Tuesday, March 13, 2018

The Keating Aggregation 1990-1993

A bit of a special feature for today ...

25 years ago today Paul Keating's Labor government won re-election against the odds, having battled a recession and fallout from the mid-term removal of the previous Prime Minister Bob Hawke.  Among all Australian elections, 1993 stands out as an oddity, the one that breaks almost every predictive election model that can be thrown at it.  If you want to get a feeling of just how unexpected it was, check out Lateline from a couple of nights before.

It is easy to forget that in office Keating was a very unpopular Prime Minister, widely considered arrogant and abrasive, and blamed for comments about "the recession we had to have".  He did not poll a single positive Newspoll netsat in 109 consecutive Newspolls on the job.  Yet history has been kind to him, in part because of this alleged electoral miracle.

Polling aggregates for all Australian terms back to 1996-1998 were available on the old Crikey Poll Bludger site (see sidebar of this Wayback archiving).  Some of the older aggregations there became possible after I recovered a large stash of historic Morgan data, also via the Wayback Machine.  But I have never yet seen an aggregation for 2PP polling over the 1990-1993 cycle, so about a year ago I decided I would prepare one and release it on this day.  And here it is.

Data sources are polls published by Newspoll, Morgan (mostly face-to-face with the occasional telephone poll), and, from the start of 1992, AGB McNair (a precursor of Nielsen).  My thanks very much to John Stirton for the later data.  Methods are a simplified version of my current 2PP aggregate, so this is a last-election preferences aggregate running off the 1990 election.  The 1990 election itself had seen a substantial preference shift as Labor ramped up its efforts to obtain Green and Democrat preferences (saving its skin by so doing) but 1993 did not. 

For this aggregate I've ignored all published 2PPs completely (rather than try to wade through which were last-election and which were respondent-allocated or something else; for most of the Newspolls there were no published 2PPs at all) so I've just calculated my own from the published primaries.  The AGB McNair face-to-face polls displayed a significant house effect relative to Newspoll and Morgan for much of the cycle; I've adjusted for this by simply applying a 0.7 point correction to the aggregate for the whole period they ran over, except when their most recent phone at the time was a phone poll.

So here it is:

Basically the re-elected Hawke government got no honeymoon to speak of and was soon behind.  It fell into the low-mid 40s by late 1990, with the recession clearly affecting government polling even before it was officially declared.  At the point marked 1 above, Paul Keating declared that Bob Hawke had reneged on an agreement to stand aside for Keating during the term, and launched a challenge against Hawke's leadership.  This was defeated and Keating went to the backbench.

Initially Hawke recovered in the polls, though at no stage was his position better than about 48-52.  In November 1991 Opposition Leader John Hewson launched the Fightback! package.  Hawke was seen as unable to respond effectively to Fightback! and government polling collapsed.  (The worst polls of the term were recorded here, including one which I convert as 59-41 to Coalition.)  Bob Hawke was removed as Prime Minister (point marked 2 above) and Keating took over.

Government polling was still sluggish at the start of 1992 but recovered markedly from early March, before dropping back slightly from late April following the Wills by-election loss.  Now and then the odd poll would even have shown Labor in the lead, had 2PP conversions been more widely practiced at that time.  However on the whole Labor continued to trail until late 1992.

In November 1992 (point marked 3) there was a sudden flip in Labor's favour, which the aggregate smooths out as less sudden than it probably was.  In a fortnight there was an eight-point swing in Newspoll and a four-point swing in Morgan, and in some following polls Labor leads as large as the equivalent of 56-44 were seen.  (AGB showed no real change through this time, but had already had Labor ahead.)  As well as rising opposition to Fightback!, a major cause of the flip was the election of the Kennett Liberal government in Victoria in October 1992, leading to enormous protests in Melbourne against its agenda.

Fightback! was taken to the panel-beaters and relaunched in December 1992.  Initially the relaunch was successful and the Coalition moved back into a small lead.  But in the last weeks even this came undone.  It is disputed what impact the infamous birthday cake interview ten days out may have had, and those disputing it tend to argue that even the final polls did not show Labor in an apparently winning position.  On my conversions, in most cases, they showed Labor in a position in which it would have been expected to probably win, but wouldn't have actually quite done so.  Although the result was called on the night, in the end Labor's win was very narrow in terms of the 2PP conversion to seats.

(The final aggregate reading comes out at 49.8, on which basis a win probably would have been forecast off the pendulum based on Labor winning the previous election with much the same, though I am still looking for a 1993 pre-election pendulum to verify this. I convert the final Newspoll and Morgan as both 50.7% 2PP to Labor - the end result was 51.4. Off the post-election pendulum 51.0 would in fact have been needed to win.)

Had polling aggregation been widely used during 1993 it's likely aggregators would have taken too much notice of Labor underperforming its polls in the previous cycles, and expected it to happen again.  So I am not sure whether aggregation would have been much of a predictive tool, especially not at the end of such a changeable campaign.

I hope this trip down polling memory lane has been of interest, and that it serves as a useful reminder that even when governments are behind in polling for nearly all a term, that is no guarantee that they will lose.

(Note added 22/05/2019: Thanks to Peter Brent for spotting an error in the bit in brackets above, which has been edited.)

(Note added 2021: See also the sequel, The Howard Aggregation 1993-1996).


  1. Very interesting Kevin and thanks

  2. When Keating retreated to the Backbench and Hawke unable to respond effectively to Fightback does that confirm Keathing's claim that it was he the economic genius behind Hawake's government and not Hawke?

    1. Probably. Hawke's efforts to replace Keating with other talent such as John Kerin were hardly a stunning success.

  3. Ahhhh, memory lane! I remember thinking that Keating deserved severe punishment for the "recession we had to have" remark - but punishing him by electing Hewson & Co would be punishing the same people that PK was being so callous about. I also remember picking up the Courier Snail on the morning of the election, with a last-few-days poll across the top of the page showing Labor level or just ahead, and thinking for the first time "Shed! (or a word to that effect) - Labor could win this!"

  4. An another leader of the Coalition could achieve such a comeback in the opinion polls and even increase the Coalition's majority at the next election. Just one issue namely massive cuts to the immigration intake could trigger a huge shift in the opinion polls hitting both Labor and lesser degree Greens support.

  5. Hi Kevin,

    I've got a pendulum for the 1993 federal election plus the 2pp swing for each seat from the AEC archives. Converting each seat back to its pre-1993 2pp figure, then applying a +0.1% swing to the Coalition, I find that only one seat would have fallen on a uniform swing (Paterson, 49.97% Liberal). Of course, I will note that the final AGB had the Coalition a bit further ahead than the Morgan and Newspoll (I get 51.9% Coalition 2pp, off primaries of ALP 43 LNC 48 Dems 3 Oth 6) and the final Newspoll did have a published 2pp (50.5% to Coalition).

    If I toss in the AGB poll and the published Newspoll 2pp instead of using a last-elections conversion (i.e. the best case for the Coalition from polling), I find a 2pp of 50.6%. Applying a uniform swing of +0.5% to the Coalition, they would be stuck at 69 seats (70 seats, minus North Sydney which was won by Independent Ted Mack) to Labor's 76 (77 seats, minus WIlls which was won by Independent Phil Cleary), with Labor still in majority.

    I can email you the pendulum data if you'd like.

    1. Yes please! That would be handy as I don't believe I ever did find a 1993 pre-election pendulum anywhere online.

    2. Just sent it.

      It's interesting to note how much of a shock Keating's re-election seemed to be to large segments of the media when in actuality he probably would still have survived had the final polls been correct. My guess is that after years of Hawke running relatively close races despite his sky-high approval and Better PM scores, few people thought Keating would be re-elected with net-negative Better PM/approval scores (if I'm not mistaken, he's just one of two PMs to do so, the other being Howard in 1998, although of course Howard lost the 2pp so it wasn't so much defying his approval score as luck + good marginal sandbagging).

      I also suspect the fact that the Coalition led Labor on primary votes probably misled many commentators. Without a 2pp + pendulum conversion, one might look at Labor being behind on primary votes and conclude that they were going to lose.


The comment system is unreliable. If you cannot submit comments you can email me a comment (via email link in profile) - email must be entitled: Comment for publication, followed by the name of the article you wish to comment on. Comments are accepted in full or not at all. If you submit a comment which is not accepted within a few days you can also email me and I will check if it has been received.