Facebook Google Menu Linkedin lock Pinterest Search Twitter

Advertisement

simulation

Jun 28, 2010

Share

Hidden among the events of Thursday’s spill was Newspoll’s quarterly release of their polling demographic breakdowns between April and June in The Oz. In what can only be described as very convenient timing, this allows us to pool these Newspoll results with the Nieslen results over the same period (plus one  unpublished phone poll I have from a company that was testing their sampling frame), to not only give us geographical and demographic cross-tabs of the last 3 months of the Rudd government – but also the demographic baseline that the Julia Gillard Prime Ministership starts from.

With these numbers, we can run them through our updated election simulation engine to see how an election would have played out on the ground were it held between April and June – providing results that can only be described as very, very interesting. You may even be surprised.

This is quite a long post, so it may be worth going and making yourself a cuppa first.

On the demographics, over the period between April and June, the Newspoll sample was 5766, the combined Nielsen sample was 4200 and our unpublished poll was 1500 – giving us a pooled sample of 11466 respondents. From that pooled sample, we can drill down and look at how the primary votes were behaving by geography, gender and age cohort, as well as how the two party preferred was operating by geography – all with fairly significant sub-sample sizes, giving us a goodly amount of certainty on the results.

First up, the primary votes. As we can see, there were pretty massive swings underway against the ALP across the demographic and geographical spectrum, yet swings which weren’t transposing into consistent primary vote gains for the Coalition. The numbers sort of speak for themselves.

alpprimarypooled

coalitionprimspooled

greensprimspooled

The Greens were the Party feasting on the spoils of ALP primary decline, particularly among the young and particularly in the capital cities – although not just directly through ex-ALP voters moving across into their column. As we saw with Nielsen preference dynamics and Essential voter firmness behaviour, there appeared to be significant compositional shifts at play, with the Greens also picking up ex-Coalition voters as the Coalition was itself picking up ex-ALP voters. If we look at what was happening at the state level of data, with the primary vote and the primary vote swing since the last election – we find that there was some quite complicated dynamics occurring between the states.

primaryvotegeography

primaryvoteswing

In Victoria, we had the interesting situation where the Greens vote had  increased by more since the last election than the ALP vote had fallen over the same period – leading to the rather odd outcome of the ALP two party preferred vote actually increasing (as we shall see in a minute). Yet the other states had the ALP primary vote falling by such large amounts that it could not be completely offset in two party preferred terms by the strong growth in the Greens primary. If we look at the ALP two party preferred results, we can see how the ALP primary decline and the Greens primary vote growth differentials came out in the wash.

alptpppooled

This is where it gets particularly interesting. As we can see, there were large swings against the government occurring in NSW particularly, and also capital cities generally. This gets us onto what the two party preferred polling actually meant at the end of the Rudd era.

If we look at our Pollytrend vote measures for all the polling done over the last 12 months while Rudd was PM, we can that there was a recent turnaround in the decline of the ALP two party preferred.

pollytrendRudd

Collectively, the polls had generally been trending upwards for the ALP over the last few cycles. Our Pollytrend measure is not a slave to any one poll or any one pollster – it looks through the polling noise and potential sampling error to find underlying trend movements in the true state of public opinion, by comparing how each pollster’s poll results change compared to their recent polling results.

Bob McMullan had argued in the party room and on Sky Agenda that the polls and their reality were being misrepresented in the media, and that the ALP was in a much stronger position than the reporting was suggesting *in terms of the way election dynamics historically play out.* A federal government has never lost an election if its two party preferred vote at the national level had a 5 in front of it – such is the power of incumbency when it comes to an election campaign. With the latest Newspoll being 52, McMullan argued, and quite accurately in that historical context, that the ALP was substantially in front of the Coalition and improving in the electoral race.

Others, one Newspaper in particular, have argued that because of the low ALP primary vote, only the ALP primary vote mattered and the two party preferred had become essentially meaningless.

That latter argument fails to grasp the complexity of what Bob McMullen was saying. Under compulsory preferential voting of the type we have in Australia, to win any seat, a party needs to get 50%+1 vote of the two party preferred vote share. That we know to be true.

With the low ALP primary but high Greens vote, the ALP was becoming heavily reliant on Greens preferences to win an election. As a consequence of this dynamic, one of two things *would* have occurred.

Either: as we witnessed with our voting firmness analysis, the Greens vote share was being inflated by soft voters and when an election was called, that Greens vote would decline and move back to the major parties, some of it moving back to the ALP and increasing the size of their primary vote (more likely than not, more of it moving back to the ALP than to the Coalition if history is any guide)

Or: If the Greens vote did not decline then those preferences would have to be distributed.

Regardless of what ultimately may have happened at the election later this year, far from the two party preferred vote being irrelevant, it was as important as it always has been – being the only measure of public opinion we have on the actual behaviour that ultimately decides elections. It just became more complicated as a consequence of an increased amount of uncertainty being attached to it – uncertainty that Greens preferences may not flow as strongly back to the ALP as they did at the 2007 election in the case of Newspoll results, and uncertainty that the stated preference of Greens voters with Nielsen polls may not be consistent with the revealed preferences of Greens voters come election day.

But just because something gets more complicated is simply no reason to ignore it. You just have to work your way through it with more sophisticated methods. This gets us on to how the ALP would have performed were an election held over the last three months and the election results matched the polling.

If we use an orthodox approach to this and plug just the latest Newspoll quarterly results into Antony’s spiffy electoral pendulum, it suggests that were an election held over the last three months, the ALP would have won around 81 seats to the Coalition’s 66 seats – give or take a couple either way as a result of the NSW Newspoll swing also containing the swing in the ACT.

A superficially good result for the ALP. However, what this result doesn’t take into account is how a large number of ALP seats sit right on the border of these swings. What we would actually see in an election is more variation on the pendulum – with some of those ALP seats on the margins of the swing falling to the Coalition and some seats the Coalition would be expected to take with those swings, not falling to the ALP.

If we move to the larger pooled sample of our demographic breakdowns, we can run those results through our election simulation engine that uses monte carlo analysis to accommodate for sampling error of the polls and the historical variation we see in the swings at the state level. This gives us a more robust, probabilistic view of how polling results transfer into likely seat numbers – as it accounts for seats with margins on the edges of the nominal polling swings, as well as the uncertainty of the polling swings themselves. Our simulation stabilised at 20,000 iterations (or 20,000 simulated elections) – giving us a nice smooth distribution of the results (click to expand).


alpseatssimfreq netchangesim

As we can see, once we account for the expected variation of the swings and polling uncertainty, the ALP was indeed in very real danger of losing an election were one held over the last three months and the result was compatible with the complete polling aggregates.

The most likely result was the ALP winning 76 seats, for a government majority of a single seat. If we turn those results into a cumulative probability curve, it becomes more stark.

alpseatsimcumulative

This chart tells us the implied probability of the ALP winning at least X number of seats. Simply pick a number from the bottom axis, trace it vertically until it intersects the red line, then trace it horizontally to the left axis to see the implied probability of the ALP winning at least that number of seats in an election held between April and June.

There was only a 65% implied probability that the ALP would have been able to form a majority in its own right.

If we rerun the simulation, but this time use Nielsen preference data as the two party preferred generator (where respondents get to allocate their own two party preferred preferences rather than have them allocated on the basis of the 2007 election flows), the most likely result would have been for the ALP to win only 69 seats, with there being only a 10% implied probability of the ALP winning government in its own right and a 30% implied probability of winning enough seats to even be able to attempt to form government with the support of all three independents that will be comfortably re-elected.

This makes the ALP “internal polling” that was being bandied around before the spill as quite important, as the purported numbers involved are entirely consistent with the polling we’ve been seeing over the last three months once we adjust for respondent based preference allocations.

It also makes them only a “moderately worse case scenario” under the assumption of preference flows repeating their 2007 election pattern – meaning that even if the internal party polling results were being slightly over-hyped to MPs and the media to encourage the dumping of Rudd, they contained not so much nuggets of truth about the most likely electoral situation, but large ingots of truth.

From a brutal hard-politics perspective, this adds considerable weight to the public opinion argument for replacing Rudd. However, it also needs to be kept in mind that the polling was improving, so the results would have been a little better than these simulations.

Not greatly better, but around half a dozen seats better in each instance.

This begs the question of whether the recent turn in the polling forced the hand of those in the Labor party wanting to remove Rudd – putting them in the position where if they did not strike last week, any further improvement in the polls, any further improvement in the polling trends, would effectively close any window of opportunity to replace Rudd with Gillard until well into the next term of government. Especially as any further improvement in the two party preferred would make many people start to question the authenticity and age of internal polling suggesting a wipeout.

Finally, and unfortunately in some respects (but entirely predictable in others), this gets us on to the tiring business of dealing with our old mate Dennis Shanahan and yet another one of his outbursts in The Australian. He states:

McMullan’s argument was that since Labor’s support crashed in April, the past four Newspoll surveys had shown Labor’s two-party preferred figures as 49, 50, 51 and 52 per cent and hence on the rise and on the way to victory. This was embraced by optimistic Labor supporters and academic or amateur pollsters who argued the primary vote and preferred prime minister were immaterial.

Read the whole article if you can be bothered, it’s the usual paint by numbers stuff.

Unlike Dennis, we analysed respondent allocated preference flows to attempt to understand and get on top of what was actually playing out with the two party preferred vote at the preference level. Unlike Dennis we actually analysed voter firmness to get a better grip on the dynamics of the large third party vote we have recently witnessed. Then again, unlike that sanctimonious windbag from The Australian, we actually do analysis here rather than word salad swimming in acidic dressing, where the substance of the meal rarely lives up to the hype of the menu.

Some of us that he’s presumably complaining about are professionals at what we do by any definition of the word – we are university qualified, we have professional experience, we have professional accreditation and operate as professionals in the fields of quantitative research –  while Dennis is, and always will be, just a journalist.

And without blowing one’s own vuvuzela, that can be seen in the content of this very article – unlike Dennis, rather than hide from uncertainty because it’s too difficult, taking the easy route of substituting proper analysis with noise, some of us have the professional skills to be able to embrace that uncertainty, adapt to the complexity it adds and come out the end with robust results from robust analysis using robust methodologies that can give rise to robust discussion. Results that actually contribute something positive to the sum total of human knowledge.

There are certainly amateurs and professionals when it comes to quantitative analysis – and then there’s those that lack the self-awareness to realise just which category it is that they actually fall into.

UPDATE:

In the next post, I’ve added another round of simulation results comparing the April-June period to Rudd’s final Newspoll.

Possum Comitatus — Editor of Pollytics

Possum Comitatus

Editor of Pollytics

Political Commentator and Blogger

Get a free trial to post comments
More from Possum Comitatus

Advertisement

We recommend

From around the web

Powered by Taboola

37 comments

Leave a comment

37 thoughts on “The electoral reality of Rudd’s last days.

  1. sprocket_

    on the gender issue, here is an unscientific sample of 2:

    mrs Sprocket_ was going to vote for anyone but Rudd, and now is rapt in Julia so much that she has tears in her eyes watching Julia’s speeches when she gets empathetic about working hard, raising kids, doing without…… definately connecting

    mother of Sprocket_ who is 79 and a life-long card carrying Liberal told me yesterday that she didnt like Julia cause she was a communist. When I asked gently what was the source of this information (which I can’t recall in the public domain), my mother said that her friends had told her that prior to joining labor Julia was in the communist party. So maybe the Obama was not born in America crowd have started similar rumours in the senior citezens clubs?

  2. The Gillard Spill « A Senex View

    […] It wasn’t the Polls – polling reality (short version, all trend measures favoured Rudd) […]

  3. Tri$tan

    This is all interesting stuff…

    Unfortunately all the spiffy analysis you can do will only predict what would (likely) have happened on that day.

    Rudd’s numbers had been trending down. They then had a slight upswing… who knows if they would have tended up or down after that. That essentially comes down to gut feel.

    I feel that Gillard will do better than Rudd against Abbott. I also feel that there was a pretty good chance that Rudd would have lost the election.

    So i think they did the right thing. Flip the switch whilst they are still in the lead and far eough out from an election to make it look all out panic. Additionally before the ‘parked’ and ‘soft’ vote that possum showed was emerging solidified.

    I think people were on the move and acting now was good prevention.

    However, we will never know.

  4. Possum Comitatus

    Lun – Antony’s calculator uses the new redistributions. My monte carlo sim uses new redistribution figures as well, except for one seat which is effectively a tie on the new boundaries and I put that in the governments column (via a “when there’s a tie – incumbent government gets it” rule that I use for these things) whereas I think Antony puts it in the Libs column by literally a handful of votes.

  5. Lun

    I may have misread things at Antony Green’s spiffy electoral pendulum, but if you reset it to the 2007 result option (52.7% 2PP) it doesn’t give the ALP 83 seats, it gives them 88 seats (Herbert, Swan, Gilmore, MacArthur and Dickson are all notionally ALP after boundary changes). That doesn’t seem to mesh well with the graphics showing Possum’s Montecarlo modelling (which appear assume a baseline of 83 seats – eg a 10% probability of a seven seat loss and consequent 76 ALP seats in Parliament). Antony Green’s judgement about notional ALP gains may be flawed a bit, but it looks as though Rudd’s last days may not have been so dire for the ALP; it may have been the case that they could “lose” 12 seats, without being back in Opposition.

  6. David Irving (no relation)

    Actually, morewest @ 13, a mate of mine proved conclusively (in about 1970) that it is, in fact, ballpoint pens that are the larval stage of coathangers.

  7. Possum Comitatus

    James – anything is possible at the moment apparently 😛

    Robbo – 1998 was an unusual election because of the flash-in-a-pan One Nation vote that pinched a lot of otherwise Coalition voters.

  8. James McDonald

    Excellent analysis, Possum. So, it will be close. Very close.
    Next question: what’s the historical likelihood that the outcome will be decided by “black swan” events in the final seven days?

  9. Robbo

    Possum, if I recall, in ’98, the coalition, after their first term, which started in ’96, was returned but with a lower prinary vote. I can’t help but wonder if Rudd had remained leader, if that is in fact the kind of scenario that would have played out?

  10. sickofitall

    My reading, though I note with interest the analysis just above, is that Mr Rudd was still on a winner – though had a better Oppo leader had come in, then perhaps…

    Australians like stable government: I think only McMahon (at the end of a long period of Coalition government) and Scullin have been booted out after one term… Keating? perhaps, though he did manage to win an election in his own right after winning the leadership 18 (?) months before… Whitlam got 2 (3, actually, then another one as Leader of Her Majesty’s Minority government or whatever he called it) elections…

    I still think that the right got it wrong: we’ll never know for sure. You can bet that if the ALP retains government, there’ll be a lot of jumping up and down ‘we were right’: of course, could Mr Rudd have won? We’ll never know. But I think he would have.

    And it goes down to trustworthiness. Mr Rudd was more trustworthy (at least in the eyes of the voters) than Mr abbott.

    Another thing is that ‘who would you have a beer with’ question… I don’t know that it is fair on either side – one, given the misogyny around, how many men would ‘have a beer’ with Ms Gillard (A lot here would, I know…). Two, if you’re me, you don’t drink beer (or any alcohol), so if asked ‘Would you have a beer with Mr Abbott?’ I could answer truthfully ‘No’, yet I might sit down and talk with him… Same with Ms Gillard.

  11. Peter Phelps

    Possum, your Quarterly Election Simulation appears to be missing ten seats?

    76 + 61 + 3 = 140

  12. Tomboy

    Just returning to Dennis Shamaham…is he the same one who once was part of a comedy act with Malcolm T Elliott on radio 2UW in the 1970s?

  13. Possum Comitatus

    Also Socrates – don’t forget the tobacco tax increase, the first by a government that I can remember where they went out of their way to claim credit for doing such a thing in the media cycle.

  14. Socrates

    Thanks Possum, a nice factual analysis as always. The Pollytrend graph is fascinating for clearly showing two things:
    – the serious decline started after the ETS backflip in April
    – if anything polls were improving through the RSPT fight

    So clearly, there is no excuse in the polls to dump the RSPT. The timing of the coup was opportunistic. If the plotters think they cna avoid doing an ETS or carbon tax deal, tell ’em they’re dreaming.

  15. morewest

    mbox@17

    Can we get a link to the BMJ article (or are you pulling our sockless legs?)

    It wasn’t an article, just a brief note postulating a possible link between the mysteries of missing workplace teaspoons reported by Lim et al, 2005 and disappearing socks.

    As always, Australian research is at the forefront in answering the really hard science questions! 😉

  16. B.Tolputt

    You gotta laugh. Primaries are wot counts, eh..?
    Last election had LNP’s primary at 42.1% – today’s Newspoll has it on 40%.
    Am I missing something – or is the Coalition, under Abbott, losing even rusted-ons?

    Yeah Abbott has lost 2% of the rusted ons… Not really his fault though, as they’re just naturally dying off. Given the reliance of the Coalition on the senior population for their votes, losing this election should mean they’ll only have 38% rusted on support come their next chance.

  17. lefty e

    That’s all very well Possum – but do you *own* it? 😉

  18. pancho

    BAM!

  19. Nipper Quigley

    You gotta laugh. Primaries are wot counts, eh..?
    Last election had LNP’s primary at 42.1% – today’s Newspoll has it on 40%.
    Am I missing something – or is the Coalition, under Abbott, losing even rusted-ons?

    That’s what it looks like.

  20. Thomas Paine

    Excuse me Possum, the Gillard transition story is a three legged table. You are not permitted to pull one of them away.

  21. mbox

    morewest: hilarious! Can we get a link to the BMJ article (or are you pulling our sockless legs?)

  22. Mr Denmore

    What’s so amusing about Shanahan is his quaint assumption that HE is the professional in this debate by virtue of the mode of distribution of his content – a dying rag kept alive only as a propaganda vehicle by Murdoch to destroy governments not dancing to his ideological tune.

  23. NickD

    Great post

  24. Venise Alstergren

    MOREWEST: I like it. I like it a lot.

  25. morewest

    David Richards@8

    …like what happens to all the missing socks in the world

    Advance details of my research into this was published in the BMJ some years ago. The short answer is that socks are the larval stage of the common wire coat hanger.

  26. Possum Comitatus

    You know Kymbos, I could probably work that into the routine.

  27. kymbos

    Ok Poss, you know I think you’re the best, but could you please provide a summary paragraph at the top of each post, rather than getting to your main point several paragraphs in? Sometimes I only have time for a glance, and I’ll come back to pieces later. But when I can’t make out your main point quickly I tend to have to forego reading a lot of your stuff unless I have time enough to read your entire piece.

  28. Possum Comitatus

    Marrickville Mauler, I may be able to count, but alas I shall never be able to spell. My sincerest apologies to Bob.

    And, oh yes – boom tish Cuppa! 😛

  29. Cuppa

    [This is quite a long post, so it may be worth going and making yourself a cuppa first.]

    You called, Poss? 😀

  30. David Richards

    …like what happens to all the missing socks in the world

  31. Marrickville Mauler

    Informative as ever Possum.

    I wonder how many political journalists actually have any idea at all what a monte carlo simulation is? (No doubt several inaccurate ideas which cannot be further elaborated consistent with posting rules on this site readily come to mind.)

    BTW there is a typo twice with the (sadly under-rewarded) Bob McMullan rendered as McMullen

  32. Possum Comitatus

    Oh, and let’s not waste anymore time on Dennis. There’s far more interesting things to fill our brainspace

  33. Possum Comitatus

    Big Ship – I actually think Dennis is effective as a “reporter” at The Australian. He fills his brief.

  34. The Big Ship

    You are far too generous, Possum, in referring to Dennis Shanahan as a ‘journalist,’ as that title, however diminished and degraded in this day and age of Murdoch-dominated media hysteria, may imply some degree of training. Shanahan may have ‘experience’ in the cess pit that is the Press Gallery, but that 20 years of experience seems to be one year repeated 20 times.

    Muck-raking hack? Jaded gossip monger? Zealous garbage sniffer?

    All a bit too harsh, perhaps …. what about hopeless, old DUD?

  35. Keith is not my real name

    Ouch 😉

  36. Tweets that mention The electoral reality of Rudd’s last days. – Pollytics -- Topsy.com

    […] This post was mentioned on Twitter by Pollytics, timdunlop, timdunlop, Oz, jonowee and others. jonowee said: Will the Ghost of Newspoll Futures haunt again? ||MT @Pollytics: The electoral reality of Rudd’s last days http://bit.ly/bRJvVb […]

  37. shepherdmarilyn

    Is shammers a journalist? Could have fooled me. He is a shill for the liberal party and newspoll and not much else.

    He is still on about ‘illegal” boats as if that word has the remotest currency and refuses to discuss it in any sense except to call people like me idiots.

    I just got an email from a very high profile, long term ALP member who has resigned.

Leave a comment