Making Seats "Marginal" At By-Elections Is Meaningless

Last night saw the Labor government get the good end of the stick in the Dunkley by-election, easily retaining a seat that was precariously above the long-term average swing for government vacancy by-elections.  It's no disaster for the Liberals who have got a modest swing with some mitigating factors but they (especially Jane Hume) were out in force last night spinning the outcome as a triumph.  Together with the usual nonsense about first-term governments not in recent decades losing seats and governments not losing by-elections caused by deaths (both based on trivially small sample sizes) I heard a lot about how they had turned Dunkley marginal and they were coming for the seat.

Marginal seat status where a seat is retained is determined by general election results not by-elections (so Dunkley is no more a marginal seat than it was before), but this made me wonder, does getting a seat inside the marginal range at a by-election predict anything at all?  I've found that such seats have historically almost always been retained by the government at the next election, although on average the election-to-election swing has been worse than the national average in such cases.  The idea that the Liberals have put Dunkley in serious danger next time with a swing that is not even bog-average for a government vacancy by-election has no basis.  

I've previously written about the impact of a change of mid-term ownership caused by a by-election.  Where there is a change of party occupancy in the middle of the term, the new party's MP can build a personal vote at the expense of the previous party.  It is also more likely in such cases that there is some lasting local issue driving the result, or that the government is losing anyway. 

But when an opposition merely makes a seat closer but the governing party carries on, does that do anything? I looked through all previous federal by-elections I could find since 1910 that fit the script.  (I do not use pre-1910 as the 2PP concept does not make sense with three main parties.)  Here they all are:

Prev - previous election 2PP

GB Swing - swing at by-election

B/E - the 2PP of the by-election

Next - the 2PP for the government at the next election

BN Swing - swing from the by-election to the next election

GN Swing - swing between the two elections surrounding the by-election (adjusted for redistribution where required)

Nat Swing - the national 2PP swing at the next election

Overall, I found 13 by-elections since 1910 where the 2PP margin has moved from outside the marginal zone (56+ 2PP) into the marginal zone (50-56) without the seat falling.  Asterisks are noted for Bennelong (recontesting incumbent), Grampians (incumbent defected to the proto-Country/Nationals Victorian Farmers Union between the by-election and the next election) and I'll come to Swan in a moment.

In every case except Swan and Grampians the government retained the seat at the next general election.  In nine cases the seat went back outside the marginal zone.  In all cases but four, the next election was less close than the by-election.  

In every case bar the irregular Grampians example there was a swing against the government from the election before the by-election to the election after.  However of the average 4.7% swing, about half (2.4%) was captured by the national uniform 2PP swing.  So on average non-marginals with a marginal by-election result were deflated by a further 2.3%.  This can be attributed partly to the loss of the sitting member's full personal vote (replaced with a part-term vote for a new sitting member), and probably also partly to such by-elections having tended to have slightly larger than average swings for a government vacancy in the first place (7.1%), but there may be something here beyond those two things. If there is, it is doubtful it applies to Dunkley where the by-election swing has been much lower. 

(Incidentally here we see another issue with the Liberals' claims about what would happen if the swing was repeated at the federal election.  By-election swings generally aren't repeated.  The average swing against governments at government vacancy by-elections is about 6%, but the average 2PP swing against governments at elections is 2.3%).

Now to Swan 1940 - the only federal case since 1910 where a seat that moved into the marginal range fell at the next election!  In this case there was a mid-term change of government on the floor of the House and so while the seat was retained by the UAP government at the by-election, by the time of the next election they were no longer the government.  The national swing was such that by uniform swing the seat would have fallen anyway, and the new Prime Minister being Western Australian could well have been the icing on the cake.  So, the exception that proves the rule.

Overall, there is no evidence that Oppositions that make seats marginal tend to make those seats competitive at the next general election - usually, in fact, the margin increases from the by-election.  There is no case where such a seat fell when it would not have fallen anyway.  

What about seats that become more "marginal"?

Dunkley (6.3%) is closer to marginal status than any of the non-marginals I found above and looks set for a closer by-election finish than all bar maybe three of them.  So for completeness I thought I'd look at cases where an already marginal seat became more "marginal" at a by-election.  These are a rare beast - because the average swing for government vacancy by-elections is around 6%, a by-election called below that level is quite often simply lost, and governments tend to avoid calling them if they possibly can.  So I only found six examples, two of them very old.

In three of the cases where a government seat became more marginal at a by-election, the government actually lost that seat at the next election.  But in the case of Bendigo 1915, the government changed from Labor to Nationalist and the national swing was such that the seat would have been lost anyway if one treats the 2PP swing as even meaningful.  In the case of Werriwa 1912, there was an unfavourable redistribution and the new government MP decamped to safer pastures; that seat too would have fallen by uniform swing.

This leaves Flinders (1982) as the one and only case I've found where a government retained in a by-election with a marginal result (whatever the baseline), lost at the next election, but would have just held on by uniform swing.  And this was a case with curious resonance with Dunkley.  Frankston Mayor Rogan Ward was the Labor candidate but the unsatisfactory 2.3% swing against a reeling government on the retirement of 16-year incumbent Phillip Lynch was considered not good enough, and Ward was replaced for the 1983 election by Bob Chynoweth, who briefly dislodged Peter Reith.  And again here in the general election there is a home state effect with Bob Hawke taking over as Prime Minister from Bill Hayden.