Comments on: Pollytrack October

By: Possum

Possum — Thu, 30 Oct 2008 07:13:28 +0000

Fair enough. The Nielsen thing last year was all about the timing – they started surveying for their final poll before the movement to the Coalition occurred on the Tuesday/Wednesday/Thursday period of the final week of the campaign.

That might be worth doing an article about actually.

By: ltep

ltep — Thu, 30 Oct 2008 06:53:14 +0000

Yes but how we view the veracity of pollsters must surely come from a number of sources, one being by looking at how close they have gone using the same methodology in previous elections, another being whether their polling technique may have in-built flaws which are tough to ‘model out’ of the equation. Using only the internet to poll would easily have the effects that it would skew the results for older people, the people with less money, away from rural areas etc. As you say they’d attempt to build there model to take these factors into account, but without having some solid result to back it up with you can’t know whether their modelling is particularly accurate. I’m willing to accept ‘online panel approaches’ as soon as I’m shown data which indicates that they are reasonably accurate. The fact that ERM seems to always have the ALP 2PP much higher than the other companies leads me to believe there quite possibly is a flaw in their sampling and adjustment models.

That’s what leads me to question whether including the ERM polls skews the results of ‘research’ like this towards the ALP. If ERM get in the ballpark at the next election I’d have no problems with it at all.

I’d like to see an explanation from ACN as to why their polls was so far out at the last election and what they’re doing to resolve the issue.

By: Possum

Possum — Thu, 30 Oct 2008 03:58:57 +0000

Being the new kid on the block is a bit weak. The online panel approach has as much potential to skew results as does a phone poll format or a face to face format. The question isnt the potential, it’s whether the company has adjusted for it’s unique problems in the same way that phone pollsters have to adjust for problems unique to phone polling.

I think they ran exit polls at the last election for Sky News which turned out to be pretty good – I’ll try to find out.

By: ltep

ltep — Thu, 30 Oct 2008 03:52:34 +0000

They haven’t been around long enough for their varacity to have been tested. That combined with its online format, which at least has the potential to skew results, makes the level that we can rely upon its accuracy dubious.

I’m aware they were around prior to the last election, what result did they have closest to the election?

By: Possum

Possum — Thu, 30 Oct 2008 03:30:00 +0000

Why are EMC polls dubious?

By: ltep

ltep — Thu, 30 Oct 2008 03:16:38 +0000

How much of this can be put down to the dubious Essential Research polls?

By: Possum

Possum — Thu, 30 Oct 2008 02:20:18 +0000

Cuppa,

The Essential polls aren’t online polls of the type they’re often accused of. What they have is a thing called an online panel where respondents are recruited via offline means (like by phone and mail) to participate in surveys that are then delivered online.

So every week, some large number of people are randomly selected from that online panel and are given an online survey to fill out. The results are then weighted by age and whatnot and become the poll results that we see.

But you have raised a good point about so called “house effects”, which I’ll start writing something on straight away.

By: Cuppa

Cuppa — Wed, 29 Oct 2008 22:19:27 +0000

Hey Poss,

Not sure whether I should put the following into Requestathon; I think I missed that boat … Some are saying they don’t put much faith in Essential’s findings, being as it is an online poll (self-selected respondents). How would the Pollytrack look with EMR’s figures excluded? And the Turnbull alleged “bounce”; would that have been more pronounced with Essential’s data (which does seem to favour Labor) excluded?

Having asked that, I also acknowledge that the bigger the aggregate sample size the more accurate, so having EMR’s figures included might just enhance the accuracy, rather than detract from it, yes? (Sorry, don’t know the terminology to explain it very well.)

Signed,

Confused.

By: Harry "Snapper" Organs

Harry "Snapper" Organs — Wed, 29 Oct 2008 12:02:26 +0000

This is turning out to be a most interesting way of analysing the polling data, wee marsupial. The more I’m following what you’re doing, the more I can understand it. Well, I think.

By: Ad astra

Ad astra — Wed, 29 Oct 2008 10:04:08 +0000

September 22 will go down in history as ‘The Turnbull Blip’. ‘Bounce’ is not apt; bounces are usually of longer duration. The problem seems to be that while he believes his own rhetoric, the public don’t.