No truth stands alone.

Why a House/Senate Split Election Is Bad for The Government

Recently, there has been a great deal of unfounded speculation (mostly on Twitter) that the federal government will attempt to split elections for the two Australian legislative chambers. Such speculation stems from a combination of the government’s currently-poor opinion polling and perceptions on the part of such users that the government must be desperate to hold onto power at any cost.

Can the government split the House and Senate elections?

In short, yes.

This highly-recommended piece by Antony Green goes into much more depth, but broadly speaking:

  • House and Senate terms operate differently. Representatives in the House have serve a maximum of three years from the time they were elected and sworn in, while Senators serve fixed six-year terms (backdated to the previous 1 July for double-dissolutions).
  • This means that the terms of the current House will expire at the latest on 1 July 2022. On the other hand, newly elected Senators must take their seats in the Senate by 1 July 2022.
  • In theory, this means that the current government has the ability to delay the calling of a House election until 1 July 2022; meaning that a House election can theoretically be held at the latest by 3 September 2022 once events such as issuing writs and campaigning are taken into account.
  • On the other hand, a government must call a half-Senate election by mid-May 2022 (to allow time for completing the count).

This discrepancy between the terms of the two chambers allows a government the discretion to split the terms, if they wanted to. But would they?

History of vote differences in the Senate versus the House

To understand why a government may not want to split the House/Senate elections in an attempt to save its own skin, let’s have a look at the history of how a government’s vote in the House tends to compare to the Senate. Data below only includes elections since proportional representation (single-transferable vote) was introduced to the Senate:

(if you’re on a mobile device, scroll right for full data or turn your device landscape. Click the Previous and Next buttons to view all data.)

ElectionSplitGovt primary, HouseGovt vote, SenateHouse-Senate gap
1949No4644.9-1.1
1951No50.349.7-0.6
1954Yes47.644.4-3.2
1955No47.748.71
1958No46.645.2-1.4
1961No42.142.10
1963Yes4644.7-1.3
1966Yes5042.8-7.2
1969Yes4738.2-8.8
1972Yes41.5N/AN/A
1974No49.347.3-2
1975No42.840.9-1.9
1977No48.145.6-2.5
1980No46.443.6-2.8
1983No43.640-3.6
1984No47.642.2-5.4
1987No45.942.8-3.1
1990No39.438.4-1
1993No44.943.5-1.4
1996No38.836.2-2.6
1998No39.537.7-1.8
2001No42.941.8-1.1
2004No46.745.2-1.5
2007No42.140-2.1
2010No3835.1-2.9
2013No33.430.1-3.3
2016No4235.2-6.8
2019No41.438-3.4
Average, no split election


Average, split House/Senate elections
-2.2


-5.2
1949 Senate election not counted as a "split" due to the joint half-Senate election held in 1949. 1972 was excluded from the average as no corresponding half-Senate election was held (double-dissolution was held instead in 1974). 1975 was excluded due to The Dismissal which meant that who voters saw as "the government" may not be as clear-cut.

The first thing to notice is that governments have historically performed worse in the Senate than they have in the House, by a little over 2%. This isn’t limited to the government; the opposition of the day has also tended to do worse in the Senate to the benefit of the cross-bench at joint House/Senate elections:

(if you’re on a mobile device, scroll right for full data or turn your device landscape. Click the Previous and Next buttons to view all data.)

ElectionSplitOpp primary, HouseOpp vote, SenateHouse-Senate gap
1949No50.350.40.1
1951No47.645.9-1.7
1954Yes5050.60.6
1955No44.740.6-4.1
1958No42.842.80
1961No47.944.7-3.2
1963Yes45.545.70.2
1966Yes40455
1969Yes43.342.2-1.1
1972Yes49.6N/AN/A
1974No45.743.9-1.8
1975No53.151.7-1.4
1977No39.736.8-2.9
1980No45.242.3-2.9
1983No49.545.5-4
1984No4539.6-5.4
1987No45.942-3.9
1990No43.541.9-1.6
1993No44.343.1-1.2
1996No47.344-3.3
1998No40.137.3-2.8
2001No37.834.3-3.5
2004No37.635-2.6
2007No43.440.3-3.1
2010No43.338.8-4.5
2013No45.637.7-7.9
2016No34.729.8-4.9
2019No33.328.8-4.5
Average, no split election


Average, split House/Senate elections
-3.1


+1.2
1949 Senate election not counted as a "split" due to the joint half-Senate election held in 1949. 1972 was excluded from the average as no corresponding half-Senate election was held (double-dissolution was held instead in 1974). 1975 was excluded due to The Dismissal which meant that who voters saw as "the government" may not be as clear-cut.

However, as the above tables show, these effects differ by quite a bit when a split House/Senate election is called. On the primary vote, a government does about 3% worse in the Senate at split elections; while oppositions have managed to turn a -3.1% under-performance in the Senate to a +1.2% over-performance at the early half-Senate election. Bear in mind that these figures compare the Senate vote to the House vote; if the government’s figures in the House slump further as a result of the half-Senate election, it is likely to do even worse in the Senate.

With just four House elections with a split half-Senate counterpart, it’s worth noting that the sample size is very small on this (the difference for both govt and opposition vote is indeed statistically significant at p = 0.012 and p = 0.004; however I’m not convinced that statistical significance is a good metric here). Despite that, no government would look at the history of split elections and come away thinking that this is a good way to rescue itself from poor polling. Any split election would likely see the government under-perform in the Senate and possibly hand political opponents control of the Senate.

For reference, if the average effects above were replicated alongside today’s House polling, there would be an approximately 8% collapse (-5% under-performance of 35% House vote) in the Coalition Senate vote, the Labor Senate vote would jump to about 38% (1% over-performance of 37% House vote), its highest since 2007, and the cross-bench would combine for 32% of the vote (versus 33.2% in 2019). Such a situation would almost certainly see Labor winning 3 Senators in at least one state, paving the way for a possible Labor/Greens joint Senate majority.

What if the government tried to sacrifice the Senate to buy more time to recover in the House?

Superficially, it may look like there is something to this. Keeping in mind the small sample size mentioned above, the two-party-preferred (2pp) swing in the House for governments is about a point better for governments who went to split House elections:

(if you’re on a mobile device, scroll right for full data or turn your device landscape. Click the Previous and Next buttons to view all data.)

ElectionSplitGovtGovt 2ppGovt 2pp swing
1949NoALP49-5.1
1951NoL/CC50.7-0.3
1954YesL/CC49.3-1.4
1955NoL/CC54.24.9
1958NoL/CC54.1-0.1
1961NoL/CC49.5-4.6
1963YesL/CC52.63.1
1966YesL/CC56.94.3
1969YesL/CC49.8-7.1
1972YesL/CC47.3-2.5
1974NoALP51.7-1
1975NoALP44.3-7.4
1977NoL/NC54.6-1.1
1980NoL/NC50.4-4.2
1983NoL/NC46.77-3.6
1984NoALP51.77-1.5
1987NoALP50.83-0.9
1990NoALP49.9-0.9
1993NoALP51.441.5
1996NoALP46.37-5.1
1998NoL/NC49.02-4.6
2001NoL/NC50.951.9
2004NoL/NC52.741.8
2007NoL/NC47.3-5.4
2010NoALP50.12-2.6
2013NoALP46.51-3.6
2016NoL/NC50.36-3.1
2019NoL/NC51.531.2
Average, no split election


Average, split House/Senate elections
-1.7


-0.7
1949 Senate election not counted as a "split" due to the joint half-Senate election held in 1949. 1975 was excluded due to The Dismissal which meant that who voters saw as "the government" may not be as clear-cut. Difference is not statistically significant, p = 0.41 (Student's t-test).

It should be noted that the finding here can’t even clear the (relatively low) bar of achieving classical statistical significance (threshold p < 0.05, here the difference is p = 0.41). In other words, there is a pretty good chance that this difference is simply a result of random chance.

Any government which attempted this strategy would essentially expose itself to a worse Senate position (and possibly a hostile Senate majority) in exchange for the < 50% chance than it improves their odds in the House. Furthermore, unlike 3 of the 5 governments who went to split elections, such a government would have to actively break the House/Senate joint election system. It would have to pass a budget including the cost of an additional election in full view of the media and an opposition (presumably) eager to capitalise on the idea of the government wasting money to keep itself alive, all for the historically-less-likely-than-not chance of a relatively small improvement on its 2pp.

(it’s also worth noting that even if the 1% improvement on 2pp came true, that alone would not be able to save a government down 46% – 47% in 2pp estimates right now)

All of this analysis so far however concedes the argument that the government is staring down the barrel of a historic defeat, and is so desperate as to grasp at election-splitting straws. Is this even the case? How predictive is today’s polling anyway?

Governments tend to recover from poor polling anyway

Our full analysis on why polls taken more than a month out from an election is available below:

But just to summarise the key points:

  1. From about a month to a year out from the election, governments tend to do worse in the polls than they end up doing on election day (or in other words, governments tend to recover from their bad polling by election day).
  2. The side that’s ahead in the 2pp – whether that be government or opposition – tends to see its leads decline (or reverse) by election day (even adjusting for 1).
  3. The amount by which the leading party’s 2pp declines is dependent on how large its lead is; so a party ahead by 51-49 tends to see less decline than a party ahead by 56-44.
  4. Polls at this point tend to much further off the final result as compared to final election polling. In fact for federal eelctions the polls generally don’t start to outperform the simple rule of “assume election is 50-50” until the final weeks of the campaign.

Let’s assume that people forget about point 4 (because politically-engaged people do tend to forget about point 4), and the political advisers and campaign staff around the Prime Minister builds him a model predicting the government’s vote based on today’s polling, correcting for findings 1 – 3.

(point 1 at least is pretty easy to find in the data; Dr Bonham found a similar effect when looking solely at Newspolls)

Given today’s polling, what would such a model forecast for the government’s 2pp, at an election held in May?

Histogram of 2pp results, predicted 25/Jan/2022 for an election held on 21/May/2022.
2-party-preferred estimated from Poll Bludger’s BludgerTrack 2022 first-preference vote data, before being adjusted for the late-term dip and lead-decline effects. Do note that this distribution will look very different for polls taken right before Election Day.

After adjusting for effects such as the government’s tendency to under-perform in polls taken too far out, the polling-only model predicts an average Coalition 2pp of 49.4%: incorporating a classical margin-of-error (or 95% CI), it ranges anywhere from 42% to 57%. Do note that this margin of error is for a poll taken ~4 months out from the election and will narrow significantly as the election approaches.

I don’t have a seat model yet, but borrowing Dr Bonham’s work on conditional probability models, let’s assume Labor’s in line for a majority at 51.3% of the 2pp (or 48.7% Coalition 2pp) and the Coalition needs a 2pp of at least 50.7% to be even-odds for a majority. The former (ALP majority > 50%) is highlighted in light red in the above graph, while the latter (L/NC majority > 50%) is highlighted in light blue, with “hung parliament more likely than not” in purple.

Now, I don’t know about you, but to me this doesn’t exactly show a government staring down the barrel of defeat. And the numbers bear this out – as much as I dislike calculating “% chance of winning” type figures, based solely off current polling, the chance that the ALP wins more than 51.3% of the 2pp is about 40% (or 2 in 5). The chance that the Coalition wins over 50.7% of the 2pp? A little over 30%, or about a 1-in-3 chance.

Just for good measure though, how would that graph shift if the House election was called at its latest possible date (3 September 2022)?

Histogram of 2pp results, predicted 25/Jan/2022 for an election held on 3/Sep/2022.
Truly, such fractional movements in the mean of a distribution with margin of error approaching +/- 7% are what Prime Ministers defy convention and risk Senates for.

The shift in the average expected 2pp is a mere 0.4% (from Coalition 49.4% to Coalition 49.8%). Such a “benefit” to the government is easily drowned out by factors such as polling error, flaws in the model or simple random events.

(If you’re surprised that the change is so small, I’d highly recommend having a look at the graphs in Rebekah’s piece, and note how many of the effects mentioned tend to level off after some point. There isn’t a huge difference in, say, the amount by which governments tend to recover in polls taken 120 days out versus 240 days out.)

Due to the historic volatility of the polls, as well as the tendency for voting-intention to move back towards the government, the polls aren’t predicting a landslide defeat for the government right now. Unless these figures hold up for a few more months (by which time the government will either have or have not budgeted for a separate half-Senate election anyway), the history of polling movement strongly suggests that the election is still a tossup, with Labor perhaps having the very slightest of edges.

(Other data e.g. approval ratings, economic performance etc may suggest a different outcome; I’m opting to focus on voting-intention polling because that’s what tends to be brought up in such discussions)

Furthermore, even if a government was so desperate, splitting the House and Senate elections has historically resulted in disaster in the Senate, with disgruntled-but-unwilling-to-change-government voters given carte blanche to express discontent without risking a change of government (i.e. the by-election effect). Oppositions tend to over-perform their House vote in split half-Senate elections, which runs the real risk of turning into a Labor/Greens joint Senate majority for a Coalition government .

Finally, even if a government was willing to sacrifice the Senate to hold onto the keys to the Lodge, the impact of such a decision is fairly minimal and highly variable. Unless something massive* changes in the coming months, it would be against the government’s interest to attempt to split the House and half-Senate elections; speculation that this may happen should not be given the benefit of the doubt unless new evidence emerges of the government considering such a move.

*: And by “massive”, I mean something along the lines of “war breaks out in the Asia-Pacific region”, not whatever might have any particular social media circle go off in ALL CAPS on any particular day.


2 thoughts on Why a House/Senate Split Election Is Bad for The Government

    1. One more example I think is relevant (but didn’t fit well into the piece) is the WA special election.

      For readers who may not be aware of the history – in 2013, 1375 ballot papers for the Senate were lost in WA, which combined with the relatively small margin by which the last seat was decided meant that the High Court ordered a new election be held. In this “split” election, both the government and the opposition took a pummeling (figures are WA 2013 -> WA 2014):

      -Liberals: 39.2 -> 34.1 (-5.1)
      -Nationals: 5.1 -> 3.0 (-2.1)
      -Labor: 26.6 -> 21.5 (-5.1)
      -Greens: 9.5 -> 15.6 (+6.1)
      -Palmer: 5.0 -> 12.3 (+7.3)

      Any form of by-election is usually bad for the government of the day, as the by-election voters (usually) don’t decide whether the government remains in power and hence dissatisfied voters are free to express their discontent without worrying about the alternative. In fact, this is something replicated overseas.

      For example, in Singapore, the opposition parties used to strategise which seats they contested so that a majority of government MPs would be returned unopposed. The theory (developed after noticing opposition candidates did better in by-elections) was that the voters in the seats they did contest could safely vote for the opposition candidate without fear of removing the government.

Add Your Comment

× Home Politics Economics Projects
×

You appear to be using an outdated browser, for which this site is not optimised.

For your security, we strongly recommend you download a newer browser.

Although any of the latest browsers will do, we suggest the latest version of Firefox.

Download Firefox here

(Click on the button in the top-right to close this reminder)