- Puzzle Me This: When Does Being Right Not Mean You're Being Right

Posted: 11:37 AM on February 27, 2011
http://www.theglobeandmail.com/news/politics/ottawa-notebook/one-of-these-polls-on-harpers-lead-is-not-like-the-other/article1920604/

One of these polls on Harper’s lead is not like the other, Gloria Galloway
Globe and Mail Update, February 25, 2011 11:06AM EST

When looking at the 'full report' for the Abacus Poll there are a number of things.

http://abacusdata.ca/wp-content/uploads/2011/02/Vote-Choice-February-2011.pdf

First is his statement

"Abacus Data: Not your average pollster Abacus Data Inc. is Canada’s newest player in the public opinion and marketing research industry."

not that 'new' necessarily means 'not as good', but, practice does make perfect.

Also, he mentions that he is using weighted results:

"Results of the survey were statistically weighted by gender, age, region, language using census data from

Statistics Canada and by past vote using Elections Canada results from the 2008 General Election. Totals may not add up to 100 due to rounding."

He then invites you to go to his Website for more details - I did but couldn't find the 'more details on weighting'.

The weighting itself can, obviously, cause different results, since otherwise what's the point of weighting.

My impression is that some results are expanded and others contracted depending on demographics.

However, then you are dealing with smaller samples that intrinsically have larger margins of error.

So, I have difficulty accepting his margin of error if it is based on the gross sample. Adjusting for this could bring his results more in line with other polls.

His states it is: +/-3.1%, 19 times out of 20, which to me is what polls seem to get when the sample is 1000 or more, without weighting

He should also reconsider his weighting anyway, since pretty soon any numbers from StatsCan will be more unreliable than not weighting.

Then there is always what % the answers are undecided.

This does not affect all the Parties the same since the Con's have a die-hard core that will answer the poll and answer Con. The larger the undecided, then the smaller the decided and the larger a % these die-hard Con's make up.

Unfortunately I never seem to be able to find Ekos' undecided.

excerpt: Lloyd MacILquham cicblog.com/comments.html