here means a "polling station". Polls in the mapping context are "polling divisions". You'd think there would be a one-to-one relationship between them, but in fact sometimes polling divisions get too large population-wise and have to be split. If they're split geographically before an election is called, you get poll numbers like "195", "195-1", "195-2", which represent different polling divisions. On the other hand, if they're not split until after the campaign has started, it's done alphabetically by family name, and the polls are called "195A" and "195B" for example. In such a case they are considered to be 2 polling stations for the same polling division (195). You can even see examples of poll numbers such as "195", "195-1A", "195-1B", "195-2". Here the original polling division was geographically split into 3 polling divisions before the election, but the second part was still found to have too many electors and thus was split alphabetically into 2 polling stations after the writ was issued.

Polling divisions can also get too small. s.538(1) of the Elections Act says that they should have a minimum of 250 electors, and so sometimes a polling division can get merged into another one, or combined with another one (a new development it appears since 2004, and I'm not sure I understand the difference between the two at the moment), or be declared "void" or "no poll held".

There are several different kinds of polling stations:

• Regular polls, pretty much what you'd expect, they are numbered sequentially starting from the number 1, and they represent the votes of people who voted on Election Day.
  
• Apartment polls, which are a subset of the regular polls, but are numbered in the 400s; also for Election Day voters.
  
• Mobile polls, which are designed for residents of long-term care facilities and the disabled, further to various provisions of the Elections Act. They are numbered in the 500s. The polling station is brought to their room, and/or remains in a central lobby area for a defined period of time on Election Day.
  
• Advance polls, which are numbered in the 600s. They include votes cast on the defined Advance voting days, but not special ballots nor people who voted in the returning office as I understand it.
  
• Special Voting Rules Group I (SVR1), which is not really either a polling division or a polling station, but is reported in a separate line. The official definition is "Includes Canadian citizens temporarily residing outside Canada, members of the Canadian Forces (except members of the Canadian Forces who voted at the polling station established for the polling division of their place of ordinary residence) and incarcerated electors", so not exactly a strategically homogenous group!
• Special Voting Rules Group II (SVR1), again neither a polling division nor station, but reported on its own line in the official voting results. The official definition is "Includes Canadian citizens residing in Canada who voted by special ballot in or outside their electoral districts"; in other words here are the special ballot voters and folks who voted in the Returning Office.
  

Neither Advance polls nor the SVR2 group has a defined count of electors, obviously. So we can calculated a candidate's share of the vote in those polls, but not their share of the electorate in that poll-type (I suppose you could express the result as a percent of the whole electorate for the riding, but I haven't done that here). Merged polls have a defined number of electors, but no results, so I had to precalculate the "merged electors" of the polls they got merged into, for database performance reasons.

You've probably also noticed the "Number of Polls" measure for each election on the riding profile page here, which is also found at the bottom of any poll-by-poll report from Elections Canada. Here's how they come up with that number:

• Take the highest regular poll number from the bottom of the printed poll-by-polls results report
• Add the number of geographically split polling divisions, eg 190-1
• Add the number of non-geographically split polling stations, eg 70A, 70B
• Add the number of 400-series apartment polls
• Add the number of 500-series Mobile polls
• Add the number of 600-series Advance polls
• Subtract the number of polls merged or combined into others
• Subtract the number of void or "no poll held" polls
• This gives you the stated # of polling stations by Elections Canada
• It doesn't include the Special Voting Rules Groups I and II for each riding

The next step of course is to assemble, model and import all the polling division boundary geospatial data, so the polling station results can be linked to their polling divisions, and the whole kit and caboodle can be mapped. Not an overnight project.

However, so that we can make use of the poll-by-poll data that's available now, I've provided a 5th tab on each riding profile page called "Poll-by-Poll". It shows the number of polls won + tied by each candidate, broken down by each type of poll (Regular, Mobile, Advance, SVR1 and SVR2), along with their share of the vote, and their percent of the electorate in the cases where that's measurable (i.e., not for Advance or SVR2, as explained above).

This allows us to see whether a certain party was more effective in pulling its support to the Advance Polls, is overrepresented in the Mobile Polls, or whether the vote shifted in a certain riding between the Advance Polls and Election Day itself.

For example, let's look at the riding of Brampton – Springdale, ON: the Conservatives won the Advance Polls by 50.3% to 37.2% over the Liberals, but by Election Day the Liberals won the Regular Polls 41.6% to 36.5%. The Conservatives won 17 Advance Polls outright (there were no ties) to the Liberals' 5, but on Election Day they won 77 of the Regular Polls outright to the Liberals' 145, plus they were tied in a further 6.

Finally, I've supplied handy links to the original poll-by-poll result reports for each electoral event at Elections Canada from 2004 forward (the earlier data is not linkable). Just click on the little Elections Canada icon next to the election of interest on any riding profile page's "Poll-by-Poll" tab to go directly to the Official Voting Results (OVR) Poll-by-Poll Results report.

OK, with that job out of the way, I'd say it's high time we got caught up on the nomination news around here, don't you.

Labels: Poll-by-Poll Results