Tuesday, March 24, 2009

So the Ipsos Reid Polling Numbers are in

The numbers indicate that there is a large gap between the NDP and the Liberals, large enough to signal a landslide for the Liberals if they hold through to the election. The Ipsos Reid numbers have been consistent for three years now.

  • Liberals - 46%
  • NDP - 35%
  • Greens - 15%
  • Others - 4%
I do not have access to the full report and therefore I can not see some of the more relevant data, specifically how many people will not vote. Typically in most polls there is a higher percentage of the population that indicates a party preference than the people going to the polls. In an 800 person poll, about 250 to 300 of the respondents will not be voting, unless they do not use these responses. Normally what I see in polls is people listed as do not know and not will not vote. The do not know numbers tend to be a much smaller percentage of the actual non-voters. The only clue we get in the polls of voter intention is the numbers that relate to how firm people are in their choice - Liberals 70% firm, the NDP 60% firm and the Greens 50%. If we use these numbers we end up with:

  • Liberals - 32.2%
  • NDP - 21.0%
  • Greens - 7.5%
  • Others - 1.0% (my own estimate based on past performance)
  • Floating voters - 38.7%

I think this gives us a better indication of what the data is saying. The NDP has only secured their core supporters at this time. The Liberals have a commanding lead. Most of the floating voters are not likely to vote.

Looking back on late 2004 and early 2005, the NDP had a higher committed vote than this time and the Liberals lower - the Greens were in the same shallow support as now.

On the basis of the Ipsos Reid numbers, I am broadly sticking with my existing projections for the election, but I will move Vancouver Kensington into a likely Liberal win rather than a toss up as I had been thinking it would be.

I know people are looking for regional break downs, I am not very interested because the data will be too erratic to indicate much at all. At 800 people the margin of error is +- 3.46% at a 95% confidence level. I am actually generally disappointed how few people that look at polls understand the math behind them. The fewer respondents you get, the broader a bell curve you get on the range of values and the less likely the 'reported' value is actually the true one. It would be helpful if the pollsters released the data with several confidence levels, it would give people a better understanding of what the numbers mean. You can do the math yourself.

The margin of error is only relevant to the statistical math going on, it does not account for other errors in the polling through people not being honest in their answer or systemic errors by the pollster. For this poll it is +-4.56% at 99% confidence and +- 2.90% at 90% confidence.

The regional breakouts of an 800 person poll would lead to +- 4.9% at 95% confidence for the whole lower mainland. For Vancouver Island we would have an error of +-8.3% at 95% confidence. What these sort of numbers mean is that any results would have very flat curves and really are no better than hamburger polls in restaurants.

I would like to see some 2000 to 3000 person polls conducted to give us some data with better data and with some real regional numbers to work with.

I am looking forward to the next Mustel poll and see what it can tell us.


Anonymous said...

Thanks for the analysis.

Its bound to tighten. NDP counting on the complacency factor.

In 2005, if only 1,700 voters in 8 swing ridings had crossed the line, Liberals would have been history.

Anonymous said...

It's always interesting to compare the February/March pre-writ period in 2005 v. 2009.

In Feb. 2005, Mustel showed the Libs with a 6% spread over the NDP. Recently that Lib spread has increased to 16%.

In Feb. 2005, Mustel had Campbell's DISapproval ratings 11% higher than his approval ratings. Recently that trend has reversed with Campbell having approval ratings 7% higher than his disapproval ratings.

In Feb. 2005, Mustel had James disapproval ratings at 20%. Recently that figured has increased to 35%.

In Feb. 2005, Mustel had Green support pegged at 10%. Recently that figure was 12%.

In March, 2005, Ipsos showed the Libs with a 7% spread over the NDP. Today that Lib spread has increased to 11%.

In 2005, Ipsos had Campbell's DISapproval ratings 10% higher than his approval ratings. Today that trend has reversed with Campbell having approval ratings 3% higher than his disapproval ratings.

In 2005, Ipsos had James disapproval ratings at 23%. Recently that figured has increased to 38%.

In March, 2005, Ipsos had Green support pegged at 12%. Today that figure is 15%.

Today Ipsos also shows that BC is on the "Right Track" by a 53% to 41% margin.

Moreover, the Libs support is the firmest of all three parties.

In a nutshell, both Mustel and Ipsos show the Lib spread over the NDP has increased considerably since the same period in 2005, Campbell's approval ratings now outrank his disapproval ratings, and James disapproval ratings have gone up considerably.

The dynamics between the 2005 election and the 2009 election have certainly changed.

Anonymous said...

Mustel uses even smaller samples then 800. The most efficient number would be 1000 participants, after that point the gains in accuracy are minimal. That said, if a polling company insures that the poll is demographically balanced or weights accordingly, then the accuracy is increased.

Regional breakdowns are good but also should be demographically weighted. Mustel doesn't publish theirs, if they even take them into account. Ipsos' are uneven small and therefore unreliable.

Your probably right, the best poll would be huge. Regional Polls at 1000 respondents should give us some idea.

Bernard said...

Going from 1000 to 2000 does make a dramatic difference, the curve becomes much more of a spike or peak. I do see a benefit from going to the higher number.

With the ability to do automated dialing, large polls become much more affordable.

For useful regional break outs, I would want a province wide poll of about 4000 to 5000