Wednesday August 14, 2013
A phone poll by Morgan on August 12 and 13 resulted in 57% TPP for the Coalition, 52% Coalition primary vote, ALP 31%, Greens 9%. n=569.
And Morgan themselves are pouring cold water on the result:
Gary Morgan says:
“It needs to be remembered that telephone polls are biased towards the party that receives the best publicity. Tony Abbott’s ‘free kick’ from News Corp over the last few days accusing Rudd of cheating in the debate by using notes has no doubt boosted the Opposition Leader’s credibility although his comment yesterday on ‘sex appeal’ was crass, definitely irrelevant and unnecessary. Prediction: You can expect all telephone polls conducted over the next few days to show a jump in support for the Liberal-National Party!”
Maybe, but 57-43?
And the only problem with Morgan trying to downplay this result is that Morgan’s last phone poll before the 2010 election actually did pretty well: a small over-estimate of the Labor primary vote (39 vs 38) and 51% for the Labor TPP (actual was 50.12%). See my GA post on this from yesterday.
In fact, Morgan phone polls have the best track record of the various interview modes they use, based on my recollection of the various analyses I’ve done on this over the years. My model assigns a pretty small house effect to Morgan phone polls (+0.6pp for Labor TPP, +/- 0.5pp). Compare Morgan’s other poll modes: face-to-face, +2.4pp Labor TPP, +/- 0.5pp; multi-mode, +2.1 Labor TPP, +/- 0.5pp.
Prior to this Morgan phone poll coming in the model was putting Labor in the high 47s. Even with the small nominal n of this poll (569), the Bayesian compromise has to move in the direction of the new data point (43%, modulo the small house effect estimate). Accordingly, my calculations suggest that the new estimate is 47.3% Labor TPP +/- 1.5pp.
New numbers and graphs will appear at the links etc in the header of my blog in the next few hours.
This poll is one of the “worst-fitting” data points I’ve got in the data set of 261 polls I’m currently working with, 2010-present. That is, given all the other polling, given the estimate of the (small) Morgan phone house effect, something like 48% Labor TPP would have been more like it, +/- 4.6pp, given the small nominal sample size. 43% Labor TPP under these circumstances is, well, an unusual result given what else we’re observing, and all that we’ve observed about the performance of Morgan phone polls in the past.
This influence of this data point will get steam-rollered by the next round of polling, but for now, there it is…