Sunday, May 01, 2011

Vote-Splitting in Ontario:: Using Cluster Analysis & Ontario Swingdex 2.0 to Assess the Risks of Conservative Gains

I have a sense that vote-splitting in Ontario because of the NDP surge may be overstated. I decided to examine Ontario ridings past voting pattern and relative support for each party to create partisanship indices for each riding. It is a refinement of the Ontario federal riding "Swingdex" I developed last month. The idea is to classify ridings by partisan support in 2006 and 2008. You may wish to just skip to the chart.

Regional Effects
I ran some k-means cluster analyses {using a cluster algorithm in Orange [v2.0b] set to a 5 cluster solution}. First, I looked at possible regional effects. While I'm not exactly sure what the boundaries for Toronto in the EKOS polls, looking at the numbers over time, I'll assume it's the 44 ridings in the GTA. Here's the 29 April numbers [pdf], showing the Conservatives, Liberals, and NDP in a dead heat in  the GTA.

29 April 2011 EKOS poll for federal vote intent for cities

Before the no-confidence vote, the 10 March EKOS poll [pdf] had the Liberals and Conservatives polling 6.8% and 7.5% higher, respectively. The NDP has jumped 10%.

10 March 2011 EKOS poll for federal vote intent for cities

A summary of the GTA numbers are::

13 Oct. 2008::  CPC 31.1% LPC 38.0% NDP 17.0% GPC 13.4% {derived from Toronto & Suburbs numbers}
10 Mar. 2011::  CPC 41.0% LPC 38.8% NDP 13.1% GPC 6.5% ±6.3%
29 Apr. 2011::   CPC 33.5% LPC 32.0% NDP 24.4% GPC 6.5% ±5.4%

The Ontario-wide EKOS numbers were:
13 Oct. 2008::  CPC 34.0% LPC 34.0% NDP 18.0% GPC 13.0% {error not stated for ON}
10 Mar. 2011::  CPC 41.0% LPC 34.0% NDP 14.4% GPC 8.3% ±3.4%
29 Apr. 2011::   CPC 38.9% LPC 26.6% NDP 26.2% GPC 6.5% ±2.9%

The 2008 non-GTA Ontario EKOS numbers are reported here [pdf]. While the sample size does not justify this, I nevertheless decomposed the Ontario numbers to create a non-GTA Ontario numbers [x = (Ontario — (GTA*.42))/.58)], where .42 represents the proportion of Ontario ridings in the GTA and .58 represents those outside the GTA.

13 Oct. 2008::   CPC 38.0% LPC 29.0% NDP 20.0% GPC 13.0% 
10 Mar. 2011::   CPC 40.9% LPC 30.6% NDP 15.3% GPC 9.6% 
29 Apr. 2011::   CPC 42.7% LPC 22.8% NDP 27.5% GPC 6.1% 

The question is whether or not the six point Liberal drop in the GTA since 2008 and the Conservative gains plus the NDP surge will cause the Liberals to lose a large number of their 37 seats. 10.9

Quick Cluster Analysis Solution
The cluster analysis yielded a solution with 5 clusters using three simple indices based on the unweighted average of the 2006 and 2008 riding results for the Conservatives, Liberals, and NDP and subtracting that from the party's provincial results.
Indexparty = [average (riding result)2006, 2008 — average (provincial result)2006, 2008]
I named the clusters, which all 106 ridings are assigned to::

  1. Liberal {30}
  2. GTA Liberal Fortress {9}
  3. Conservatives/Split Left {22}
  4. Strong Conservatives/Weak Liberal {25}
  5. NDP {20}

All of the 37 Liberal ridings are in clusters #1, #2, and #5. Given the erosion of Liberal support and relative strength of the Conservatives, I chose to focus on how the election might cause the Liberals to lose seats. I see the Liberals or the NDP possibly picking up at most 3 seats in clusters #3 and #4.

Revised Ontario Swingdex:: Vote-Splitting Chart
The following chart summarizes the Ontario races. The first three columns are the three indices calculated for the Conservatives, Liberals, and NDP. The winning margin is the average winning margin for the 2006 and 2008 elections, where * denotes a party switch occurring. The Swingdex 2.0 column denotes a swing riding {22 total}.

Ontario Federal Swingdex 2.0-swing riding chart,
click on image for larger view

The Liberal Cluster:: Risks of Loss
The Liberal cluster of 30 ridings has a high average percentage of Liberal votes {2006 and 2008} at 47.2%. The Conservative and NDP votes being 33.9% and 12.7%, respectively. Examining the Liberal cluster {which has 5 ridings that have switched to Conservative but have a Liberal history}, I have identified 9 ridings that the Liberals should be very concerned about. The rationale are in the notes column, but the main criteria are close margins, the patterns of support in the indices, patterns of regional party support in EKOS polls, and volatility {riding changing parties}. Seven of the 9 are in the GTA, where the Liberals and Conservatives are statistically tied. There is overlap {5 ridings} with the Conservative party's targeting of GTA ridings and overlap with 4 of the original 21 Swingdex ridings. These nine represent all but one of the Ontario ridings that I see could go to the Conservatives. The tenth is Guelph, discussed in a later section.
  1. Ajax—Pickering might be at risk, but incumbent Mark Holland should hold off a challenge from a cookie-cutter "star" candidate in Chris Alexander.
  2. Brampton—Springdale, Ruby Dhalla's riding, where the years old "nannygate" might be overshadowed by Conservative opponent's "immigate" with allegations of offers of immigration visas for votes.
  3. Brampton West, Andrew Kania won in 2008 by 231 votes in the most populous riding in Canada, which also has 54% visible minorities.
  4. Don Valley West, Rob Oliphant's riding.
  5. Eglinton—Lawrence, Joe Volpe trying to hold off a concerted Conservative effort to puck up this riding.
  6. Kingston and the Islands, no incumbent, in the wake of Peter Milliken's retirement. I feel the NDP also has a shot here.
  7. London North Centre, Incumbent Glen Pearson may be technically be in the danger zone, but the Conservatives would need a sizeable swing along with Liberal defections for the riding to go blue.
  8. Mississauga South, Incumbent Paul Szabo, Chair of the House of Commons Ethics Committee
  9. York Centre, Ken Dryden being challenged by Mark Adler.
The Liberal Cluster:: Possible Liberal Pick-Ups
There are 5 ridings that have been historically Liberal that have switched Conservative, with four being in the GTA. All of these 5 were in my original Swingdex and three are targeted by the Conservatives for their bid for a majority.
  1. Kitchener—Waterloo, Liberal Andrew Telegdi lost by 17 votes and hopes turnout will turn things around.
  2. Missisauga—Erindale, a riding with another close race in 2008 determined by 397 votes. Liberal Omar Alghabra is seeking to get his seat back from Conservative Bob Dechert.
  3. Oak Ridges—Markham is yet another close race with the Liberal and Conservative candidates squaring off on a local national park proposal.
  4. Thornhill, Liberal candidate Karen Mock is seeking to unseat Conservative incumbent Peter Kent.
  5. Vaughan, Fantino or Ferri? Will Conservative riding association resignations over a health care project grant help to return the riding the Liberals from the ex-Ontario top cop, Fantino?
The GTA Liberal Fortress:: NDP Cashing In?
This cluster of 9 ridings should be safe for the Liberals (55.7% average Liberal votes in 2006 and 2008), with the Conservatives faring poorly in these ridings. The sole threat is from the NDP::
The Wildcards: Seven Selected Ridings from Three Clusters
Three of these are being targeted by the Conservatives on the quest for majority, Kitchener Centre, London West, and Guelph. The first two are held by the Conservatives and Guelph is Liberal. All seven were in the original Swingdex, so 16 of the original 21 ridings made it into Swingdex 2.0. Dropped were Essex, Brant, Huron—Bruce, St. Catherines, and Ottawa—Orléans.

Kenora, Kitchener Centre, and London west are part of a cluster that trends Conservative with the Liberals and NDP splitting the vote. In this Conservative/SplitLeft cluster, the average 2006-2008 Conservative percentage is 42.9, while the Liberal and NDP percentages are 34.6% and 15.6%, respectively. These three ridings are volatile, previously Liberal, but going Conservative in 2008.
  1. Kenora, strategic voting {SV} for the NDP could defeat the Conservatives. The riding has a history of relatively high NDP support.
  2. Kitchener Centre, rematch of a close 2008 race {1.4% margin} with Stephen Woodworth {Conservative incumbent} up against Karen Redman, Liberal.
  3. London West, this is most likely a two-way race between the Liberals and Conservatives. Vote splitting could allow the Conservatives to hold on to this seat.
Oshawa is in a cluster of strong Conservative ridings with average 2006-2008 Conservative votes of 51%. This cluster is characterized by a Strong Conservative/Weak Liberal voting pattern. Oshawa is a riding that may be vulnerable, given the NDP surge. The Schwa was Ed Broadbent's old riding, but back when GM had a much greater presence.
  • Oshawa, Colin Carrie is trying to hold on, while Chris Buckley of the NDP tries to knock him off.
These ridings are curiously in the NDP cluster while being currently held by Liberals. The cluster of 20 ridings has an average 2006-2008 NDP voting percentage of 39.9%, while Liberal and Conservative percentages are 31% and 22.3%, respectively.
  1. Beaches—East York, Matthew Kellway is taking up Marilyn Churley's cause of taking Maria Minna's seat.
  2. Guelph, a riding that had strong showings from 4 parties, Liberals, Conservatives, Greens, and NDP. Guelph was the site of a special ballot controversy, where the Conservatives tried to get student votes invalidated, over allegations that the polling place was illegal. This is a riding that vote splitting could lead to a Conservative pick-up.
  3. Parkdale—High Park, Peggy Nash {NDP} is trying to return to office by unseating Gerard Kennedy {Liberal}. I'm sure many in this riding feel that proportional representation is in order, as both the Liberals and NDP have left-progressive candidates.
Final Thoughts
Ontario will be critical with respect to how well or poorly the Conservatives do in this election. The current breakdown for the 106 seats in Ontario is Conservatives 51, Liberals 37, NDP 17 , and independent conservative 1. The theoretical maximum pickups based on the cluster analyses here for each party in Ontario are:: Conservatives 10, Liberals 7, and, NDP 11. Technically, it would be 11 for the Conservatives, given that Simcoe—Gray is held by Helena Guergis, an independent conservative. 

The latest projections expect a few net pickups for the Conservatives, upwards of 10 seats lost by the Liberals, and a handful of seats for the NDP. The projections are all in the order of Conservatives, Liberals, & NDP::

LISPOP:: 56 26 24 {1 May}
TooClosetoCall:: 55 30 21 {1 May}
EKOS:: 60 25 21 {1 May}
Canadian Election Watch:: 63 21 22 {1 May}
DemocraticSpace {avg.}:: 63 19 24 {29 April 1 May}
308:: 55 31 20 {1 May}

I expect the Conservatives to gain a few seats {net +3-4}, the Liberals lose a few {net -6}, and the NDP picking up 4-5 seats in the province. The latest EKOS poll [pdf] has the NDP surge holding at 26% in Ontario. I'm not expecting the Liberals to collapse entirely in the Liberal cluster, let alone the GTA Liberal Fortress cluster, with the NDP support allowing Conservatives to come up the middle. I think the cluster analysis is doing a good job of identifying the at-risk ridings, based on underlying patterns of support in the riding. I might be dead wrong...particularly if Liberals are so demoralized they stay home. The EKOS poll doesn't support this::

Twitterversion:: [blog] How much -splitting to expect in Ontario? Cluster analysis tries to address this question.  @Prof_K

No comments: