Pages

10 November 2015

Polls and Projections #nlpoli

If you want to get a sense of how accurate polls were in the last federal election take a look at the ones we have in public and compare those to the actual result on polling day.

In each of the tables below,  we are using the official vote for each party as a share of eligible vote.  Basically, that’s what the pollsters surveyed.  They asked people who were eligible to vote what they would do.

SJSMP

Eligible

MQO
Oct 15

Lib

39

39

Con

03

03

NDP

25

35

Not Vote

33

22

The MQO poll in St. John’s South  - Mount Pearl was conducted between October 9th and 14th. MQO hit the Liberal and Conservative numbers exactly. The results show decided and leaning.  That means that MQO asked the “undecided” respondents which way they were inclined to vote.

MQO showed the NDP vote 10 points higher than the final result and it had the  “no answer, will no vote, undecided” cluster at exactly 10 points higher than it actually was. That suggests that in the last week of the campaign a chunk of potential NDP vote stayed home while the Liberals got every single one of their voters to the polls.

 

Avalon

Eligible

Mainstreet
Sep 21

MQO
Oct 16

Lib

35

37/43

31

Andrews

11

13/19

22

NDP

09

16/19

10

CON

07

13/14

09

Not vote

38

17/00

28

Avalon gives us a bit of an interesting situation in that we have polls conducted about a month apart using what appears to be a couple of slightly different methods.

Mainstreet Research called a mix of cell phones and land lines in a single night in September using a computerised calling system.  They contacted more than 670 respondents.  This was before the campaign got into full swing. All responses are on the left of the column.  The decideds and leaners figure is on the right of that column.

MQO contacted 400 respondents between October 9 and 15. We don't;t know the precise method but it appears to have been human contacts by telephone. MQO numbers are only “decideds plus leaning”.

Here the responses are harder to compare easily.  If we compare the same numbers – decideds and leanings – you can see quite a variation over a month.  Mainstreet was quite a bit higher for the Liberals than the final, but MQO was short by four percentage points even less than a week before voting day. 

MQO was in spitting distance of the Liberal and Conservative numbers but their number Andrews was double what actually showed up.  Mainstreet was out to lunch on both the NDP and Conservative numbers – way too high – but Mainstreet’s decided number for Andrews was pretty much what he got, in the end.

Projections

Take the MQO numbers in each race because they are the closest in time to the final result.  They are reasonably close to the final result for the top finishers.

If you had access to the raw data you might be able to figure out if the poll results were off because of an error in weighting. In other words, it is possible MQO weighted to population instead of weighting to likely voting. We might also find that the NDP vote in SJSMP was skewed because it had more young voters in it than the others.

We don’t know any of these details because the polling firms did not release the sort of detailed poll reports you’d expect if they followed the industry association standards.  It’s a common problem in polls released in Newfoundland and Labrador.  You often run into problems with a firm telling very little about the poll and how they conducted it. 

That lack of transparency makes it very hard to assess the polls and compare one set of results reliably to another.  For media purposes, those details are largely irrelevant.  But if you were trying to do something like project the potential number of seats a political party might get,  poll results are a source of important data.

Seat projections use a mathematical formula to combine poll results with voting results over time to make a picture of what a given level of poll support might mean in seats during an election.  While the voting results are consistently reliable,  the poll results are another matter altogether.  They come from different companies, these days using different methods,  with different weighting or any of a raft of other variations.

Eric Grenier is one of the best known seat projectionists in Canada these days.  He writes a widely-read blog and turns up on CBC and in the Globe and Mail offering what is legitimately and accurately described as his expert projections. 

In a post last week, Grenier wrote about the difficulties in modelling elections based on which historic results you use. He offers the observation that his modelling works best when things don’t change much from one election to another.  Prediction is easy when things don;t actually change.

But invariably, Grenier’s job is much harder when he has to rely on the skimpy public polls for the raw data on which he bases his picture of current events.  If the polls don’t pick up shifts in public opinion, then it’s much harder for Grenier to know that he is dealing with a new situation, and not one precisely like the last one.

As good as Grenier is,  his modelling may not work effectively in a small place like Newfoundland and Labrador with skimpy information.  Look at his most recent projection.  Based on Abacus Data’s poll,  affirmed presumably by Forum Research’s similar numbers, Grenier still gives us a projection that covers a range of 25% of the available seats.

The middle of his range is a reasonable outcome of 34 Liberals, three New Democrats and three Conservatives.  But if you go through his district by district projections, you will see that some of the districts he allocates to blue or orange are actually only in that category by the very slimmest of margins. It wouldn’t take much to turn even a district like Topsail from blue, as Grenier currently has it, to red.

SRBP has used seat projections before and you can bet you’ll see more of them.  The difference between Grenier and what you will find here is a bit of extra commentary that admittedly Grenier can’t model.  it comes form experience and local district knowledge, which may not be perfect either, admittedly.

If you want to use Grenier’s current projection, take any seat that he has with less than a 60% confidence level and focus your attention there.  The closer Grenier’s confidence level gets to 50%, the more likely the seat is to change colour.

Starting from the bottom of his list up, that looks like this:

Virginia Waters – Pleasantville, currently coloured orange at 53% is more likely than not to go red.  Liberal organizational strength could come into play here on the ground in the district.  The NDP don’t have the resources to fight on the defence in more than a couple of seats.  All their weight is going to go to keeping Lorraine Michael and Gerry Rogers.  Bob Buckingham – as potentially decent an MHA as he would be  - is actually playing a red-shirted crewman on the landing party here.

Topsail – Paradise  - blue 56% – could go either way.  This is another dead heat = dead meat scenario.  If the incumbent is functionally tied with the challenger at this stage of the campaign, consider him a candidate for the unemployment line.  The fact that it’s the Premier makes it all the worse for Davis.

St. John’s Centre – orange 56% – is another dead heat = dead meat district.  The fact that such a popular incumbent appears to be in such trouble by Grenier’s model doesn’t bode well. 

Mount Pearl North – red 58% – is a battle between two candidates who’ve been mayor of Mount Pearl. They’ve also both been Liberals, although Conservative Steve Kent – a man who has toyed with every party but the Dippers – switched teams in 2007.  The old political gut puts this one as either blue or too close to call. 

Labrador West – orange 53% – is a very tight race with the incumbent already in bad shape.  This one is too close to call and while Grenier has tipped it to the NDP, it could easily go Liberal as well.  Much will depend on the organization on the ground.  The Conservatives beta the Dippers here in 2007 because the Dippers took the Thanksgiving weekend off.  That’s how tight things could be again.

Cape St. Francis – blue 56% – gives lots of votes to the NDP.  It isn’t clear how much of their performance last time was the result of having a good local candidate.  Federally the Liberals did extraordinarily well in Torbay.  If that’s a clue, then put this one in the red column.

We’ll come back to these seats as the election progresses and see if there is a need to revise the numbers.  Just so that you know,  one of the SRBP seat projection models produced results similar to Grenier’s after some allowance for floor crossings and by-elections. The other model gave the Liberals and easy sweep of 40 seats.

In other words, the reds are looking at something between 35 and 40 seats, with 40 being just as likely to be the case as not. At the low end of the range, the other five seats are split between the blue and orange three to two or two to three depending on the happenings on the ground.

We might well wind up with a single opposition seat occupied by Lorraine Michael. And frankly, in that case, we’d be looking at a by-election within one to two years there anyway.

There’s a thought for you.

-srbp-