27 November 2015

Weight Problems #nlpoli

If you want to know why Forum Research’s poll is out of line with the other polls done on the provincial election, you need look no further than the data tables for the questions.

This is why pollsters should give out this information. Lots don’t.

Polling firms adjust their sample so that the sample matches the population as a who for sex, age, geography, and so on.  It’s called weighting.

Forum notes that where “appropriate,  the data has been statistically weighted by age, region, and other variables to ensure that the sample reflects the actual population according to the latest Census data.”

That’s where you get the problem.

On page four of the detailed poll report, you can see the sample of 790 people eligible to vote breaks down to include 23 people in Labrador, 107 in western Newfoundland, 110 in central Newfoundland, and 550 on the Avalon.

The actual population is roughly 50% on the Avalon and 50% off the Avalon.

Forum’s sample is weighted 70% on the Avalon and only 30% for the rest of the province.

That’s enough of a problem to throw the entire poll result off.  There’s no way to correct or adjust the poll without having the complete table of all the individual results so basically

There’s your explanation of why a raft of polls all gave roughly the same number but Forum’s second poll showed a dramatically different result. it was also dramatically different from Forum’s first poll. 

And in that first Forum poll, the company didn’t showed detailed regional breakdowns.  Odds are that the first poll was correctly weighted since its results are close to the other polls.. 

Quick double check.  Take a look at Abacus’ regional breakdown in their second poll and compare that to what Forum calls its provincial numbers.

 

Abacus 2 (Avalon)

Forum (“Provincial”)

Liberal

57

52

Conservative

25

29

NDP

18

19

D’oh.

But wait.

It gets better.

Over at 308 dot com,  seat projectionist Eric Grenier couldn’t wait to revise his seat projections for the lection base don Forum’s numbers.  Even as Abacus released a second poll, Grenier stuck with his projection that the Liberals would now likely win 30 seats with the Tories and Dippers splitting the other 10.

Grenier gave the latest Abacus poll half the weight of the Forum results or less. Grenier weights Forum very heavily in his projections, as you can see from this little table on his blog (retrieved.26 Nov 15).grenier weighting

Here’s an excellent example of why seat projectionism is more of an art than a science.  Grenier’s latest projection displays perfectly the extent to which projections are dependent on the input data.

Or, as the experts would put:  GIGO.

Garbage in.

Garbage out.

Grenier missed the point, apparently, that Forum’s poll has a fatal flaw.  A single night IVR is useless not because it’s a single night IVR but because the poll itself is frigged in the same way some polls in elections a couple of year’s were wrong:  they got the weighting wrong.

Go through Grenier’s individual seat projections and you can see the extent to which is model is seriously frigged up.  Monday will confirm this contention.  Grenier makes no mention of the weighting cock-up, incidentally.

How badly effed up is Grenier?

Well, look not further than Waterford Valley, where incumbent Liberal Tom Osborne is up against a placeholder Tory who got into the race late and who has campaigned very little.  She’s an associate of Tina Oliviero, by the by.

Anyway,  Grenier now projects Osborne will lose by seven points.

Huh?

Grenier’s model considers this a Tory seat because of the trending that had chunks of it going Tory before.  Problem:  Osborne used to be the Tory who held a chunk of the district. Grenier’s model puts too much emphasis on the party choice in previous elections.  Plus and discounts entirely the candidate.  Put some garbage data in the front end, in this case an incorrectly weighted poll, and Grenier’s model spits out nonsense.

There is absolutely nothing that would independently confirm Grenier’s projection in Waterford Valley.  Nothing.  No trends.  No events. Nothing that you could point to  - intuitively  - that would say Waterford is going to suddenly reverse direction 180 degrees based on a single poll.

Yet, there you’ve got Grenier predicting a Tory miracle and NDP and Tory miracles elsewhere.

Nonsense.

As you will see on Monday night.

-srbp-