Where have all the pollsters gone?

From memory, most political polls are conducted as loss-leaders by market research companies. While they don’t make money on them, they don’t cost too much (as they’re often conducted by adding questions to a consumer survey which is being run anyway) and are great advertising for their commercial services. I guess that Canberra is too small a market to justify the expense of getting a sample together.

Jakez, just did a quick poll of the office. everyone thinks Stanhope is a negative for Labor. One of the more extreme views was “Stanhope is a F@$%#*& tosser” another ” He’s a Knob” small sample, only 10, but might give a guide to the outcome of the election.

(Stuffed up my terms: confidence *level* = 95% in this example, confidence *interval* is the +/-)

@jimbocool – someone who knows what they’re on about! Though – the confidence interval gives you the +/- for the *survey* results, had you interviewed everyone in the population (i.e. ACT voters) rather than just 800 of them at random. Not that it predicts the *election* results to within that +/- (the stats won’t tell you how many people will be lying to pollsters or change their mind on election day). Though that’s probably what you meant.

Thanks jimbocool. It’s been a long time since I did statistics and the correct terms long ago departed my brain.

how about the riot do a poll on each elec.. each week and see how they go

actually 800 is plenty for statistical significance – a random sample can be as low as 16 and still be statistically significant. What you really want to know is what is called, erroneously, the “Margin of Error”. Most published polls use a 95% confidence interval (that is 95% of the time the results of the poll will reflect the actual result plus or minus the MoE) and the MoE on a sample of 800 is around +/- 3.5%.
Terry Aulich from UCan usually publishes a poll (I can’t remember the brand name they use) in the week before the election, and while he claims it’s always right, my memory is that the methodology is pretty suss.

Another part of the problem with the ACT is the Hare Clark system makes it hard to get specific results – if you spend good money on a 800 sample you need to break it down to electorate level which is 270 odd, so the error gets larger as you look at the seats themselves. This is significant as the final seats have in the past been decided by mere handfuls of votes – from memory Zed was only a couple of hundred votes ahead of Hettinger in 2004, and Dunne was less than 100 ahead of Hird in 2001.

caf said :

jakez: The major expense is in paying people to sit on the end of a phone, read out questions and write down the answers. You need to pay for one-on-one time with at least 800 respondents.

Yep good call (and 800 is pretty low for statistical significance as well).

jakez: The major expense is in paying people to sit on the end of a phone, read out questions and write down the answers. You need to pay for one-on-one time with at least 800 respondents.

Polls cost the same to run in the ACT (you need the same sample size to get a reasonable result) as they do in NSW or Victoria, but the newspapers in those markets have a lot more revenue to pay for them. Even WA and SA don’t have polls of great quality, and NT and Tasmania usually get nothing, like us.