## New day, Newspoll

According to today’s Newspoll, if an election were held tomorrow, the ALP would win 56% of the two-party preferred vote, winning by the largest margin since 1966Â World War II. Newspoll is effectively telling us that the Coalition’s odds of winning are below 1 in 10,000.

According to Centrebet, Labor pays \$1.90, which means Labor is a 49% chance of winning the next election.

One reason I prefer markets to polls is that they’re more accurate (see here and here for evidence). But another is that polls just say some loopy stuff. I find it impossible to believe that either party ever has less than a 1 in 10,000 chance of winning.

This entry was posted in Australian Politics, Media. Bookmark the permalink.

### 17 Responses to New day, Newspoll

1. Sacha says:

There have been quite a few recent state elections with very high two party preferred votes for Labor so a 56% result is possible.

How is the table in the “effectively telling us” link calculated – is it through a binomial distribution with standard deviation 0.5 centred on the “Poll Result”?

2. Mork says:

“Newspoll is effectively telling us that the Coalitionâ€™s odds of winning are below 1 in 10,000.”

I hate to be a pedant (NOT!), but isn’t the thing that Newspoll is effectively telling us is 10,000 to 1 on is not that Labor will win the election, but that more than 50% of the population would, if asked today, tell a pollster that they planned to vote Labor at the next election.

I actually think that’s a more than semantic difference.

3. Andrew Leigh says:

Sacha, yes.

Mork, interpreting Newspoll as saying something about electoral realities isn’t my own misleading spin. Here’s Shanahan on Monday: “KEVIN Rudd and the ALP are riding an environmental wave to the party’s best electoral position in six years as the new Labor team’s honeymoon with the voters continues.”

4. Peter Brent says:

Andrew, can you give the respective odds for a 2pp of 55, 54, 53, 53, 51?

5. Peter Brent says:

Cancel that, found it, I leapt before I looked.

6. Peter Brent says:

I agree with the man from Ork. An economist or statistician might reckon that a 56 to 44 poll result means a 1 in 10,000 chance of the Coalition winning the next electon, but no-one else takes it that literally. That’s a straw man. (Even allowing for Shanahan-hyperoble.)

Count me as a sceptic.

The reason Centrebet odds have narrowed since Rudd took over is largely because punters have been looking at opinion polls, particularly Newspoll. If Newspoll instead showed the Coalition ahead, you can bet the money would be moving the other way.

Andrew, I read this paper and while it is excellently written, powerful and persuasive …. speaking as a non-economist it seems to include a kind of self-sustaining, circular thingee that economic arguments sometimes do.

For example, you note the betting markets reacting to forest workers cheering Howard on the news, therefore “Labor’s Tasmanian forestry announcement provided the Coalition with a substantial boost.” Why?

(I doubt the forestry stuff had much effect on the result at all.)

If throughout 2007 the polls keep showing Labor way ahead, the party’s betting odds will continue to narrow. They’re a lumbering lot, not great at thinking ahead, these punters.

(btw 1966 was a bigger result than 56 to 44.)

7. Andrew Leigh says:

Peter, thanks for the 1966 correction, and the kind words on our ER paper. As to the “lumbering punters” argument, I offer you this quote, from a recent paper by Robin Hanson:

[I]n every known head to head field comparison between speculative markets and other social institutions that forecast, the markets have been no less, and usually more, accurate. Orange Juice futures improve on National Weather Service forecasts, horse race markets beat horse race experts, Oscar markets beat columnist forecasts, gas demand markets beat gas demand experts, stock markets beat the official NASA panel at fingering the guilty company in the Challenger accident, election markets beat national opinion polls, and corporate sales markets beat official corporate forecasts.

8. Sacha Blumen says:

Andrew, that’s a very interesting quote. I can’t resist saying that Richard Feynman, a physicist, was the one who worked out what went wrong in the Challenger accident. Did the stock markets beat him in fingering the cause of the accident?

9. Sinclair Davidson says:

Sacha, see here

10. Sacha Blumen says:

Thanks Sinclair. Interesting that the market got it “right” in the ’86 disaster without “knowing” the actual specific cause of the accident, and got it “wrong” in the more recent accident.

11. Chris says:

One sec…..if an election were held the day after the poll, all things remaining equal, it would be extremely unlikely that the LNP could win…perhaps not 1:10,000, but extremely, extremely unlikely.

The reality is that polls can only pick the mood of the electorate at a specific time, not the election result itself what with grandiose campaign promises, muckraking and cheesy campaign ads in the final month.

12. Andrew Leigh says:

Chris, this is a standard defence by pollsters. There are two comebacks.

1. As you acknowledge, 1:10,000 must be wrong. That means that their published confidence intervals are too small. Maybe they should be +/- 4%, rather than +/- 2%. Maybe bigger still.

2. If polls aren’t a useful forecasting tool for the actual election, why are they on the front page? Let’s face it, readers aren’t falling over themselves to know what would be the result of a hypothetical election. The newspapers are using polls as a forecasting tool – as such, they do very badly.

13. Chris says:

Hi Andrew,

1. Considering a normal distribution with sample size of 1000 (common for political polls), we’re looking at a margin of error of +/- 3% 19 times out of 20. Eliminating any questionnaire or sampling bias, that’s just cold statistical fact.

The problem with polls is that there is nothing at stake. Most research asks your opinion, and opinions have no consequence. Voting in an election is vastly different, as there is much at stake for the individual, whether ideology or a mortgage. Thus, polls measure a mood within statistical limits, but not necessarily an election outcome because they cannot adequately replicate an election voting scenario.

One can’t apply simply odds or statistics to election outcomes because people do not behave as rationally as mathematics. Human behaviour or psychology is the unknown dimension

2. Oh I completely agree. Polls are used to kill time on news bulletins, fill space in newspapers and to provide ammunition for much parlour talking on blogs.

However, polls within 7 days of an election can in fact be used to measure prevailing mood in that week and thus give an indication of outcomes.

As for odds, well I have a Centrebet account that is handsome purely due to betting on boxing “long shots”. I don’t bet unless I can get at least \$2 odds, and win a lot more than I lose. Statistically that is improbable because I go squarely against the market, but I happen to follow boxing closely and thus can narrow the odds with my nous alone. I like to win.

That’s the X factor, something at stake, much like the political polling example.