Saturday, September 28, 2019

Australia's Closest Federal Elections (2019 Wasn't One Of Them)

This article is brought to you by the following quote from Brendan O'Connor, as Labor continues to grapple with its unexpected 2019 federal election loss and continues trying to work out whether what it did wrong this year was hardly anything or almost everything (or something in between):

"Some of the critiques to date, especially from outside the party, remind me of those absurd footy match reviews where despite the margins being very close, extol only the excellence of the winners and denigrate the virtues of the vanquished, even when there was just a kick in it."

He's right, for the most part, of course.  Analysis which praises everything the winner did (because they won) and pans everything the loser did (because they lost) is a massive problem in electoral commentary.  I refer to it as "annotation by result", a chess term for the same thing.

But there are a couple of big caveats here.  Firstly if you're up against Richmond or GWS, you might think a loss by a few points was a decent effort and that with only a little fine-tuning, if you catch them on a bad day next time round, you'll beat them.  But if you think you're a good team and you lose by a goal to the Gold Coast Suns, you might be sacking more than the captain.  One of the hard things with elections is that you can say how much one side won by, but that doesn't tell you if both sides campaigned well or if they were both hopeless.  Before the election the Morrison Government hardly looked like Grand Final material!

The second one is: was 2019 really comparable to a one-kick loss?  In some ways yes, because the government only scrambled to a three seat majority.  But on the other hand, the government won nine more seats than Labor, healthily won the two-party preferred vote, and did so well at sandbagging marginal seats that it wasn't even sweating on the post-count to know that it had won.  (In fairness to O'Connor he didn't actually say that the 2019 result was comparable to a one-kick loss, but that was just the media-received meaning of the comment.)

So through this last week I've been thinking about a multi-factor process for ranking the closeness of federal elections.  There's no one indicator that captures everything about how close an election was, or that is faultless, so when in doubt, aggregate.  I've aggregated the results of five different measures by two different aggregation methods (one based on rankings, one on magnitude), and then reweighted and averaged the two aggregates to produce what I think is a decent attempt at ranking which elections were close and which were not.  Of course, there can always be arguments; some questions aren't meant to be answered.  I've confined this ranking to the elections from 1910 on, since 1910 marks the emergence of a two-party system that has pretty much survived until this day.

Winning elections is mainly about winning seats.  That's a post hoc justification for seat tallies being three of my five indicators.  But it also matters whether a comfortable win in seat terms was a product of general voter satisfaction, or a result of a lot of luck in close contests.  So two of my indicators make use of two-party results.

I've worked all this out quickly and don't guarantee anything to be error-free.  But because I am aggregating across multiple indicators, any errors probably won't matter too much.  Here's a link to the spreadsheet I used, but there's no need to contact me about any minor errors that won't affect the overall rankings.  The five indicators are:

1. Size of government majority as a share of all seats

Obviously a government that has won with a big majority has done very well for itself while no government wants to (and few governments have had to) depend on capricious crossbenchers for confidence and supply.  So this indicator is the number of government MPs minus the number of all other MPs, divided by the number of seats.  Some judgement calls have been made in cases of state branches of government parties that may or may not have been formally parts of the Coalition.

Incorrect claims about the size of the government's majority are common in media reporting.  The majority is typically determined as the number of government MPs minus the number of all others, so that the Turnbull government started with a majority of two and the current government with a majority of three.  To say the current government has a majority of one is silly, because if it lost a seat to Labor it would still have 76/151 seats, which is a majority - albeit not a floor majority, so it would need the Speaker's casting vote to pass legislation opposed by the opposition and crossbench.

One issue with majority as a measure is it doesn't deal with differences over time in the size of the crossbench.  A majority of three is much more comfortable with six crossbenchers than it would be with none.  At many times federal elections have had few or no crossbenchers, but at times there have been many (particularly in the early days of the Country Party, but also more recently).

Another important issue with this one (and also the following two) is that what parties are part of government is sometimes determined post hoc.  So in 1922, the Nationalist-Country coalition had a majority of five.  But it acheived this only at the cost of Billy Hughes departing as Prime Minister as the condition for the Country Party joining.

The closest elections by this measure were: 2010 (a majority of -6), 1940, 1919, and 1913 tied with 2016.  1919 is a good example of why this is a dodgy metric, because the Hughes Government could well have formed a coalition with proto-Country elements if it had to.  The least close were 1975, 1917, 1977, 1925 and Labor's biggest win in 1943.

2. Difference between government and opposition seats as a share of all seats

This one does a better job of accounting for variation in crossbench size than number 1.  What it doesn't do so well is deal with the rare cases where a crossbench of decent size is known to support one party much more than the other.  There were a few cases where the Country Party did not go into coalition with the proto-Liberals but nonetheless clearly had confidence in their government.

The closest elections by this measure were 2010, 1913, 1961, 1969 and 1990.  2010 is the only case of a government not having more seats than the opposition.  The least close were the same five in the same order as for method 1.

3. Expected margin of confidence

This one gives an estimated margin for a vote of confidence in the new government, again as a proportion of the size of the parliament.  In general it assumes that Greens, Lang Labor and independent Labor candidates will give confidence to Labor governments, and that Country, independent Country/Nationalist/UAP (original)/Liberal candidates will give confidence to the Coalition.  However there are exceptions - eg I treat the Nationalists who voted (in effect) to dump Stanley Bruce's government as giving confidence to Labor.  Usually I ignore independents not clearly aligned to a party, but in some cases they made statements on confidence clear enough to assign them to a party (in some cases I have even braved to assign Bob Katter!).  Zali Steggall only said her "preference" was to deal with the Coalition, so I ignored that.

The reservation attached to this indicator is that a government reliant on crossbenchers who give it confidence is more vulnerable than one that has a majority in its own right.  And indeed one government that started out with a margin of confidence of two votes actually fell without losing any seats along the way.

The closest elections by this measure were 2010 tied with 1913, 1961, 1940 and 1974.  The least close were 1931, 1934, 1975, 1917 and 1977.  1931 was not so well picked up by the indicators above, because the Country Party was on the crossbench.

4. Two party preferred result for government

Here I've used the AEC 2PPs for 1983 onwards, just the shares of major party primary votes for the few first-past-the-post elections, and for the remaining years the average of 2PPs computed by various authorities and posted by David Barry here.

The main issue with 2PP is cases like 1998, in which a government uses personal vote effects and sandbagging to win reasonably easily despite losing the 2PP.  After all parties know the 2PP does not decide elections by itself and that what decides elections is winning seats.  Nonetheless, attempts to run a seat-focused strategy at the expense of 2PP quite often come up short.

Another issue with 2PP is that early elections often had uncontested seats and there is no right way to deal with those.  In some cases the uncontested seats were much more on one side than the other so leaving them out gives one side a 2PP advantage.

A third issue is malapportionment.  The Coalition had a malapportionment advantage at times prior to the 1970s; this was especially apparent at times during the Menzies years.  Often the malapportionment was not deliberate but a product of lax review schedules and overly generous variation allowances.  Some Coalition wins in this time were less close than the 2PP makes them appear.

The closest elections on this criteria are those where the government actually loses the 2PP (even if these are not the closest 2PP results), so the closest five were 1998, 1954, 1961, 1969 and 1990.  The most lopsided were 1931, 1943, 1966, 1929 and 1975.  It's notable that four of the top five here came after the previous government had collapsed on the floor of the House.

5. Swing to change result assuming uniform swing

This one produces some interesting results.  Sometimes governments win by a few seats, but win the seats they need comfortably.  Sometimes governments win comfortably in seat terms, but have a lot of close seats go their way.  This one asks: how much worse would the government need to have done on a uniform basis in order to lose or at least reach a deadlocked parliament?  Australia hasn't had a deadlocked federal parliament for any length of time, but it's doubtful a government would last long in one with a permanent floor minority if it provides the Speaker.

For this one I've had to make a lot of assumptions concerning who would and wouldn't provide confidence (as with method 3) and I've also had to cut some corners concerning non-classic seats.  In lopsided results especially, the pendulum gets crammed with contests where one of the majors fails to make the final two and so there is no 2PP result.  I've usually assumed that these seats would not fall with any swing, but I've made an exception for Wentworth 2019 where Kerryn Phelps is a reasonable proxy for Labor (at least to the point of assuming a 2PP swing of, say, 2% would have seen her retain the seat).

The main issue with this indicator is that a win that is comfortable in seat terms but close in uniform swing terms is still further from defeat than one that is less comfortable in seat terms, because an event affecting many seats is less likely than one affecting a few.  However the other indicators will pick up that.

The closest results on this indicator are 1961 (the infamous if overrated Communist preferences story!), 1940, 2010, 1993 and 1954.  1993 is a surprise inclusion - although Keating's ALP won re-election with a seat result of 80-65-2, there were eight seats they won with less than 50.5% 2PP, making this a closer win by this method than the one-seat margin in 1913.  The least close come out as 1931, 1966, 1955, 1975 and 1934.  1931 comes out as a huge outlier here on a margin exceeding 13%, but this is likely to be exaggerated by the numerous non-classic contests.  In any case, that helps cancel out the artificially narrow majority of the Lyons Government because it did not need Country Party support.

David Barry's site was immensely useful for this section, as were the pre-1925 pendulums on Wikipedia.  Any errors in use of the data are mine.

The closest and least close elections

After all this I get the following ranking for the closest ten elections:

1. 2010: Labor wins a government in minority under Julia Gillard after weeks of post-counting and negotiation.  This result is the furthest into the red in seat terms, the only one where a government has not had more seats than the opposition, one of the weakest positions on confidence and not much to write home about on the 2PP front either.  Plus one more seat loss and it wouldn't have happened at all.

2. 1961: The credit-squeezed Menzies government survives with 62 voting seats to 60 after winning the last seat by a whisker and losing the 2PP.

3. 1940: The first Menzies government survives for part of a term after losing its majority and being reliant on two crossbenchers.  Although this is a minority government and not a majority as in 1961, it ranks as less close than 1961 in confidence terms because the parliament was smaller, and also the 2PP was slightly higher, as was the margin by which chaos was avoided in the last seat won.

4. 1913: Joseph Cook takes office by a single seat, only to throw it away a year later when his attempt to get a bigger majority is ruined by World War I.  The only one-seat majority doesn't rank as high as it might because in those days the parliament was smaller.

5. 2016: Malcolm Turnbull underperforms in a long and tedious campaign and eventually struggles to a two-seat majority with a minor 2PP win and a substantial crossbench.

6. 1954: Robert Menzies' government comes back from a horror budget and some of the worst polling ever seen with a free kick from the Petrov affair right at the end (though by then he was probably winning anyway).  Menzies' majority is seven seats but he loses the 2PP and is very close to defeat in swing terms.  There's a case for moving this one below the next few because of uncontested seats affecting the 2PP (see comments).

7. 1974: A close call for Gough Whitlam at his only re-election attempt as PM (he'd been sacked before the second one) - he stays in with a five-seat majority.

8. 1969: A calamitous collapse for the Gorton government which survives a massive swing with a seven-seat majority despite losing the 2PP.

9. 1990: The Hawke government loses the 2PP and only holds off Andrew Peacock with an eight-seat majority.

10. 1998: Although John Howard gets back with a 13-seat majority, his is the worst winning 2PP for a federal government, and on a uniform basis another 1% of swing would have been the end of it.

Ten biggest thumpings

The following are the ten least close elections on my rankings:

1. 1975: The post-Dismissal election sees the Labor Party smashed by Malcolm Fraser's Liberals.  Although not the biggest win in 2PP or swing-to-change terms, it is the biggest win in terms of government majority and the third biggest on confidence margin.

2. 1917: Billy Hughes leaves the Labor Party, merging with his opponents to form the Nationalists, who thrash the Labor remnants in a lopsided wartime election.  Again not so huge on 2PP or swing-to-change but an enormous majority in seat terms.

3. 1943: The other wartime pasting; this is John Curtin's Labor government cleaning up the rabble of the Fadden-Hughes coalition which had collapsed mid-term in the previous parliament.  A little less overwhelming in seat terms and swing-to-change terms than 2PP terms, perhaps because of malapportionment.

4. 1931: Joseph Lyons comes to power in a landslide after Joseph Scullin's Labor government collapses under the weight of the Depression and internal chaos.  The biggest win in confidence and 2PP terms and way ahead of the pack on swing-to-change (though probably inflated there as noted above).  However gets marked down on the other indicators and avoids being top of the list because the Country Party didn't join the government.

5. 1977: Another massive win for Malcolm Fraser as Labor wastes an election by maintaining its rage with Gough Whitlam.

6. 1966: Harold Holt's thumping win against a dated and xenophobic ALP under Arthur Calwell (who only five years earlier had been very competitive).  This ranks second, albeit a distant second, on my swing-to-change measure.

7. 1934:  Joseph Lyons retains office against an ALP still reeling from its Depression loss.  Although Labor got a big swing back on 2PP at this election, the swing-to-change measure suggests that the 2PP result flattered the Opposition.

8. 1925: A thumping win by Stanley Melbourne Bruce's Nationalist-Country Coalition, ranking fourth largest in terms of government majorities.

9. 1929: Finally another Labor win and this is Scullin taking office after Bruce's government collapsed.  The fourth-biggest win by 2PP but little better than average on the swing-to-change measure. 

10. 1958: Menzies had two very easy wins in the 1950s, 1955 and 1958, in both of which the swing-to-change measure suggests an even easier win than the 2PP suggests.  There's a strong case for including 1955 in this top ten too - the 2PP was blighted by uncontested seats.

It will be noted that only two of the ten biggest wins are by Labor.  Labor's next biggest wins are 1946 (14th) and 1983 (16th). 2007 doesn't even make the top half, perhaps because of personal vote effects from the 2004 drubbing.

What about 2019?

Here is the detailed scorecard for 2019:

* 7th closest on government majority
* 9th closest on government lead over opposition and expected confidence margin
* 17th closest on 2PP
* 20th closest on swing to change

On the 2019 post-election pendulum, the closest Coalition wins were Bass (0.41), Chisholm (0.57), Wentworth (1.31 vs IND) and Boothby (1.38).  But even assuming the Coalition lost all of those, it would probably still have governed in a 73-71-7 house, at least initially, though the whole thing would have been a total mess and could well have led to a mid-term change of government or a fresh election.  It takes a much bigger uniform swing of 2.69% to knock out another seat, Swan, at which point the outcome is anyone's guess.  So on the swing to change measure, this was actually a typical federal election result, and on 2PP it was only slightly closer than the median.

My rankings have 2019 11th (and a long way clear of 10th).  That places it around the 25th percentile.  Checking the frequency distribution for AFL games, this means elections as close as 2019 happen as often as 12-point defeats.  So Labor didn't lose by a kick; rather, they lost by two straight kicks.

2022 Update

I have fed 2022 into the spreadsheet and rank it:

* =7th/8th closest on government majority (with 2019 of course)
* 17th closest on government lead over opposition
* 19th closest on expected confidence margin
* 22nd closest on 2PP
* 32nd closest on swing to change

I have regarded all the non-Green crossbenchers as neutral on confidence, there being arguments in various directions on many of them.  (Post-election confidence offers from independents at a point where only Labor could win are considered to have been made for the sake of stability and ignored.)  While the election is close in terms of government majority, it is not close overall, and on balance I rank it the 19th closest out of the 44 elections in the sample (more or less in the middle, and less close than 2007.)  This seems a fair summary of an election where Labor scraped a majority over all others but the Coalition recorded the worst seat share result for the other side since 1943.


  1. Excellent and most interesting and I endorse your conclusions. Two queries:
    In 1954 six safe Coalition seats were uncontested against 1 ALP seat, and about 3 other ALP safe ALP seats were uncontested by Liberals or effective Liberal surrogates, boosting the ALP vote. I wonder therefore if any 2PP is effective (and there was no concurrent Senate poll to fill in the gaps as occurred in 1955).
    Secondly did malapportionment work for the ALP as much as the Coalition in 1966? The lack of a redistribution had perpetuated such shrunken ALP seats as Dalley, West Sydney, previous Scullin, etc.

    1. Yes most of the 2PPs up til 1955 are affected by candidates winning unopposed meaning that there is no 2PP for their seat. The most serious imbalances I can find are 1955 (10-0 conservative-Labor), 1928 (9-3), 1954 (6-1), 1937 (0-4), 1917 (7-3), 1929 (3-6), 1922 (4-1). Taking that into account 1954 would move down a few rungs on the closest elections list and 1955 would come into the least close seats list. A related issue which I haven't looked at is cases of seats that were uncontested on a 2PP basis but had a token candidate contest (one major party vs Communist for example).

      For 1966 the Coalition's advantage on swing-for-different-result compared to 2PP comes in similarly to other elections around the same time (about a point, with a fair amount of variation between elections.) However Labor should indeed have been more favoured by malapportionment as the average seat won by Labor had about 11% fewer formal votes. The discrepancies in that election between Bruce (with over 110 000 formal votes) and some others with only around 25K are quite amazing by modern standards.

  2. What about state/territory elections?

    I know in 2012 in the ACT, Liberal and Labor got nearly identical primary vote, but I don't think it was that close in terms of Liberals being able to grab another seat

    1. There have been some incredibly close state/territory elections. One could probably start by listing all those won by 1-2 seats and ranking them in order of vote margin in the closest seat. Queensland 1995 would rank highly as it was initially decided by one seat won by 16 votes, then that was thrown out in court leaving the election as effectively a tie until the same seat voted again in a by-election. In Victoria 1999 the Liberals were the same number of votes away from getting exactly half the parliament, at which point the rural indies would probably have supported them and not Labor.