Making sense of the polls

Making sense of the polls

On Betting Markets and Party Expected to Win

Betting Markets

Are betting markets really a good way to predict the result of an election? In most Federal elections, they have got the result right, but an exception was 1993, when Keating won what seemed an unwinnable election. In 2010, Labor was heavily favoured, but only barely scraped a win. At the state level, betting markets erred in predicting that the government would retain office in the Victorian 1999 and 2010 elections and the WA 2008 election. Peter Brent writes about this here, here and here. To sum up, Brent says that the markets follow the polls and the conventional wisdom, but that they need many polls to change their minds. In the Victorian 2010 and WA 2008 state elections, the opposition did not lead until the final polls, and this was too late for the betting markets.

At this election, the conventional wisdom is that Labor has no hope, and this is currently reflected by the betting markets. As of Sunday 18 August, the Coaliton’s odds were $1.11, and Labor’s $6.50 on Sportsbet, implying a Coalition win probability of 85%. In 2010, the conventional wisdom was that Gillard would win easily.

Do Public Expectations of the Winning Party Matter?

The evidence is that they do not matter. In 2007, the public expected Labor to win, but the late swing was to the Coalition, and this happened again in 2010. Nobody gave Labor any chance at the 1999 Victorian state election, but they had a dramatic late swing towards them and won it. In the US, if people on one side think they are going to lose, they may not turn out, but Australia has compulsory voting. The best Australian politicians take care to keep public win expectations down, even when they are heading for a landslide. Not doing this can imply arrogance.

The concept of political momentum is closely connected to win expectations. According to this concept, a good poll generates positive press and will thus lead to more good polls. However, this has zero real-life evidence to back it up. Polls that miss the overall trend in either direction will generally come back to the trend at the next issue. Of course, sometimes the trend can move towards an outlier, but in these cases the trend changed because of other factors, not because of a good or poor poll.

A Suspect Lindsay Poll

An automated pollster took a poll of Lindsay for The Guardian on Wednesday night from a sample of over 1000. They found that the Liberal candidate’s primary vote support was a staggering 60% against 32% for incumbent Labor member David Bradbury. However, this poll seems skewed because it weighted for the last election’s results. Weighting for past votes is standard practice in the UK, but the problem is that respondents will misremember who they voted for at the previous election. Generally, more people will say that they voted for the winner than actually did. Essential asked a question on the last election’s results in June, and more people recalled voting for Labor than actually did.

If more people recalled voting for Labor in Lindsay than actually did, but the pollster weighted their results to ensure the past vote was near the last election’s results, this would bias the poll against Labor. However, if the Liberal primary vote is anywhere near the 60% estimated in this poll, the Liberals would still win Lindsay despite any bias.