Friday, February 12, 2010

EKOS Gone Awry

I often reference my "election matrix", which is a spreadsheet of the 2008 election results riding by riding, region by region, province by province. When I receive new polling data, I redistribute the 2008 results (which are the most accurate possible numbers available on voter preference) to reflect new polling data (it does include several algorithms and functions, and is much more complicated than adding and subtracting, hence the computations of first past the post projections). I use the 2008 federal election results as my base of reference, unlike Frank Graves, who likes to compare current outliers to past outliers. "The Tories are down to a projected 114 seats from their October projection of 180 seats" (my own projections had the Tories peaking near 165 seats in October).

I decided to plug all the EKOS numbers into my matrix to double check their seat projections. They break down all their numbers regionally, and when I started measuring alleged shifts in voter preference, the numbers were insane. I got through Ontario, BC, and Quebec, and was left with no other conclusion than the EKOS numbers are virtually impossible. The Tories are down 14% in BC, but only down 8% in Vancouver where half the provincial votes are? The NDP is up in Montreal where nearly half of the Quebec votes are located, but down 140,000 votes everywhere else? It doesn't compute.

The "other" parties (not Tory, Liberal, NDP, Green, or Bloc) are up by a quarter of a million votes nationally. The Marxist Leninist Party must be juiced! The Greens managed 3% of the Quebec vote in the last election, but have since added 167,069 votes? The Tories are down 200,000 votes in Quebec, but most of those are in the Montreal area where they have no seats. Last April when EKOS had the Liberals creeping into majority territory, I computed the national numbers (which proved to be yet another outlier) and reported it. But this time breaking it down regionally, the numbers are insane!

I suppose that we should put this outlier in perspective in terms of the EKOS numbers. They report that they polled roughly 3000 people nationally. Based on the proportion of total number of counted votes in 2008 by province, here is an estimate of how many people were polled in each region.

Ontario 1113
Quebec 791
BC 391
Alberta 277
Atlantic Canada 234
Sask/Man 193

Which equates to one person polled for each 10,000 Canadians. 391 BC residents on the EKOS call list indicate that the Greens are up to 15%, and the Tories are down by 14%.  That means that only 55 people polled were required to say Green instead of Conservative for the Tories to fall by 14% provincially.  If you know who in your database is likely to answer what, it does not take much effort to willingly skew your polls. It is all about perspective.

Argh, I'm giving Frank Graves more attention than he deserves. Granted, the only reason that I am giving him this much attention is because his friends at the Globe and CBC are parading this PR opportunist as the Oracle of Canada. This guy is the busiest pollster in the country. He seems to be always releasing a new poll, but why ask why? Is the CBC paying him for "exclusive" polls? Or is he just telling them what they want to hear so he can get on television and pump his company?

You decide...

1 comment:

  1. I'd still trust Nanos over EKOS. Oddly enough, the latest Nanos poll doesn't show any of the EKOS "action". Probably why it's not "newsworthy".

    ReplyDelete