Monday, February 28, 2011

The EKOS Outlier

On February 23rd Liberal pollster Frank Graves released a poll that the Tories had fallen dramatically to 32.4% support in the week following the public release of Odagate. This allowed Evan Soloman to muse about "the Oda effect". On Feb 21st, Nanos had the Tories at 39.7% and two days later EKOS reported a drop. Then on Feb 25th Abacus and Angus Reid had the Conservatives at 38% and 39% respectively. So was what Frank measured real or wrong? Did the party just lose 7% support for a few days and bounce immediately back, or was it an outlier that just happened to cull a non random sample?

The ironic thing is that the PMO sent out an internal memo warning caucus not to get too excited about an earlier EKOS poll because some pollsters report high numbers only to follow them with large declines that they and the media get excited about. This makes their poll a bigger story. The bigger the swing, the bigger the story. I am not accusing Graves of deliberately manipulating his own numbers, but rather indicating a likelihood that they have a flawed method of randomization. Randomization is a key necessity for opinion sampling and estimating the true parameter values.


  1. The look on Graves face when he gave the result was odd...he looked guilty of something.
    Speaking of the look on faces, check this out from P&P tonight.
    First segment near the end of the segment...Ralph Goodale finally getting the boots put to him by Rick Dkystra.

  2. You are being too generous to a known Liberal hack.

  3. I wish that we had a body language expert to do a take on those ranters on the political shows, or the ranters i QP. Bill O, has one one now and again and it is very interesting to watch.
    Those ranters can't disguise their hate or their lies, their eyes, body language, and facial expressions give them away.
    Mary T.