Once again today there was only a single new poll released today to be incorporated into our poll tracker and that was the Nanos poll that showed the Liberals up to 32.7% and the NDP down to 15.9% with the Conservatives steady at 39.1%. That boosted both the Tories and the Liberals in my poll tracker, but since the poll doesn't take a straight average the Liberal gains were moderated.

Not to be a party pooper, or let my personal bias interfere with my analysis, I do have some concerns about the validity of the Nanos poll. The increase in the Liberal support is indeed startling, more so because it is a three day polling average, which means the difference between the Liberal support on Sunday and on Wednesday must be at least 12% in order to account for an increase of 4% over a single day. Similarly, the NDP support must have dropped by approximately 12% over the same time period to explain their sudden drop in support.

Because of the low NDP support initially, this means that at least 1 in 2 NDP supporters on Sunday decided to switch their votes and go with the Liberals instead by Wednesday, dropping the NDP support to around 10%. Does that sound realistic to you?

Mathematically the problem is very interesting since there is no way I can see to determine for certain what the actual daily values that were used to calculate the averages; there are 8 variables and only 4 equations available.

We know that the average of A, B and C must be 28.7, the average of X, Y and Z must be 19.6 while the averages of B,C and D and Y, Z and W must be 32.7 and 15.9 respectively. By arbitrarily choosing a few values for A, B, X and Y we can figure out from there what the remaining values must be.

Choosing initial Liberal support of 27 and 28 would mean that the third and fourth days would have Liberal support of 31 and 39 respectively, a shockingly high number for the fourth day. If the NDP support was 20 and 20 for the first two days, then it must be 19 and 9 on the third and fourth days, a shockingly low number. I tried another more reasonable possibility that wound up with Liberal support at 35% and NDP support at 12% but it still requires an unbelievable drop in NDP support.

Try it out for yourself, see if you can find some more reasonable numbers than I did.

## 6 comments:

Good stuff ! I tried to do similar calculations last election, and ended up, like you, having to guess at the daily percentages based on Nik's samples of 400.

Three polls from now, the big number for the Libs will drop off. So let's say they get two days of level numbers, followed by a drop in the third day. That will combine to produce a pretty big drop in the 3-day average--a big number gets replaced by a small number.

So like I've said before, there is going to lots of daily volatility in Nik's numbers.

Day 5 polling from the 2008 election was quite similar. As the campaign progressed the gap tightened. Then voting day came and we saw the real numbers.

Day 5 Nanos poll 2008 election (CPAC-Nanos nightly tracking completed last night (September 12th) shows the Conservatives eight points ahead of the Liberals among decided voters (CP 38%, Lib 30%, NDP 15%, BQ 9% GP 8%). The Tories and Liberals continue to be statistically tied in Ontario (CP 38%, Lib 37%) with the Tories narrowing the gap in the Atlantic provinces to 7 points (Lib 40%, CP 33%).

Calgary Junkie:

There's no mathematical way to solve the problem as I can see it. Adding another day only creates the problem of having 10 variables and 6 equations.

Paulsstuff:

I'll agree, last election Nanos wasn't very reliable but the day to day variability is part of the nature of the beast. A 400 sample poll has an accuracy of only 100%/sqrt(400)= 5%, so there can be a large degree of variability day to day.

My analysis indicates somewhere between a stronger Conservative minority and a strong Conservative majority at this point. Conservatives ending up somewhere between 152 and 172 seats. It's a question of the efficiency of the vote, which is something the Conservatives have right now - in spades.

In a few days, we'll know if there's a shift happening from NDP to the Liberals. That's all fine and good, and it may mean a few seats changing hands here and there, but the important information is the regional voting patterns, and it looks -- at least at this point -- like the Conservative support is rock solid and is strong in the places that really matter.

scbritton: I'm not really confident that the polls are good enough to base any real seat projections on. In any case, I prefer the seat-by-seat analysis done by people like electionprediction.org

I tend to believe myself that the Tories will end up with between 150 and 155 seats. I really hope its the latter since it will give us some respite from the silliness that comes from minority government periods and constant campaign mode.

It is too early to predict an outcome on turnout given the hype on this poll. I like to use all of them to look for consistency in their snapshot.

http://www.wlu.ca/lispop/fedblog2011/?page_id=10

The following seat projection is based upon a weighted blending of polls from Nanos, Harris-Decima, Abacus and Forum Research that were drawn from March 24-29. The overall sample was approximately 6000.

http://www.wlu.ca/lispop/seatprojections.html

## Post a Comment