Should the ALP come to power in the December 2007 election after winning 58 seats from the Coalition, I'll come back and eat some of my words on this blog posting.

In the meantime, I might start to say that a marginal 1-8 seat ALP majority might be possible. (That's the first time I've conceded that an ALP victory is anything other than an uphill - almost impossible - battle.)

Today, the Australian and Newspoll reported a strong swing to Labor. Extremely strong.

The balance is now 59 percent to Labor and 41 percent for the Coalition (two party preferred). In their last poll it was 55 percent ALP vs 45 percent ALP.

Martin O'Shannessy (Newspoll Chief Executive) told Marius Benson on Newsradio that the swing in today's poll was "well outside the margin of error". The implication was that it must be true.

Well, well, well. Somebody's being a little lazy with probability.

Let's recap how polls work.

A poll of 1000 people gives you a result with a likelihood of + or - 3%, where the + or - 3% has a 95% likelihood. Thus, if you ran 20 polls, 19 times you'd get a result within the + or - 3% range. And one time in 20, you'd get a result outside those boundaries. It's not a sign that you were bad at choosing your sample. It's not a sign of a sudden swing to one side or the other. It's just what happens. No matter how well you've chosen your sample, this type of inaccuracy is inevitable if you are only sampling 1000 people.

Newspoll is doing one poll every fortnight. 26 per year. So the odds are that they will be off by more than plus or minus 3% a bit more often than once a year. Every year. So things might not have changed at all since the last poll (ALP 59 today vs ALP 55 a fortnight ago). If you include the other polls (Galaxy, Morgan, AC Nielson), you're going to get a poll that randomly fluctuates by about this much every quarter or so.

I'm guessing that maybe the ALP vote is fairly solid, but it hasn't suddenly shot up. When the next Newspoll comes out, the ALP will be back down around 55/56 percent, and every commentator will come up with a reason: Maybe Howard will gain an "APEC boost" or Rudd will say something wrong. But it will merely be the random fluctuation of probability.


(This probably counts as my most feeble excuse for a photo in one of my blog entries yet. I'd better explain it. This is a blog about Opinion Polls. I'm punning on the word Pole/Poll The sign refers to the Pole Plot Track. That, in turn, is almost certainly a pun by somebody in Victoria's forestry department on the name Pol Pot, the genocidal leader of Cambodia from 1975 to 1979. Coles Ridge Road in Sherbrooke Forest. More photos here.)

Footnote 1: "... ALP come to power in ... December 2007". Howard has ruled out a January 2008 election. Most people are predicting a November 2007 election. I now reckon Howard will go late - either the last week of November or the first week in December.

Footnote 2: For completeness, even though it's not important in this example: The rule that a sample size of 1000 gives you a 95% chance of being within plus or minus 3% only applies if the split is roughly 50/50. If the response is skewed (eg 95/5) then the variability range changes.

Comments (0)
Anthony Holmes September 4th, 2007 07:39:05 PM

No Comments Found
 Add a Comment
Web Site:
Comment:  (No HTML - Links will be converted if prefixed http://)
Remember Me?