The big political news yesterday was a national opinion poll showing, on its face anyway, that Labor would easily win an election held right now. you didn’t see it? I wouldn’t have either, except that it was in my Twitter feed for about five minutes and I happened to be looking at it. AFAICT, none of the major national news organizations even mentioned it. There seem to be a couple of possible reasons for this. One is that some people don’t like Morgan as a pollster (I don’t follow the polls closely enough to have a view on this).
The second is the idea that a result so far out of line with other polls (52.5-47.5 for Labor) must be a “rogue” or “outlier”. This reasoning reflects the fact that political journos still don’t understand stats. It would be sensible to ignore a poll finding if it was the result of a breakdown in sampling procedures, or a biased question. But this is the same poll Morgan has been running for many years, presumably with the same procedures. What is more likely is that, by chance, this particular sample population was more pro-Labor than the population as a whole. Every sampling procedure is subject to this kind of error. But the correct response is not to discard the data, but to collect more, or combine it with existing evidence.
Given four or five of recent polls with results around 53-47 for the coalition, simple averaging suggests that the best estimate would now be around 52-48. A better procedure would be to use a Bayesian model. This guy has done it, and Hey Presto, concludes that the best estimate is 51.9-48.1.
Coming back to the statistical illiteracy of journos, the problem may be put as follows. On the one hand, they know that it would be silly to run a “Labor ahead” story. On the other hand, they don’t have the technical chops to explain Bayesian updating, or even weighted averaging, and to do so would make it impossible to write future stories suggesting that small variations in poll numbers have any meaning.
Even so, I think Tony Abbott has had a lucky break here (as has, in a secondary way, Julia Gillard). The only thing keeping him in his job is the perception that, while he may be unpopular, the LNP are sure to win. Even a single poll challenging that could pave the way for a spill. And if the result were to return Malcolm Turnbull, the outlook for the Gillard government would suddenly get a lot worse.