Just when you thought the next federal election was a done deal along come the opinion polls.

Last week an ACNielsen survey, published in Fairfax papers, showed the coalition leading Labor 52 to 48 on a two-party preferred basis.

That would be an even bigger win than the last election.

Two days later, Roy Morgan Research produced the same numbers but transposed them to give a Labor victory.

And this week Newspoll in The Australian split the difference to give us 50-50. Who can we believe?

We must, before answering that, address the Morgan Question.

Morgan once enjoyed possibly the best reputation of all our pollsters.

Then came the 2001 election campaign, during which Morgan alone predicted an ALP win by a big margin.

Managing Director Gary "We are the best in the business" Morgan's characteristic assertiveness raised the stakes further, and when election day went to the government, albeit by a modest margin, Morgan's credibility suffered.

Morgan is also unique in sending interviewers into people's homes; the others phone them.

On the one hand, face-to-face over a cuppa elicits more honest responses.

On the other hand, phones have a wider reach, particularly among rural voters, who tend to favour the conservatives.

Morgan says this can be corrected by weighting; the other outfits point to the last election.

In any event, Morgan needs a good federal election to redeem itself.

Sample size is another variant. Morgan's, at about 2000 a pop, are twice the size of the other two. This should make them more correct.

And preferences, with which two-party preferred numbers are determined, are of increasing importance because minor party and independent votes are growing.

Here it is Newspoll that is out of step, only measuring primary votes and notionally distributing them as they flowed at the last election. The other two outfits measure full preferences.

Other factors, such as sampling methodology and rotation of questions, tend to remain close to pollsters' chests.

But the real answer to last fortnight's discrepancy is this: there is none.

To give context, recall the NSW election in March.

Nielsen's final poll gave 58 to 42 in Labor's favour, Morgan's was 60:40 and Newspoll's 60.5 to 39.5. ACNielsen proved to be the most accurate, but all were in the right ballpark.

As all three are today, but the ballpark is around the half way mark.

Notwithstanding Simon Crean's dreadful personal ratings and his party's primary support, the main feature of published polls in this current term is the closeness of the contest after preferences.

God didn't create opinion polling for us to take literally, just to give us an indication.

A federal election today would probably be close, with the Coalition favoured to win. That's what the published polls are saying, anyway.

Peter Brent is editor of mumble.com.au, a website that looks at electoral behaviour.