According to pollster Gary Morgan, last year’s election showed that voters “don’t allocate their second preferences as they say they will,” and that therefore pollsters should just extrapolate the preference flow from the last election to get two-party-preferred estimates (read his view on the Morgan website here).
Some background here: Morgan’s polling became notorious at the 2001 election when he confidently predicted a Labor win that didn’t happen. Last year, by contrast, his final pre-election poll was very accurate, at least as far as primary votes go. His two-party-preferred forecast, however, was 51-49 in Labor’s favour (against the reality of 52.7-47.3 for the Coalition), and it is this discrepancy that he is now trying to explain.
Peter Brent (Mr Mumble) explained the whole debate very well in a December 2004 article in Walkley Magazine (here), commenting that “how [Morgan’s] preferences went that way is anyone’s guess.” Now we have Morgan’s explanation – that voters mislead the pollsters. At least, they mislead his pollsters, just as they did on the big picture in 2001.
It may be special pleading, but what Morgan says actually makes a lot of sense. Intending minor party voters are a very small sample in the first place, which makes the results more prone to error. Moreover, we know that a lot of people just follow the how-to-vote cards without thinking about where their preferences are going, so there’s no particular reason why that should match up with what they tell pollsters.
Extrapolating the distribution from the previous election could well give a better result, but it’s not perfect either: minor parties can change their how-to-vote recommendations, or new minor parties can appear (eg. Family First in 2004). Whichever method they use, poll results still have to be interpreted with a large measure of commonsense.