- My book: Bayesian Analysis for the Social Sciences (Wiley; Amazon).
- 111th U.S. Senate ideal point estimates:
- 111th U.S. House ideal point estimates:
- Next Australian election, Centrebet prob. of ALP win: 0.78 (2010-02-10, time-series)
This morning’s Australian aggregates the last two Newspolls to form state-by-state estimates of vote intentions. Here is Labor 2PP:
Lets repeat the exercise I did on the weekend, converting these shares into swings, assuming uniform swing within states, and read off some seat turnover estimates from the state-by-state pendulums.
Incidentally, analysis of the swings in 1996 (the last Big Swing election) suggest that about half of the variation in swing is between-state variation, while the other half is within-state variation (i.e., uniform swing within state, conditional on “knowing” the state swing, isn’t a completely stupid first approximation, at least for the 1996 data).
Thats 10 + 5 + 14 (Qld!) + 5 + 2 + 2 + 1 = 3839 seats. Labor on 98 99 seats in the new parliament.
I doubt that it will be this big. Insert usual caveats: its just a two week aggregate of polls, with pretty big confidence intervals around each state-level estimates; swings won’t be uniform within states: local factors, incumbency, candidate gender, hair color, dress sense, ballot order, etc.And if you believe sampling error and/or any Newspoll bias favors Labor (no on the former, maybe on the latter), then it won’t be this big; likewise, if you believe that departures from uniform swing within state favour the Coalition over Labor (maybe, maybe not), then yes, it won’t be this big.
But it will be big.
RSS feed for comments on this post.
Sorry, the comment form is closed at this time.
Powered by WordPress
Bad Behavior has blocked 397 access attempts in the last 7 days.
Yay, I’m first (maybe). I am also going to challenge your arithmetic….I make it 99, as you list five seats in Vic, but only count four. Obviously beyond your model which assumes uniform swings within states, but I wonder about seats like Kooyong, Higgins, Ryan etc. Kooyong esp because it actually swung to Labor in 2004, so the big question is, will soften up even more, or has it defied the rest of the nation by swinging at the wrong time???
Comment by Les Patterson — Thursday November 22, 2007 @ 10:03 am
Thanks Les. Fixed. Uniform swing is like fitting an average to a set of data, ignoring the variance. Then the next move in the argument is to say, errors cancel out on either side of the average. In electoral terms thats the assumption (hope?) that seats like the ones you identify, that won’t swing as large as the average swing, are offset by seats that swing higher than expected (i.e., bigger than average swing).
Thats the idea. Its kind of trivially true, in a “by construction” kind of way: i.e., for a set of data that has a roughly symmetric distribution, the mean is approximately the median, and half the data will lie above it, and half below it. Of course, this depends on knowing where that point is. Newspolls give us an estimate, but its a bit noisy in any given state, given the sample sizes.
Comment by jackman — Thursday November 22, 2007 @ 10:43 am
Hi Simon,
I’m an applied biostatistician and I’ve been following your contributions to the election prediction with some interest.
I’m interested in what you have to say about a couple of statistical contributions to the debate.
Firstly, Poll Bludger has a link to an analysis by Geoff Lambert:
http://www.pollbludger.com/lambert2007.pdf
Geoff is a medical researcher and not does not appear to be trained as a statistician, but he’s done this pseph stuff for many years. Using a combination of linear regression and motecarlo analysis, and taking into account state and even individual swings, he’s done a thorough analysis. He also has form. (An interesting statistic in his analysis is is the distribution of electorate swings.)
Socondly, this site has a go at predicting outcomes on a stochastic basis:
http://orgburo.com/elections/mockstateelection.php
I tried it with the Newspoll estimates on possumcomitatus for “current Newspoll cumulative in terms of State swings and the seats it would deliver”, see:
http://possumcomitatus.wordpress.com/2007/11/22/the-newspoll-cumulative-final-edition/
The estimated number of ALP seats is 100 based on state-by state swings:
http://www.abc.net.au/elections/federal/2007/calculator/?swing=state&national=0&nsw=7.2&vic=7&qld=11.1&wa=5.4&sa=9.4&tas=4&act=4&nt=4&retiringfactor=1
The estimated number of ALP seats is 92 based uniform national swing of 7.7% – the corresponding national swing
http://www.abc.net.au/elections/federal/2007/calculator/?swing=state&national=7.7&nsw=7.7&vic=7.7&qld=7.7&wa=7.7&sa=7.7&tas=7.7&act=7.7&nt=7.7&retiringfactor=1
I used http://orgburo.com/elections/mockstateelection.php to generate 40 outcomes using the state based swings and another 40 using the national swing.
Fot state based I got a mean of 96.1 ALP seats (SD=3.1, SEM=0.50 range 89-104) – the expected is 100.
For a uniform national swing the mean was 93.3 sd=3.1, range 87-100 – the expected is 92.
(True the underlying programming might be wrong. I don’t have the time to set this up myself.)
While the 2nd of this is perhaps reasonable the first isn’t.
I’ve drawn a few conclusions:
1) State based swings are important to consider.
2) More important is to consider the random error involved in translating swings into seats in a statistical model of seats. This is an additional variance component.
3) Simply using the electoral pendulum may result in a biased estimate of the number of seats in the new parliament. I’m assuming the orgburo.com is set up correctly.
What are you thoughts?
Comment by Bruce — Thursday November 22, 2007 @ 8:55 pm
One further comment.
I agree that the way to calculate the ALP Expected Seat Count is by summing the implied probabilities from the betting agencies. But have you done this for the Independents.
I did this analysis for Centrebet data on 31 Oct and got ALP favoured in 75 (LNP 73, Indep 2), expected ALP wins 78.2, LNP 69.2, Indep 4.6 – so there was at this stage clearly an error in this technique estimating independent outcomes. I took it to mean that the house premiun (vigorish of about 5%, http://en.wikipedia.org/wiki/Vigorish) was not distributed in proportioon to the implied probability of winning. In particular independents carry a heavier load. QAlternatively punters are not good judges of Independent’s chances.
Thoughts?
Comment by Bruce — Thursday November 22, 2007 @ 9:10 pm
I think the Coaliton will hold one of the 2 W.A seats you mentioned and lose Kalgoorlie, in fact I’ve stuck my neck out and predicted it.
http://australiavotes2007.blogspot.com/2007/11/prediction-labor-to-take-kalgoorlie.html
Comment by paulyt — Thursday November 22, 2007 @ 10:50 pm