This morning’s Age reports the results of another Galaxy poll. Not voting intention this time, but something just as interesting: gay marriage.

Three years ago, when Newspoll surveyed attitudes to gay marriage, it found a slight majority opposed: 44% to 38%, with 18% undecided. But Galaxy now reports a huge swing in favor, with supporters outnumbering opponents 57% to 37%.

I don’t doubt that society is moving towards acceptance of gay marriage, and that attitudes are much more tolerant than our political leaders would have us believe. But a movement like that in just three years is simply not credible. One of those polls is wrong.

If you have to choose, Newspoll is more likely to be right. Firstly, because we already know from last week’s debate that Galaxy engages some dubious practices. Secondly, because this particular poll was commissioned by an organisation, GetUp!, with a strong vested interest in the result.

The problem is not that polling organisations deliberately doctor their results to fit what the client wants (although maybe some do). The problem is that polling is a difficult art; there are lots of ways that bias can creep in, and it would hardly be surprising if decisions were subconsciously being made in the interests of whoever was paying the bills.

This is an important reminder that there are two fundamentally different sorts of polling: those where there is some sort of external check on accuracy, and those where there is not. Polls of voting intention are in the first group — when the election is held, we can see which polls were right and which were wrong. As a result, there are strong competitive pressures towards polling accuracy.

As I said a couple of years ago, election results “are our equivalent of the cocaine traces in the river” — referring to an Italian study where testing for chemical residue provided higher estimates of drug usage than survey results.

But with attitudinal polls, like this one on gay marriage, there is no external check. The same goes for “preferred prime minister” surveys, and for much internal party polling. They could be getting it horribly wrong, and no-one would ever know.

If today’s poll leads to reduced discrimination, that’s a good thing. But don’t take the results as gospel.