Opinion poll opinions

With eight months to go before the next Federal election just how important are the opinion polls?

 Obviously they are important to the members of the media who write relentlessly about what they mean. Equally they are important to polling firms who get brand exposure which benefits the many other research activities they undertake. But how important they are in forecasting results is another question altogether.

 Many people prefer to use betting markets such as Intrade. Australian betting markets currently show a 73.1% probability of a Coalition win in the next Federal election. Betting markets have a good record although in the US thin betting volumes allow candidates, in primaries for instance, to move the market with smallish investments and then generate media coverage and comment about alleged momentum. On the other hand they get the benefit of insider trading in that people with good party research sometimes have better insights into outcomes. But the emphasis has to be on ‘sometimes’. The most recent Victorian State election result was unexpected by both parties as the swing apparently didn’t surface until the last 48 hours and even then it was a close run thing.

 Now the odds are very definitely still in favour of a Coalition win in the next election although if you get good enough odds (a friend offered me 4 to 1 which was too good to refuse in a two horse race; and another friend told me he got given 10 seats in which is also attractive) a wager on Labor is not stupid.

 However, forecasting any result eight months out from an election is always difficult. In recent Federal elections the government, if performing in the polls the same as the Gillard Government currently is this far from the election, has won in more cases than not.

 In Nate Silver’s book The Signal and the Noise he shows a chart (page 63 or on his blog fivethirtyeight.blogs.nytimes.com if you are an online New York Times subscriber) of the probability of a US Senate candidate winning based on the size of lead in polling average. The data is drawn from elections between 1998 and 2008 and it shows that if a candidate leads in polls by 10 points a day before an election the probability of winning is 99.7%; 98% one week; and 79% and 67% respectively six months and one year out. For a five point poll lead the probabilities are: 95% for one day out; 81% for one month; 66% for six months; and 59 % for one year. A one point poll lead gives probabilities of 64% one day out from the election; 57% one month out; 53% six months out and 52% a year out.

 Now some provisos: Senate elections in the US are different to Australian elections for many reasons – not least the voluntary voting; the average of all polls in a Senate election are a significant sample; and, we are eight months out from the probable election date of September/October and the Silver chart doesn’t do eight months. But it is possible to extrapolate and suggest that the probability of a Coalition win (depending on which poll you use) is between 47% and 59% and the probability of a Labor win are between 53% and 47%.

 All in all – very similar to the odds in tossing a coin – and a bit closer than betting markets suggest. Of course, one can also take the view of J.K.Galbraith who was pestered to give a prediction on an election outcome believed to be close. He refused – saying he would be certain of the result after the election.