While we were all busy covering the by-elections, the Hill Times was kind enough to publish a column of mine yesterday. It's reprinted here with kind permission.

Byers’ formula fails in his own province

Michael Byers recently urged the Liberals and NDP to strike a one-time deal where each party would stand down from running candidates in ridings in which the other party placed higher. But his vote switching theory doesn’t fly.

By ALICE FUNKE

OTTAWA—University of British Columbia professor Michael Byers has studied climate change, the war in Afghanistan, and Canada’s North. But after a provocative op-ed in last Monday’s Toronto Star, I think he should go back and study the actual numbers—at least the fields of voter behaviour and vote switching.

Byers recently urged the two main English Canadian opposition parties, the Liberals and NDP, to strike a one-time deal where each party would stand down from running candidates in ridings in which the other party placed higher, with the objective of unseating the Conservative government and installing a form of proportional representation.

The proposal has been roundly criticized by columnists and bloggers in the NDP, Liberal and even small-c Conservative ranks, who deemed it unwise, undemocratic, and/or politically unfeasible.

But would it even work? A look at switches in party voting in Byers’ own province of British Columbia, from 1988 to 2008, suggests not.

In a multi-party system, dissatisfied voters have the choice of staying with their previous party, switching to another party, or staying home. New seats can be won in one of two ways: either by increasing your own share of the electorate, or by decreasing your most serious opponents’ share, whether by causing them to switch to a third party, or just stay home.

Using riding-level data from the PunditsGuide.ca database, and calculating each party’s vote and the number of non-voters as a percentage of the number of electors in each riding, it was found that, over the seven general elections held in the last two decades, many previous party supporters would rather stay home than switch.

And the switchers there were, did not usually move where Byers assumed they would. Because the Liberal and NDP shares of the electorate both dropped in most seats in 2008, those voters either moved to the Conservatives or Greens, or stayed home. Only in Saanich-Gulf Islands is it clear that a large number of previous NDP voters switched to the Liberals. And only in Kamloops-Thompson-Cariboo, Burnaby-New Westminster and Victoria did at least some previous Liberal voters switch to the NDP, although more switched to the Conservatives in the Kamloops seat, while more stayed home in Victoria or Burnaby-New Westminster.

There are currently 36 seats in B.C. The Conservatives hold 22, of which 13 were won with more than half the ballots cast. For Byers’ strategy to work at all, the agreement would have to focus only on the seats the two opposition parties have the best chance of winning from the Conservatives. Because if the parties were to stand down in each other’s held seats, many would be at risk of falling to the Conservatives instead, undermining the whole point of the exercise.

[For the purpose of this exercise, I’m also ignoring the campaign effects of trying to run both together and against each other at the same time. But it’s safe to assume there would be a strong reaction from the Conservatives.]

The NDP finished ahead of the Liberals in 23 of the 36 B.C. ridings. It is arguably in striking distance of winning about five of them next time, but would either need to win over virtually every remaining Liberal supporter (unlikely given historical patterns), or would need former Liberals to leave the Conservatives, and then return to voting Liberal, vote NDP, or stay home.

2008 vote swings in five B.C. seats that would run only NDP candidates under the Byers plan

Percentage of Electors	NDP	Lib	Grn	Cons	NV
Kamloops – Thompson – Cariboo	+ 2.8	- 9.8	+ 2.0	+ 3.9	+ 1.0
Surrey North	- 6.6	- 3.1	+ 1.3	+ 4.9	+ 3.7
Pitt Meadows – Maple Ridge – Mission	- 2.5	- 9.0	+ 2.4	+ 5.4	+ 3.9
Nanaimo – Alberni	- 1.6	- 7.3	+ 4.1	+ 1.8	+ 4.0
Vancouver Island North	- 1.3	- 5.9	+ 1.9	+ 2.2	+ 2.6

Meanwhile, the Liberals finished ahead of the NDP in 13 of the 36 B.C. ridings. In its next five best seats currently held by the Conservatives, the NDP share of the electorate fell in 2008, but the Liberal share fell more. Three of the five were Conservative pickups from the Liberals last time, won on the basis of Liberal-Conservative switchers. To win those seats, the Liberals would find more likely supporters amongst those who switched to the Conservatives last time, while in Fleetword-Port Kells the two opposition parties are nearly evenly split.

2008 vote swings in five B.C. seats that would run only Liberal candidates under the Byers plan

Percentage of Electors	Lib	NDP	Grn	Cons	NV
Saanich – Gulf Islands	+ 8.6	- 15.4	+ 0.1	+ 3.3	+ 2.8
North Vancouver	- 4.8	- 3.0	+ 1.9	+ 2.4	+ 3.4
West Vancouver – Sunshine Coast – Sea to Sky Country	- 8.8	- 4.6	+ 4.9	+ 3.6	+ 5.0
Richmond	- 8.0	- 1.7	+ 0.8	+ 4.1	+ 4.3
Fleetwood – Port Kells	- 4.3	- 2.3	+ 2.1	+ 4.9	+ 3.9

The riding of Saanich-Gulf Islands, in which Byers now lives, is a special case inasmuch as almost 80 per cent of NDP supporters last time abandoned their party when their candidate resigned. Of the 19.4 per cent share of the electorate earned by the NDP in 2006, 8.6 per cent voted Liberal in 2008, four voted NDP, 3.3 per cent voted Conservative and 2.8 per cent stayed home. This move was still insufficient to change the outcome, and in fact it strengthened the Conservatives’ hold on that seat.

But perhaps there’s another approach.

Using the per cent of the electorate rather than per cent of the vote allows us to look at the fastest-growing “party” of the last 20 years in British Columbia: non-voters. In fact, if the non-voters were a political party, they would have won every B.C. seat in 2008 except Saanich-Gulf Islands and Surrey-White Rock-Cloverdale.

Looking at elections results this way, we notice that Conservative support has remained fairly constant in most B.C. federal ridings since 1988. It’s just that the turnout is down.

A big drop in turnout occurred in 1993 across every B.C. riding. Although the Liberals picked up five seats, in fact they typically held their share of electorate while other parties’ supporters stayed home, or split across two other parties.

A recent master’s thesis (opens PDF) from the University of Waterloo (by Maria Mavrikkou) has found that NDP support is most likely to drop when turnout drops, since their support correlates with low-income status, and that correlates with lower voter participation. And in 1993 as the NDP vote plummeted in virtually every B.C. riding, about half of the drop could be accounted for by big increases in non-voters. But in 2004, the B.C. seats showing the biggest gains for the NDP were also seats where turnout went up.

So, Byers may want to go back to the drawing board to find a more successful formula, and perhaps a different dance partner.

It might not be the one he originally had in mind.

Alice Funke is the publisher of PunditsGuide.ca

Labels: Data Analysis, Electoral Coalitions, Strategic Voting, What-if scenarios

Still battling a cold here, which demands sleep, and that cuts into precious blogging time. I've made a few database updates tonight, including the victor in the contested Conservative nomination in London North Centre last night (Susan Truppe). Thanks to a reader for bringing this clipping to our attention.

The list of official candidates for the 4 by-election ridings was released yesterday, and it appears that no-one got disqualified, while a new Independent candidate has shown up in Cumberland – Colchester – Musquodoboit Valley CCMV), Kate Graves. I've added in all the finalized candidates to the database now, and updated the graph on the main page. There are 23 candidates running across the 4 by-election ridings, 8 of them women, with both the Liberals and NDP running an all-male ticket. Hochelaga has the highest number of candidates running (8), while New Westminster – Coquitlam (NWC) has just 4 running. Each of the major parties is running a full slate. Five of the candidates won contested nominations in order to run, including 3/4 in NWC, the Liberal in CCMV and the Bloc candidate in Hochelaga, two were appointed (the 2 Liberals in Quebec), while the rest were acclaimed or were nominated as independents.

In other news, here are a few items worth following up on:

• Regular readers will recall that the Canadian Election Study (the academic exit poll that has been conducted after each general election for years) publicly releases its raw data a year after the election was held. I wrote to Professor Elisabeth Gidengil to inquire when the data would be released, and she was kind enough to reply earlier today that the data would be made available on their website shortly (probably on this page, along with many of their publications). They have made earlier datasets available in SPSS format (so, it's not for the casual user), but still and all, it is being made public, which is great.
• On a related note, former M.P. Dennis Gruending in his blog Pulpit and Politics has written the first of a two-part series examining data on voting trends by religious affiliation, looking at a recent study by the Evangelical Fellowship of Canada. In a subsequent post he plans to look at the data collected by the Canadian Election Study on that subject as well, so stay tuned.
• For those who can't get enough of mapping, there have been some very detailed coloured poll-by-poll maps of the 4 by-election ridings posted on Babble (see comment #77) by someone who looks like she or he does this professionally. Here are the links:
  • Cumberland – Colchester – Musquodoboit Valley, NS - (this one is for the 2006 results, since Bill Casey won every poll but one of the Indian Reserves in 2008, which he lost to the NDP)
    
  • Montmagny – L'Islet – Kamouraska – Rivière-du-Loup, QC
    
  • Hochelaga, QC
    
  • New Westminster – Coquitlam, BC
    
  There are also other maps linked to by that Babble user in other threads as well, including Outremont from 2006 and 2008.

Thanks to everyone for their patience, and for staying in touch. I hope to be back in fighting form for a productive weekend's work.

Labels: Data Analysis, Nov 9 2009 By-elections

Readers may recall I attended some presentations at the 2009 Canadian Political Science Association meeting in Ottawa last May, and wrote about Professor Elizabeth Gidengil's paper on the 2008 Election in English Canada.

In updating my Links page just now, I noticed that the full paper ("The Anatomy of a Liberal Defeat" in PDF) is now available on the Canadian Election Study website. It's a lengthy work which I haven't had time to fully review yet, but looks to be well worthwhile. Most notably, it answered my question from last May about whether the drop in Liberal support amongst visible minorities from 2004 to 2006 was a percentage drop or a drop in the number of percentage points (it was the latter), and so I'm updating my old post accordingly.

Professor Gidengil told me in May that the full dataset for each Canadian Election Study is made available through their website on the first anniversary of each election, and so the data for the 2008 election will be available mid-October, 2009.

It's usually in SPSS (Statistical Package for the Social Sciences) format, and so not for the faint of heart. Better to read the academic publications, and leave that kind of number-crunching to the folks with the expensive statistical software, I say.

Labels: 40th General Election, Data Analysis, Election Resources

I knew it would not take much time for the very good programmers out there to start to make some really good use of the recently-released Elections Canada polling division shape files, and Conservative blogger Stephen Taylor has demonstrated he's no slouch in the mapping wow department.

He has taken the polling division results and geographic information from the shape files and built an application for Google Earth, demo'ed in a YouTube HD video here. He's also planning to make the information available as an API (that's programming lingo for "shareable to other programmers").

Hats off to Taylor for a beautiful demo and a very solid contribution. I'll be adding some Google Maps functionality to the Pundits' Guide soon myself, but Taylor has definitely set the standard here. Bravo Stephen!

Labels: Data Analysis, Elections Canada Data

For some time I've been collecting and organizing the party financial data for inclusion in a systematic way into the Pundits' Guide database. Now, I've finally had enough uninterrupted time to assemble it into its own module. Presenting, for your interest and amusement: the first version of the "Browse Finance$" page.

Here are some things to keep in mind:

• Annual contribution data is available from 2000 to 2008 at present. Of course, the 2009 data won't be available until around this time next year.
• Quarterly contribution data is available from 2005 (when its collection was first mandated) forward, on its own tab.
• You can drill down by Party or Year, or any combination of both, on either tab. The tables and charts will redraw themselves appropriately.
• Remember that there were a number of rule changes along the way, so that prior to 2004 contributions came from Individuals, Corporations, Labour organizations, and a variety of Other changing categories as well that I've lumped in together. Post-2004 only Individuals could contribute to federal political parties.
• Most bars in the charts break down contributions by donation size as well. Technically there are 3 categories: (i) anonymous contributions of <= $25/$20 (it used to be $25, but later became $20), (ii) contributions of <= $200 where the donor is known to the party, but their names are not reported to Elections Canada, (iii) named contributors of > $200 over the year, whose names must be reported. In some older data, the donation size is not always reported alongside the donor category by Elections Canada in their datasets (the case for 2000 here). So the convention I followed was that I lumped (i) and (ii) together in the '$200 and under' category on the charts, and I lumped the 'size unknown' in with the 'over $200' category. I find it nice to always see the proportion of funds coming from these two general groupings.
• In every table, you will find little Elections Canada logos, which are links that point to the parties' annual or quarterly reports as appropriate, so you can jump directly there to consult the original data (and proofread my work, if you feel up to it). I find this makes it a lot easier to find the return I want, rather than navigating through their search engines.
• All the data was manually entered by me from returns available on the Elections Canada website, where it is only available in "as submitted" form (none of it is "as reviewed" there at all, yet).
  

As always, there's more to come on this front, but with this module and core dataset in place it should be easier to add quarterly reports in future, and to add more data to the module as well (such as the public subsidies, net transfers in and out of the federal party, and the parties' financial statements). I'll also be providing some additional metrics and analysis in the associated pundit queries in due course.

I hope you like this new page at the Pundits' Guide. Please feel free to provide feedback in the comments section or to me directly by email. And take a second to pick one or two quarterly or annual returns and proofread my data entry work, to help other readers of the Guide.

Labels: Data Analysis, Party Finance

Susan Delacourt had an interesting story in Saturday's Toronto Star (and a follow-up on her blog) on the differential rates of election of women between urban and rural ridings. The research she's citing from Louise Carbert, of Dalhousie University in Halifax, apparently looks back several decades. But I had the data handy to examine the last 5 general elections, with the following results.

[For the purposes of this analysis, I've used the same urban vs. rural riding designations as for last summer's analysis of party success done for the Chronicle Herald (with the one exception that I later realized I had accidentally miscategorized Wascana, SK as an urban riding). Notice that I've grouped the Canadian Alliance/Reform and Progressive Conservative parties together for the years 1997 and 2000 in the following charts.]

The first observation we can make is that women have been twice as likely to be nominated in urban ridings compared with rural ridings, although the number of women nominated in rural ridings is increasing overall. This was mostly due to increased number of rural Liberal women in 2008, while their number of urban women candidates stayed roughly constant. The Conservatives also had their best year ever for urban female nominees in 2008. Between 2004 and 2008, the Liberals, NDP and Greens were nominating between 2/3rds and 3/4rs of all urban women candidates, but by 2008 virtually all of the rural ones. Given that these parties traditionally have their bases of support in urban Canada, it's clear that the majority of those rural women candidates were facing an uphill battle.

Turning to the number and distribution of women who got elected, however, we can see that the number of women elected in both urban and rural seats stayed fairly constant, although the distribution between parties changed a bit. And as the article indicates, the Conservatives did manage to increase the number of women they elected in rural seats in 2008, increasing the overall number there slightly and leaving the Liberals with just one woman elected in a rural seat anywhere in the country (Judy Foote in Random – Burin – St. George's, NL). The number of rural women elected by the Bloc Québécois also declined in 2008, mainly due to retiring incumbents.

Obviously, if parties are going to meet their commitments to increase the number of women candidates they run, it will be by definition in non-incumbent ridings that will therefore be harder to win. If more women are to be elected in rural ridings, it will either be because retiring Conservative incumbents are replaced by women candidates (as happened for example when Candice Hoeppner replaced Brian Pallister in Portage – Lisgar, MB), or because women candidates from the other parties can win over those seats (although the Conservatives only lost 6 seats overall last time, 2 to women, both urban).

One picky point to finish with, though. The sidebar to the Star article refers to Agnes MacPhail as the first woman to be elected a Member of Parliament (and in a rural riding), saying that she "ran in the first federal election in which women had the vote, for the Progressive Party, a forerunner to the Progressive Conservative party." However, Ms. MacPhail herself did not join the PCs, but rather had belonged to the Ginger Group faction of the Progressive Party which went on to participate in the formation of the CCF, the precursor of the modern NDP. She ran under the banner of the "United Farmers of Ontario-Labour" party in 1935 and 1940, ultimately joining the CCF in 1942, and it was the NDP's fund for women candidates which bore her name.

Labels: Data Analysis, Rural vs Urban Seats, Women Candidates

A website that many readers might find interesting but may not know about is the International Elections forum on the "US Election Atlas" site.

Right now there's a thread on the go where various amateur and professional map-makers are taking the polling division shape files released the other week and making riding and even city-wide maps. It's great stuff, and they are having a lot of fun with it all. Go and take a look.

Thanks to commenter "the506" for bringing this work to my attention.

Labels: Data Analysis

In a very good demonstration of why I stay away from polling data altogether, readers might be interested in a series of blogposts that started as a methodological debate between the author of the Paulitics Polls blog and the vice-president of Ipsos-Reid, and have intensified from there.

Essentially, Paul was asking statistically and methodologically why, in his examination of Ipsos-Reid polls versus their competitors', a consistent gap appeared between their results for two of the political parties. Ipsos-Reid VP John Wright challenged the assertion, but did not respond to Paul's methodological points in as thorough-going a fashion as Paul had hoped for. Paul introduced more data to prove his points, while Wright amended his originally claimed source for his data, but otherwise demurred from further detailed discussion. U.S. pollster John Zogby also appears to have entered the debate in the comments section.

There are a number of good aggregations of public domain polling data for federal politics, including:

• The list of polls maintained at the Wikipedia page
• NoDice.ca federal polls page - maintained by David, a graduate student at St. Francis Xavier
  
• Paulitics Polls - maintained and written by Paul, an M.A. in political science from Carleton University
  
• PollingReport.ca - (which I've just discovered you can now follow on Twitter) maintained by Kerry, another M.A. grad Kelly, an M.Sc. grad with a B.A. in Math [thanks to the commenter for correcting me]
  
• ThreeHundredEight.com - maintained by Éric, whose academic credentials I don't know, but he publishes his methodological approach here
• John Wright earlier claimed there was a complete set of polling results at Elections Canada. Although I can't claim to know every nook and cranny of that website, I do know it better than many folks, and I've honestly never come across any such data there.
  

The points of debate between Paul and Wright seem to boil down to:

• in making claims about differences between Ipsos-Reid polls and other firms' polls, what is the source of the other polls, and which source has the most complete set of other polls
• in making claims about differences between Ipsos-Reid polls and other firms' polls, is it valid to use an unweighted average of averages
• why has Ipsos-Reid not followed the example set by other polling firms in publicly releasing polling data details
• why would Paul invest so much effort to investigate this question

The two have a history of sparring over the same topic, but the current set of blogposts does make for an interesting and real-life question about the validity of different ways to assess polling methodologies. I know I learned a lot from reading it all.

Here are the blogposts in what I think is chronological order:

• That’s some mighty interesting polling data you’ve got there
• A response to Ipsos Reid, Part I
• A response to Ipsos Reid, Part II
• A Response to Ipsos Reid, Part III
• Ipsos Reid: 0; bloggers everywhere: 1
• Bloggers everywhere: 2; Ipsos Reid: 0

Labels: Data Analysis, Opinion Research, Polling Data

As I surf Canadian political blogs for nominations information and local colour, I occasionally come across some interesting entries that guide readers might find interesting. Here are two that surfaced in the last day or so:

• From Hoser-to-Hoosier, a thorough-going look at the application of game theory to the current parliamentary impasse.
• From Not an Official Green Party Canada Site, a look at the relationship between candidate spending and vote share for Green Party candidates in Ontario ridings in the last federal election.

I like to hear about sources of data analysis in Canadian politics with strategic significance, so if you're aware of more work in this vein, drop me a line.

Labels: Data Analysis

With Tuesday's second reading vote on Bill C-280, and forthcoming votes on the estimates and opposition day motions raising the temperature on end-of-session election scares, maybe it's a good time to think about indexes of election readiness and what we can glean from the available data.

Plenty of bloggers (e.g., Far and Wide, Calgary Grit, and ThreeHundredEight.com) already regularly read the entrails of public opinion surveys, and so I would not presume to cover that well-worn territory again here. But what other boots-on-the-ground indicators can we generate from Elections Canada and other data sources?

1. Nominations progress - This is one contribution my site can definitely make, since the parties themselves don't like to release candidate lists until nominations close during the campaign period and the Elections Canada database doesn't contain appointments or other candidate selection methods that did not entail a nomination meeting. The network of Pundits' Guide readers coupled with my set of Google news alerts, web surveys and blog browsing has proven to yield a fairly reliable set of data on party nominations.
   
   I've been tracking overall numbers of nominations, counts by gender, and nominations according to how that party placed in the last election (i.e., how many candidates have been nominated in seats the party came 2nd in last time, etc., see the "Search the Database" page, 2nd query under "Nominations Progress"). Another way I could report those data is to show the number of incumbent vs. non-incumbent nominations, and I may add that to the front page, as well as in a new Nominations Progress query on the Search page. And this time I'll also be able to track them according to the date the candidate was confirmed as well.
   
   
2. Central party finance - There has been a lot of interest in the relative fundraising capacities of the various parties and the strength of their balance sheets. Thus far I've just been covering that in the website blog, but I have been collecting the raw data and getting it ready to properly model in the relational Pundits' Guide database for reporting and querying here ... I only need a nice long uninterrupted stretch of time to finish off that work. We won't have the parties' 2008 financial statements until the end of June, so best guesses now would entail looking at the 2007 balance sheets for net current assets, adding the 2008 quarterly fundraising numbers, subtracting actual 2008 campaign expenses from the election returns filed a few months ago plus some kind of estimate for 2008 expenditures, and then repeating the same for the first two quarters of 2009.
   
   I'm told that the Conservatives are closely tracking their fundraising performance to identify any effects of either the recession or partisan disaffection with the government's spending plans. The Liberals have been actively spinning their fundraising efforts; for example they released their first quarter fundraising results before their convention last month. And the NDP engaged in a bit of pre-warning spin themselves in this week's Hill Times, given that their decision to withdraw from fundraising in favour of its provincial cousins engaged in leadership contests and election campaigns is now expected to negatively impact the second quarter of 2009 as well.
   
   I've been thinking that expressing a party's most recent net worth as a percentage of the national spending limit would be a good indicator of how ready it is in this department, but taking a look at the debt-to-equity ratio might also give us a good indication of how able they are to raise financing if required.
   
   
3. Riding level finance - Registered party electoral district associations (aka "EDAs" or "riding associations") must also file annual returns with Elections Canada, and the 2008 ones were due at the end of May. I've been tracking their entry into the Elections Canada database, and so far am counting entered returns as follows 66-Liberal, 112-NDP, 77-Green, 30-Bloc, 90-Conservative.
   
   I'm going to be looking at the ridings' net worth as a percentage of their last riding spending limit as an indicator of their financial strength, and the amount of money they raise from individuals in contributions of $200 or less as an indicator of their organizational prowess. Looking at the average net worths of the ridings, they are considerably down from 2007 but that's hardly surprising when an election was held in the last six weeks of the fiscal year (see chart below). Perhaps the better comparison year is 2005 when the fiscal year-end fell in the middle of the 2006 election campaign period.
   
   
4. Candidate finance - Long-time Liberal backroomer John Mraz raised another important point in a recent column for the National Post's Full Comment blog, namely that many candidates have not yet received their riding rebates from Elections Canada as yet. Candidate returns were due in the middle of February, and I've got a pretty good dataset of them so far although that needs another pass-through to finish. For indicators, I will be looking at the campaigns' net deficit/surplus including rebates, but excluding transfers to and from the riding association, and adding that to the riding level finance indicators for that fiscal year.
   
   
5. Party Preparations - Has the central campaign booked a plane, hired an ad agency, completed its platform development work, etc. We won't really know, except based on journalists' reports such as the story in yesterday's Hill Times on the state of Conservative pre-election preparations, stories that are usually planted by the parties as part of the election scare war dance. So, while I'll anecdotally note those in the blog, there's not much to measure in that regard.
   

So, what do readers think of these indicators? Are there others you could think of that could be tracked and calculated based on publicly available data?

As a further update to the situation with Liberal incumbent nominations, the Hill Times reported on Monday that 3-4 incumbents may not have met the membership and fundraising criteria in order to be unopposed when they are nominated. However, the Liberal party did not issue any news release as to their status, and is not planning to publicly confirm whether a Liberal incumbent is entitled to run unopposed or not, leading some to wonder how they would know whether they could challenge an incumbent or not. Another Liberal quoted in the story suspects the 3-4 members who missed the deadline in fact may not be planning to run again, in which case the other speculation is beside the point.

Here is the raw data on Average EDA Net Worth by Party and Year. Note that while it's been widely assumed based on the central party financial statements that the Bloc Québécois is highly-dependent on the public subsidy funding, in fact they had quite well-endowed riding associations in 2006 and 2007.

UPDATE: One reader reported being unable to see any colour in this graphic. Is anyone else having similar difficulty? Please let me know if so, and I'll replace it with another file format.

Labels: Data Analysis, Election Readiness

On Friday morning, I attended a presentation on the first data from the 2008 Canadian Election Study, given by Professor Elisabeth Gidengil of McGill University at the Canadian Political Science Association annual meeting at Carleton University.

[UPDATE: (August 30, 2009) The full paper is now available online as a PDF.]

Echoing the title of her 2002 book "Anatomy of a Liberal Victory", Friday's talk was entitled "Anatomy of a Liberal Defeat". With her kind permission, I'm going to try and cover the main points of the paper here. For those who don't know, the Canadian Election Study collects survey data before and after every general election, for academic study. The data is made publicly available one year after the election, and thus the 2008 data (in SPSS format) will be available at the CES website on October 14, 2009.

Professor Gidengil and her co-authors use a "multi-stage analysis" to examine the data: since the factors that affect voter choice are closer in time to the vote than longer-term demographic variables, they add the variables into the analysis in stages. Of course individuals' demographic characteristics don't change, but their effects can change over time, and this was certainly one of the main findings of the 2008 election survey.

Three main factors are argued to lie behind the Liberals' unsatisfactory results in 2008 (at least outside of Quebec, as that province will be reported on in a later paper):

1. First, the data point to significant losses in support for the Liberal Party amongst two main groups in the last election:
• Visible minorities - Liberal support dropped by 14% in this demographic group between the 2000 and 2008 elections, the first wave to the NDP in 2006, and the next wave to the Conservatives in 2008. [Although I was writing quickly ... it could have been that they dropped 14 percentage points over this time period, rather than 14%. UPDATE: It dropped by 14 percentage points, according to the actual paper which is now available online (PDF).]
  
• Catholics - Liberal support dropped by 24% in this demographic group [but subject to the same reporting caveat as above; UPDATE: it actually dropped by a "massive" 24 percentage points, the actual paper clarifies] over the same time period. Once the almost exclusive domain of the Liberal party (possibly for class reasons as well), by 2006 Catholics were as likely to vote Conservative as Liberal, and by 2008 they were more likely to do so. Nevertheless Catholics are still less likely to vote Conservative than Protestants are.
• The change in Catholics has occurred more rapidly than with visible minorities, but is more striking amongst Christian fundamentalists. However, their original move away from the Liberals did not appear to be the result of same-sex marriage as an issue, but rather the very strong impact of the so-called "sponsorship scandal" (2004-->2006). Only afterwards (2006-->2008) did moral issues such as abortion and same-sex marriage emerge as important motivators for this group.
• Gidengil suggests that there may be a new voter cleavage emerging, in which the older cleavage of Protestant vs. Catholic is being replaced by Fundamentalist vs. Secular.
  
2. Second, the Liberals have across the last four elections lost their head start over their competitors in "Party ID" (i.e., answers to a question that asks something like 'thinking of federal politics, which party do you normally feel closest to').
• In 2000, the Liberals were far ahead of all others;
• By 2004, the newly-merged Conservative party was up in Party ID, as was a reinvigourated NDP; while the Liberals started to drop;
• In 2006, they were tied with the united Conservative party; and
• By 2008, the Conservatives were ahead of the Liberals on this question, although not by enough to give them a huge head start either.
• One-third of Liberal partisans in 2004 no longer considered themselves Liberals in 2008.
• Also, in 2008 one-third of the remaining Liberal sympathizers nevertheless voted Conservative.
  
3. Third, the Liberals didn't own a single election/policy issue in the last election. In fact, Professor Gidengil and her co-authors claim that "But for the economic downturn, the Liberals would have suffered a worse defeat in 2008".
• The "Green Shift" was viewed as hurting the economy. Randomly interspersing that policy description with the term "Carbon Tax" in the survey, moreover, proved that the latter was perceived as even more damaging to the economy than the former. Gidengil called it a "strategic blunder" for them to focus so much attention on the Environment, which did not appear as a priority issue amongst the voters surveyed. The policy suffered from "bad timing and was poorly explained," she maintained.
  
• The two leading issues, meanwhile, were Healthcare (owned by the NDP, along with Social Welfare issues) and the Economy (owned by the Conservatives).
• The Liberals did not place first as the best managers of ANY issue in the 2008 survey, and only placed a very distant second to the Green Party on the Environment, barely ahead of the NDP.
• On the question of Leadership, both NDP Leader Jack Layton and Conservative Leader Stephen Harper were rated ahead of Liberal Leader Stéphane Dion. It is interesting that a positive evaluation of Jack Layton did not have an effect on voting Liberal, however; "luckily for the Liberals," she adds.
  

The good news? Liberals were the overwhelming second choice of Conservative voters. However, the most popular choice for Liberals and NDPers are each other, so the Liberals should be wary, Gidengil says. The big strike against the NDP is its weak positioning on the economy, however.

Still, these findings argue that the Liberal Party needs to do more than just change its leader and dump the Green Shift to turn things around, Gidengil believes. They have to rebuild their partisan base.

Labels: 40th General Election, Census Data, Data Analysis, Ethnic Communities, Liberals

I like to give a shout-out whenever I come across political bloggers who do some interesting numerical analysis, and this week's post from the Winnipeg-based Prairie Topiary blog is worth checking out along those lines.

Following on Lawrence Martin's column of the other week, the blogger assembled data comparing the parties' national election vote-shares going back to the 1972 General Election, and plotting them against the Annual Unemployment Rate and Quarterly Real GDP Growth, in order to help answer the question about whether the left has historically done better or worse when the economy is down. There is a great chart there that you should check out, but to quote from the blogpost's conclusion:

The unemployment rate curve shows two dramatic peaks – one in 1983 and one in 1993. Interestingly, the two biggest electoral shifts over the period also occurred around those times. The first, in 1984, swept Mulroney to power with the largest majority in Canadian history. The second, in 1993, did the opposite, nearly wiping out the Progressive Conservatives and vaulting the Chrétien Liberals into power.

What's striking about the GDP growth curve is the relative flatness since about 1981 and especially during the 1990s. In the past ten years, quarter-to-quarter growth has been mostly positive, but is relatively inconsistent, with 9/11 having a noticeable negative impact in 2001. The plunge at the end of 2008 appears dramatic when compared historically.

So can we conclude any relationship between swings to the left or right and economic downturns? Not likely, at least not based on the data I've presented. Interesting shifts in the popular vote do appear, but they don't seem to be dependent on economic upturns or downturns. Economic shifts probably do affect voter behaviour, but not in ways that are entirely predictable. Furthermore, the traction that a party may get politically isn't always evident in the electoral numbers. Having an influence on the policy process and in the marketplace of ideas – regardless of which party ultimately takes credit – is something that can’t be measured here.

Labels: Data Analysis

Last month I ran some statistics examining the relationship between riding-level election spending and the associated share of the vote obtained, for the 1997-2006 general elections.

Now the bluegreenblogger has in added a few more variables and run some correlations on the dataset of Green Party federal candidates in Ontario from 2008 (reviewed here by NDP blogger the Jurist at Accidental Deliberations). Bluegreenblogger examined both vote percent and raw vote as dependent variables, and tested election spending vs. riding association (aka EDA or "electoral district association") assets at the end of the previous calendar year vs. campaign contributions vs. transfers into the riding as independent variables. I've summarized the results below:

Correlations between election finance and riding association development, Green Party federal candidates, Ontario only, 2008 election

Correlation (r)	Raw Vote	Vote Pct
EDA assets as at end of 2007	0.575	0.534
Candidate/EDA election spending	0.751	0.766
Transfers to Candidate/EDA campaign	0.698	0.727
Campaign contributions	0.329	0.277

Source: bluegreenblogger, at "Not An Official Green Party Canada Site"

I always like to report on other efforts to quantify election data, so thanks BGB for this contribution.

Labels: Data Analysis, Election Expenses, Election Finance

Friday February 13th will be doubly spooky for 1601 candidates and their official agents this year, because 12:00 midnight is the deadline for filing their 2008 Election candidate campaign financial returns.

The returns detail how much money they raised and from whom, and how much they spent on Election Expenses and Candidate Expenses, and in what spending categories. Failure to file by an M.P. can cause them to forfeit their seat, so you can be sure that party officials are following up very diligently with their caucuses to make sure everything gets filed on time.

Candidate election expense data from previous years' returns is entered in the Pundits' Guide database: it is either "as reviewed by Elections Canada" wherever available, or "as submitted by the Candidates" otherwise ("as submitted" data is shown in italics on the Financial Metrics tab of any riding profile page, and it can be internally inconsistent and a little more unreliable).

This expense data is used to calculate the "Percent Spent" (percent of the candidate expense limit for that riding spent by each candidate), the amount "Spent Per Voter" and the amount "Spent Per Vote" (i.e., per vote received by that candidate), as shown on riding profile pages (for example, here is Vancouver Centre, BC).

It's also used to calculate the number of ridings in which a party's candidates spent >50% of their limit, and the number of ridings where they spent >70%, as shown on the "Browse Parties" page.

While we wait for the new data to become available, here is a bit of a statistical overview of how parties have spent in previous campaigns:

• What percent of the limit, on average, did the parties' candidates spend in each general election, and what was the standard deviation on that data (i.e., how widely did their spending patterns vary around that average), and
• How well correlated was the Percent Spent with the candidates' Vote Percent, across parties and elections.

You may find some of the answers surprising:

• For one thing, the Bloc Québécois is now able to win seats in spite of spending a smaller percent of the candidate spending limits than they did in the past. Longstanding incumbency in many of their seats probably explains a good part of this result.
• The NDP spends, on average, a lot less than its competitors, but with a very high variability which, looking at the correlation matrix, is probably explained by the very high correlations between its level of spending and its level of support.
  
• The Liberals have gone from spending close to the limit across much of the country in 1997, to a more targetted effort by 2006; while the Conservatives (and their predecessors in the Reform/Canadian Alliance and Progressive Conservative parties) appear to have moved in the opposite direction over the same period of time.
  
• The Green Party (a number of whose activitsts have been trying to track riding spending levels against their raw vote lately in order to determine the best strategy going forward) will not derive much direction from this dataset, since only 50% 25% of the variance in their Vote Percent levels can be explained by their spending [UPDATE: Just looked up my old stats notes: it's the square of the correlation that explains the amont of the variance], and, in spite of growing expenditures, that pattern has not changed much across the last 4 general elections.
  

Average Percent of their Spending Limit Spent by Candidates, by Party and Election

Avg % Spent	Lib	NDP	Grn	BQ	Cons	CA	PC	Oth
1997 GE	73.86	28.38	0.88	78.81		46.75	47.01	2.98
2000 GE	72.06	20.07	2.63	82.15		46.66	23.66	2.42
2004 GE	73.85	25.40	3.92	57.97	60.99		5.28	5.94
2006 GE	61.45	26.26	3.02	61.31	70.37		6.15	5.62

Standard Deviation for Percent of their Spending Limit Spent by Candidates, by Party and Election

Std Dev	Lib	NDP	Grn	BQ	Cons	CA	PC	Oth
1997 GE	19.23	31.28	1.66	22.35		28.32	29.03	8.52
2000 GE	23.47	27.72	4.46	22.62		32.99	27.71	9.03
2004 GE	20.35	25.91	9.17	25.60	31.19		6.30	13.91
2006 GE	30.59	28.85	5.15	21.95	28.39		17.84	13.33

Correlation Matrix of Percent Spent to Vote Percent, by Party and Election

r	Lib	NDP	Grn	BQ	Cons	CA	PC	Oth	All
All	0.505	0.841	0.505	0.518	0.439	0.700	0.753	0.668	0.836
1997 GE	0.228	0.854	0.570	0.675		0.721	0.624	0.696	0.827
2000 GE	0.522	0.892	0.597	0.804		0.688	0.810	0.611	0.865
2004 GE	0.425	0.856	0.571	0.774	0.552		0.714	0.830	0.811
2006 GE	0.658	0.882	0.556	0.522	0.262		0.755	0.495	0.817

Labels: Data Analysis, Election Expenses

I've added some new "Nominations Progress" queries on the Search the Database page tonight, which will help readers follow the parties' preparations in several different categories of "ridings to watch".

The media love "ridings to watch", because picking a few of them helps boil down a big complicated story of 308 ridings into a few colourful local scrimmages. But what makes a riding worth watching?

• Professor Michael Geist has identified 27 ridings containing universities that were previously won by 10% or less as ones to target in his Digital Copyright campaign, for example.
• The Hill Times wrote this past February about how parties might be targetting the 49 ridings previously won by 5% of less, and called them "swing" ridings.

The trouble with these approaches is that they assume past voting behaviour in a riding can predict future voting behaviour or at least future electoral jeopardy. Demonstrably this is not the case when a government changes hands, at least not in every riding.

"Swing" ridings, in my view and contrary to what the Hill Times suggested earlier, are not necessarily "close" ridings, but rather ridings that change hands when the government changes (aka "bellwether ridings"). I don't have a dataset spanning enough elections to determine which they all are, but anecdotally the most commonly identified "swing" ridings in English Canada are Peterborough, ON and Ottawa West-Nepean, ON (both currently held by Conservatives).

I can say, though, that being a close riding in the previous election did not necessarily predict a seat changing hands party-wise in the next election, because I ran an analysis myself not long after that story ran. Only half of the ones that were close in 2006 had been close in 2004, and more ridings that wound up changing hands had been won by > 5% than by <.

This is why I prefer indicators that look at incumbency and competitiveness between parties, rather than simply "close margins". But I do love a good race too, so today I'm adding a few new queries and therefore a few more perspectives to the two already identified under Queries for Current Election(s). In order of potential volatility and therefore excitement, I would rank them as follows:

• Nominations Progress in Ridings with Retiring Incumbents
• Nominations Progress in Ridings with First-Time Incumbents
• Nominations Progress in Ridings with the Closest 4-Way Races (MargPct MargPct_1_to_4 <= 25%)
• Nominations Progress in Ridings with the Closest 3-Way Races (MargPct MargPct_1_to_3 <= 15%)
• Nominations Progress in Ridings with Close Races (MargPct <= 5%)

I've also improved on the earlier versions by adding riding counts, party candidate counts and party women candidate counts for each category.

As a note to the data, the ridings shown here in the "Nominations Progress" queries for the current election may be slightly different than the ones that appear when running the "Elections Pundit Queries" with the same names. This is because the latter looked at the electoral event in question. The former looks at the last electoral event for each riding.

For example, Outremont, QC was one of the 3 closest four-way races in 2006, yet it doesn't show up in the Nominations Progress query on 4-way races because the 2007 by-election was its last electoral event and that was not a 4-way race (however, the riding now shows up under First Time Incumbents). Similarly, Vancouver Quadra, BC now shows up as a 4-way race (it barely meets the mathematical criteria ... really it was a 2-way race for first place and a 2-way race for third place), which it wasn't at all in 2006.

------------------------------------

As your Guide was a golf widow last week, the lion's share of it could be spent fixing database performance issues, resolving some charting problems, and fixing various bugs that required some deeper concentration.

When I got bored or tired with that, I started looking up and adding web-site addresses where available for candidates in the above categories, or anywhere else that readers have sent me addresses for. As there are almost 1000 candidates identified to date, this will be a long, drawn-out process, so be patient ... and sending me any candidate web-site addresses you do have would be a big help. Or, you could invite my partner out golfing to give me more time to work.

Meantime I have a pretty long list of new functionality and new datasets I'd like to add to the database ... with one eye on the Parliamentary Calendar and one eye on the Election Fever Meter at National Newswatch (personally I think it should be a little higher than 20% though). If readers have suggestions for items they consider higher priorities than others and would like to see here soon, leave a comment, drop me a line, or stand by for a survey I'll try and compile for a separate blog post in the near future.

[UPDATE: I had to correct the criteria for the 3-way and 4-way race definitions, which were written far too late into the night.]

Labels: Data Analysis

Another of the questions frequently raised in discussions about the recent by-election results was the issue of which party was "taking votes away" from which other parties, and whether it was different in different parts of the country. Several steps are required in order to rigourously quantify the answer to that question.

As a start, I've added two new measures to the summary tables found on both the Browse Regions and Browse Elections pages, namely (i) the number of ridings in which a party's percent of the vote held or went up since the last election, and (ii) the number of ridings where it went down. The last election is determined for each riding for each election, thus the last election for Lethbridge, AB in 2006 was the 2004 general election, but for Labrador, NL it was the 2005 by-election instead.

For example, click here to see the results for just the 2006 General Election. Not too surprisingly, the Liberals went down in more ridings than they went up (278 vs. 30), while the Conservatives showed the opposite trend (down in 25, up in 283). The NDP increased its percentage of the vote in 233 seats, while the Greens showed an increase in 186, and the Bloc dropped in 73/75 Québec ridings (exactly reversing their pattern from the 2004 election, where they gained in 73/75).

That's the first step. But next, we might want to drill down to see things in, say, Montréal West Island, or the Golden Horseshoe, or Northern Ontario, or Edmonton, or Vancouver Island.

Or the entire province of Manitoba, or everywhere North of 60.

After that, we will want to see the numbers of seats in which one party went up while another party went down, and also be able to measure the amount of the average gain and loss. That's the functionality I will be coding this weekend, so users can test out their own theories about the Greens vs. the NDP and/or the Liberals, in different parts of the country, and then chart it, pour over the numbers, and see exactly which ridings were involved. Hours of fun for political junkies!

Meantime, here's a teaser ... I've compiled counts of the numbers of seats nationally in which each party went up and down together and in opposite directions as against each other party during the last General Election (shown below). Pundits may believe that things have changed since then, and no doubt they have, however during the last election, the NDP vote and Green vote moved up or down together more often that it diverged, and where it did, the NDP vote increased at the expense of the Greens more often (85 ridings) than the converse case (37 ridings).

On the other hand, the Green Party gained in 176 ridings where the Liberals lost ground, while Liberals gained at the expense of the Greens in just 21 ridings. The NDP gained ground in 224 ridings where the Liberals fell back, with only 21 ridings following the opposite pattern.

Also, note that the Bloc lost ground to the Conservatives, NDP and even the Greens ... here is a case where I really need to add an indication of the magnitude so we can quantify that, as well as drilling down to the area of interest to see local differences.



Please see the relevant entries on the Pundit Metrics page for further methodological details, and some qualifications to the interpretation of the data in earlier electoral events.

---------------------

The Guide was cited and used by Don Newman on Wednesday's CBC Politics broadcast, which my mother saw and pointed out to me. This just means I'm going to have to redouble my efforts to stay ahead of all the nominations news (stay tuned for a big update this weekend), and otherwise provide counterintuitive but interesting fodder for pundits everywhere to mull over. Thank you to everyone who has written to offer positive feedback and thoughtful suggestions. It's truly appreciated.

Labels: 39th (2006) General Election, Data Analysis

The recent by-elections have raised some issues about how to assess the performance of political parties who don't win or place second, but nevertheless may be growing or not at others' expense.

Your Guide does caution that extrapolating from by-elections having extraordinarily low turnouts to a general case is very unsafe, since there is no reason to believe that the by-election turnout is a representative sample of the turnout in a general election ... and especially not in the case of the northern Saskatchewan seat . We would need to see turnout data on a poll-by-poll basis up there, and it would have to be read by folks with a lot more local knowledge than me. Especially in that riding, one would need to consider that recent changes in voter identification laws might have had an impact, and/or that political parties, being largely viewed as inventions of white folks, wind up being far less relevant than the geographic and aboriginal background of the individual candidates in such a riding.

Nevertheless, it is helpful to have a wider array of metrics to draw on for comparison over time, and so I've started to add a few of them into the database now.

• Strong 3rd-place finishes. A count of the cases where a party's candidate placed third in a riding defined as having had a "close 3-way race" (i.e., the Percent Margin between first-place and third-place was less than or equal to 15% points). As more political parties become competitive, this pool of seats should also be closely watched as potential targets of the 3rd place party as well as the first- and second-place finishers.
• Ridings where the Party's candidate was eligible for a Rebate of election expenses. A count of the number of seats in which a party's candidate achieved a high enough percent of the vote to qualify for a public rebate of their election and candidate expenses. In the old days, we used to call this "getting your deposit back" (as in "poor candidate X, he lost so badly he didn't even get his deposit back"), although the rules have changed a lot since then. When the practice was first instituted, parties had to get 15% of the vote or more in a riding (and this is still the case in many provinces) to "get their rebate". Federally the riding rebate threshold changed to 10% with the 2004 General Election. Parties themselves also now become eligible for rebates with separate thresholds involved, but I haven't included or analyzed that data here yet.

To draw out the second metric a bit further, I've assembled an overview table here of the second metric. Remember that the threshold for receiving a rebate decreased from 15% to 10% in 2004. A couple of things stand out very clearly from the table.

• the very large increase for the NDP between 2000 and 2004 that continued to grow in 2006
• the value of the new lower threshold to parties like the Greens (compare with the Bloc Québecois for whom it seemed to make little difference in the short term)
• as of 2006, the "conservative" parties are stronger together than they were apart in 2000
• the Liberals, who could previously count on a 10% or better showing in every seat but 2 or 3, fell short of that watermark in 25 seats during the last general election.

Count of Candidates Eligible for Rebate of Expenses by Election and Party

Elec	%	Tot	Lib	NDP	Grn	BQ	Cons/Ref	PC	Rest
2006	10%	308	283 (244)	214 (145)	7 (0)	73 (70)	303 (291)	( )	4 (3)
2004	10%	308	307 (298)	195 (134)	3 (1)	74 (72)	251 (227)	( )	6 (5)
2000	15%	301	(298) 288	(86) 57	( )	(73) 69	(208) 182	(143) 87	(3) 2
1997	15%	301	(299) 292	(121) 87	( )	(73) 70	(192) 161	(233) 189	(3) 2

This analysis has got me interested in categorizing party performance by percentage of the vote in deciles or something like that, and seeing how it changes over time, but that's a project for another day.

For now, the metrics "Strong 3" and "Rebate" appear on both the Browse Regions and Browse Elections summary tables, and I've freed up space in those tables by pulling out some summary data into stats tables above them (slightly different versions of the stats table appear on each of the two pages).

Next up: the number of seats in which a party increased its percentage of the vote, vs. the number of seats in which its vote declined, and I'll let users compare two parties against each other on those measures, in order to assess the contest between them.

Labels: 2_How to Search the Database, Data Analysis, Election Finance

This morning's Hill Times writes about how the two main parties are targetting the ridings won by less than 5% of the vote last time (i.e., MargPct < 5%). This criterion also appears as one of "Election Pundit Queries found on the Browse Elections page in this database, allowing you to see quickly which ridings those are for all past elections, called "Closest contests (MargPct <= 0.05)".

At least one writer tonight, however, is questioning whether a close margin should be the only criteria for targetting ridings likely to change hands again.

Naturally, it makes sense to consider the 2006 results as a relevant factor in allocating resources. But any party which relies unduly on 2006 alone as its baseline figures to miss a significant portion of both the risks and the opportunities facing it next time out. -- Accidental Deliberations

So, I decided to do an analysis that compared seats previously won by margins of < 5% against Ridings that changed hands party-wise in an election (another Election Pundit Query), to see how well a narrow margin predicted later riding turnovers. Here are the results:

Distribution of seats by Hold vs. Change and Previous Percent Margin of Victory, 1997-2006

	Under 5% Hold	Under 5% Chng	Over 5% Chng	All Under 5%	All Chg
2004 -> 2006	33	25	30	58	55
2000 -> 2004	11	28 (24)*	39	39	63
1997 -> 2000	21	15 (11)*	22	36	55 33

* reflects adjustments between the 2000 General Election results on the 1996 Redistribution boundaries and the 2000 Transposition results on the 2003 Redistribution boundaries.

The statisticians out there can run the chi-squares on this, but my Coles Notes version is that, in each of the three cases here, more seats changed hands where the previous margin of victory had been over 5% than did under it.

[UPDATE: Fixed transposition error in the table.]

Labels: Close Margins, Data Analysis, Seats that changed hands