Since returning from Ottawa, I've developed a growing interest in politics outside of Alberta. The recent Ontario campaign was the first I'd followed with anything resembling attention but it certainly won't be the last.
I've chosen to focus on Ontario for this series of posts for one very clear reason. Ontario is a unique creature in federation, as it is the only province with near identical numbers of provincial and federal seats. This greatly simplifies the process of comparing federal and provincial results, as common boundaries eliminate the need to map and remap poll results.
The success of the Ontario Liberal campaign and the failure of their federal counterparts presents a very unique opportunity for academic analysis. As I gradually track down data, it is one I will hopefully be able to share with any readers of this site.
For students of geography, the fact that location matters is not news. Our discipline is defined by the role of spatial processes. For those in other corners of academia, this simple fact is a new and exciting revelation. Paul Krugman won a Nobel Prize for reminding Economists that distance is a defining factor in international trade. Questions of distance and space have popped up in Political Science, but have not yet been explored to the depth I would prefer. (For any curious readers: Johnson, Eagles, and Belanger all offer excellent readings on these topics).
The integration of spatial tools offers considerable potential both for academics as well as campaign managers. Understanding the spatial distribution of votes can help explain the success or failure of a given campaign. The major news networks have jumped on the bandwagon, and now include basic chloropleth maps in elections coverage. The maps, while rudimentary, provide an excellent introduction to electoral geography.
The map shown below is a simple visual representation of the winners of the 2011 Federal election in the GTA. The colour scheme is tied to that of the winning parties, and the riding boundaries are drawn from Elections Ontario shapefiles. Despite the lack of a legend, the relative simplicity of the map offers a starting point for analysis into the May 2nd campaign. The Conservative advance into the GTA can be seen as a product of spatial characteristics, in addition to socio-demographic, and other concerns.
I would contend that the 2011 election was not a collapse in Liberal GTA support, but rather a restructuring of spatial and socio demographic processes. The actual difference between Liberal and Conservative vote totals in the ridings shown here was minimimal. In the heart of Toronto, the Liberals and the Tories were neck and neck, at about 35% a side. The NDP ended up with less overal support than the LPC, but secured an additional two seats. These results are not uncommon in a First Past the Post system. Elections with single member districts can be won with less total votes, but more optimal voter distributions.
Over this series, I intend to demonstrate that not only do spatial distributions matter in elections, but that we can quantify their effects, and explain how, where and why smart campaigns can overperform in First Past The Post systems.
The Numbers:
| |
Conservative Party |
Liberal Party |
New Democrats |
| Vote Percent |
30.05% |
33.52% |
31.91% |
| Total Votes |
179,276 |
199,998 |
190,384 |
| Seats Won |
7 |
4 |
6 |
I've mapped out and clipped the results by poll for the 2011 federal campaign. (This data is not without limitations, as it does not include the advanced polls, or mobile polls) In the seventeen ridings shown above, one would expect a different distribution of seats based off the raw numbers. Based on raw vote totals in the election day polls, the distribution of of Liberal Party votes was not optimal. While I can't prove it as of yet, I am almost certain the same less than optimal vote distributions won McGuinty a third consecutive mandate.
The Evidence:
On the left is a breakdown of Liberal vote percent by poll. A first glance would indicate that the Liberal party performed admirably across the region, but did not have any clear clusters of support. Stronger liberal polls are scattered across 15 of 17 ridings, and it is only Toronto Danfort and Trinity Spadina where Liberal vote collapsed completely. The scattered nature of Liberal votes support is not unique to the GTA. Similar patterns are found both in Montreal, as well as Vancouver.
The lack of strong clusters of support in these ridings is closely tied to the failure of the Liberal Party to hold key ridings in the region. York West, Etobicoke North, and Toronto Center all show highly clustered Liberal support, and also saw Liberal victories. We would expect Eglinton Lawrence to go Liberal based on the strength in the western half of the riding, but this is as much a product of CPC strength rather than Liberal weakness. While I do not yet have data for the 2011 provincial campaign, I expect we will see similar scattered results for the Ontario PC Party, albeit without winning the total vote. The NDP vote is higly clustered in their six ridings, and speaks to the ability of a targetted campaign to overperform.
The Math:
I wouldn't be an aspiring academic if I didn't at least touch on the implications these results have for quantitative electoral analysis. The 2011 election saw massive swings, both in party vote support, but also by riding. Recent developments in spatial statistics allow us the ability to measure the spatial processes underlying electoral success. The folks at ASU have released a wonderful set of tools which you will hear referred to often on this site.
I've run the poll results through GeoDa, and while I won't go into the math underlying the results, I will say that I am pleasantly surprised. Moran's I is a measure of spatial autocorrelation calculates the relationship between an object, and surrounding objects in space. In simple terms, it can serve as a measure of clustering in spatial data.
| |
Conservative Party |
Liberal Party |
New Democrats |
| Moran's I |
0.81 |
0.6427 |
0.8579 |
The two parties with the strongest measures of clustering outperformed their respective vote totals, while the party with the lowest measure signficiantly underperformed. Space matters, and the geographic distribution of votes can mean the difference between government and opposition. In the next post, I'll break the results down by riding, and offer a glimpse at the applications for campaigning. As always, your questions and comments are welcomed.
- Justin
Posted on:
October 19, 2011
