GPC's Relentless Rise: Ipsos-Reid Under-estimates Green Support by 2%: Inflates Conservative's Support 5%+

There's been some discussions on polls, polling companies and trends lately -- so I think it's time to do a series of posts on these topics.

This First Approach Really Highlights the Trends
Jim Johnston has blogged about polling and he has a very interesting approach. Every quarter since the 2006 election Jim has averaged all the publicly released national polls from every polling company. He then plots the averages (see graph below).

What's great about Jim's approach is that it filters out all the noise of individual polls and the crisis du jour in the media -- and really highlights the long term trends.

Jim then takes the data from the graph above and plots the quarterly difference in support from the 2006 election and the result is stunning:

The graph above is very instructive: it shows only one party has been consistently rising since 2006 -- the Green Party!

Jim's blog In support of support at http://greenparty.ca/en/node/3203 is great. A few things are interesting to note about the graph above. In November 2006 there was a big bump in Liberal support -- why? The leadership convention was November 29-December 3 which focused media attention pretty much exclusively on the Liberals.

Which one of these isn't like the others?
Below is a reprint of a killer funny blog by the guy named Paul who blogs and runs Paulitics at http://paulitics.wordpress.com/

5 November 2007

One of these things is not like the others.
One of these things just doesn’t belong.
Can you tell which thing is not like the others
By the time I finish my song?

Ipsos-Reid just released another poll which, yet again, for the fourth time in a row, shows a dramatically divergent level of support for the Conservative Party than all of the other polls released by every other major polling firm.

The graph to the left presents the average results for the Conservative Party of Canada from five polling companies during the period of October 14 to November 1, 2007.

UPDATE: Since this post on November 5, Ipsos Reid released another poll on November 8 which put the Conservatives at a whopping 42 percent while only three days later (with telephone interviews actually overlapping with the Ipsos poll) Strategic had the Conservatives at 32 percent.

Just a note for those of you who aren’t up on your statistical theory, polls with a 1000 person sample have a margin of error (MOE) of roughly +/- 3.1%. This blog is reprinted from from: http://paulitics.wordpress.com/2007/11/05/one-of-t...

The hot topic of the day in the polling community is why Ipsos-Reid's federal numbers are so far off everyone else for the Conservatives. More later perhaps. . .

My Comments
While Ipsos-Reid overestimates the Conservatives support the company's methodology also under-estimates the Green Party support.

The final poll released by Ipsos-Reid before the Ontario election projected the Green Party of Ontario would win 6% but GPO won 8% -- so whenever I see an Ipsos-Reid poll I add 2% to the Green Party numbers.

To get a more accurate read on actual party support Paulitics averages the five most recent polls from major research firms to get a rolling average. This is something that Greg Morrow at DemocraticSpace also does. I really like this approach because it embraces the different methodologies of different polling companies, is then based on different samples, the aggreggate number of interviews is five times larger, and therefore the margin of error is lower and it smooths out variation.

Below is Paulitics' the graph of each party's rolling average (heavy coloured solid lines) with the individual polls in dotted light lines. As Paul points out, in October and November 2007 we have entered the time of "weird polls" because Ipsos-Reid's polls are so out of whack with the rest of polling firms.

Ipsos-Reid's polls pull the rolling average for the Conservatives up -- given that Ipsos-Reid's polls are usually one of the last five -- Paulists' average for the Conservatives itself will always be 1% too high (i.e. if we assume Ipsos-Reid's methodology consistenlty over-estimates Conservative support by 5% -- and Ipsos' polls are always one of five being averaged to get the rolling average, then the rolling average will always be 1% too high for the Conservatives). In the case of Ipsos-Reid's Nov 8 poll which is 10% higher for the Conservatives than other polling companies -- it lifts the Conservatives average support 2% higher in Paul's rolling average.

Another thing to consider is that the overwhelming support for the Conservatives in Alberta (typically 60%+ -- see http://paulitics.wordpress.com/polls-provincialreg... and scroll down to Alberta) distorts the national average. The Conservatives could win 90%+ of the seats in Alberta with only 40%+ of the vote in the province. So that extra support doesn't gain the party any additional seats but inflates the Conservatives support across the country. Taking Alberta out of the picture and projecting all 28 Alberta seats for the Conservatives, gives a very different picture for the rest of the country.

Percentage Change . . . . . .-3% . . . . .+2% . . . .-12% . . . . -12.4% . . .+104%

Image from http://paulitics.wordpress.com/ with last line (percentage change) added by me.

I really like how Paul summarizes the difference (+/-) from the 2006 election results at the bottom of graph in the small table.

There's an important point to make here: While the Conservative slip of 1.1% from the 2006 vote may not seem like much -- when you look at it as a percentage of the Conservative's 2006 support the shift is more significant: it's -3%. The Liberals marginal gain of 0.6% is a gain of +2% of the party's 2006 support. The NDP -- and here it gets interesting -- is down 12%, and the Bloc is down 12.4%; and the Green Party is up 104%! No other party has this magnitude of gain!.

In one of the next blogs I'll focus on the most accurate/inaccurate polling companies in Canada and the Ontario provincial election. For those who follow polls here are some helpful reference links:

Aggregators of Polling info:
http://www.nodice.ca/elections/canada/polls.php
http://www.hillwatch.com/PPRC/Polling_Station.aspx
http://democraticspace.com/
http://paulitics.wordpress.com/
http://www.angus-reid.com/polls/ -- aggregates 16,700 polls from around the world

Interesting Polling Analysis:
http://www.nikonthenumbers.com/

Polling companies:
http://www.decima.ca/en/
http://www.thestrategiccounsel.com/our_news/polls.asp
http://www.sesresearch.com/main.asp
http://www.angusreidstrategies.com/index.cfm?fusea...
http://www.ipsos-na.com/news/
http://www.ekos.ca/media/default.asp
http://erg.environics.net/media_room/
http://www.legermarketing.com/eng/ -- often conducts Quebec specific polls
http://www.mustelgroup.com/ -- conducts BC specific polls

If I am missing any or there are other resources please comment and I will add them in!