It seems we have to go through this every three years. Once we get into the second half of a government’s term and people start taking the opinion polls seriously, they seem to forget everything they learnt in the previous cycle. So each poll has to be invested with huge meaning on its own, independent of any sort of trend or context.
So let’s review some basics. Start with Peter Brent (who now blogs for The Australian), explaining things in the 2007 Crikey election guide:
“If news items were given the emphasis they deserve, political polls would [be on] say, around page eight … But opinion polls cost a bomb to produce, so onto page one they must go. Then everyone must pretend that’s where they belong, adding several hundred words of interpretation — turning them over, looking for meaning, interpreting them as good or bad for someone or other, pretending you can identify why the numbers move over a fortnight.
“Don’t get excited about a dramatic movement in any single opinion poll.
“Each is just an imprecise dip in the ocean. Wait for another, and then another.”
Because polls are proprietary products, newspapers tout their own rather than the competitors’, although the most sensible results come from averaging several different polls. If you want to really understand what’s happening you need to look at the trend, since any one poll on its own is basically meaningless. And obviously (but this story is all about people who ignore even the blindingly obvious), polls become more useful as you get closer to an election.
Fairfax is probably the worst offender, because its poll, by AC Nielsen, comes out the most infrequently — today’s is the first for five weeks. As by now you’ve no doubt heard, it’s bad for Labor. But lack of frequency doesn’t make a poll any more accurate, and even taken in conjunction with a similarly bad Newspoll last week, it doesn’t really justify the avalanche of commentary.
To see why, look at Nielsen’s trend lines. (Don’t look at the big red and blue numbers beside them, because they’re wrong — two-party-preferred should be 43-57, not 47-53.) The red line, showing two-party-preferred vote, shows a gradual but sustained improvement from Labor’s low point in the middle of last year. Yet pundits seem in denial about this simple fact.
Don’t get me wrong; it’s entirely possible that the past few days of polling are the first sign that the trend has kicked into reverse. But it’s also possible that they represent just a temporary Queensland-induced blip that will wash out in the next round or two of polls. We don’t know — and on the basis of only one poll (even one from each pollster) we can’t possibly know.
Five years ago we had the opposite problem. The polls showed the opposition consistently ahead by huge margins: Nielsen in March 2007 put them at 61% to 39%. Yet most pundits gave them short shrift; they were bewitched by John Howard’s comebacks from behind in 2001 and 2004, and remained convinced that he might be able to do it again. The Australian in particular conducted a fierce campaign of disinformation to deny the conclusions of its own polls.
But looking at the trend explained the difference. In early 2001 Howard was well behind, but he was making up ground; in 2004 the polls were erratic. In 2007, however, the trend line flattened out with Labor well ahead, and sure enough, although things tightened a little during the campaign, that’s the way it turned out.
Alert readers will already have noticed a problem with the dates in this story. 2001, 2004, 2007 — we have three-year terms, so the next election is in 2013. So why are we obsessing about this now?
That’s the other big problem with today’s coverage — it completely fails to convey the fact that we’re still some 17 months away from voting, and that there is plenty of time for a lot of things to change.
Michael Gordon does mention this morning that the trend (on the assumption that today’s result is just a “blip”) showed “a recovery that could see Labor competitive if it continued until polling day.” But in reality if you just project the trend in a linear fashion (a simplistic way to do it, but Fairfax is in no position to accuse anyone else of being simplistic) it would have Labor not just “competitive” by late next year but in line for a landslide victory.
Similarly Michelle Grattan on Friday drew all her comparisons with 2001, when the proper yardstick would be 2000. At that point, Howard’s polls weren’t awful — they turned really toxic in the second half of that year, when the GST implementation proved to be a disaster.
If implementation of the carbon price is that bad, Julia Gillard will be in trouble. If it goes relatively well, she’ll have ample time to turn things around. Or any number of other things, good or bad, could happen to upset calculations.