Tuesday, September 15, 2020

Age And Canberra Are Still Killing State Governments

 Advance Summary

1. One of the most important factors in state election outcomes is the influence of whether the governing party at state level is in Government or Opposition federally.  To be the same party as the federal Government is a disadvantage.

2. Another important factor is the age of the state government, with governments tending to do worse the longer they have been in office.  

3. In the last six years, all same-party state governments that have faced elections have lost seats in significant numbers.

4. In the meantime, two of the three opposite-party governments gained seats (though one very old opposite-party government was defeated, but with a 2PP swing to it.)

4. It might seem logical that if the federal government at one state election was the same as at the previous election for that state, then this factor would not generate further swings against an incumbent state government of the same party, or further protection for one of the opposite party.

5. However, the evidence suggests otherwise.  It appears that more voters continue turning against state governments that are of the same party as the federal government, over successive elections, even when the same federal government had been in office at the state election before.



---------------------------------------------------------

During the most boring phase of the COVID political shutdown I started revisiting some old themes that had been covered on this page.  This article (rising to about 4/5 on the Wonk Factor scale; beware) is a sequel to what I consider to be one of the more important articles on this site, What Kills State Governments: Age Or Canberra? (2014)  The theme of that article is the massive impact of which party is in power federally on state election results, an impact that has increased in the last few decades.  State governments are disadvantaged when the same party is in power federally, and usually lose seats in this situation.

At the time of writing that article, only six state governments had lost office while their party was in Opposition federally since 1969.  However, state governments of the same party as was in power federally had had a little better than even strike rate.  The pattern had become more acute after 1990, with same-party governments retaining office just 35% of the time, and opposite-party governments retaining in 89% of cases.  I found that both whether a state government was in power federally, and how old the state government was, were useful in predicting changes in seat share for that government, and that the two worked much more powerfully in combination than either did alone.  I call the former factor federal drag.  It results at least partly from voters using state elections to give federal governments of the same party as their state government a kicking, but probably also partly from voters feeling that state governments of the same party as the federal government will be too compliant with Canberra's wishes.  I should stress that I did not discover it; my inspiration was this Antony Green blog post.

I wanted to revisit this theme in the leadup to the Queensland state election, especially because of the extreme shortage of good polling for that election so far.  (Yes there have been at least four Queensland polls released lately, but they are generally internal polls, weird commissioned polls, by irregular pollsters, etc - I'd wait for some mainstream statewide polling first.)  Especially, I wanted to take a look under the hood of the federal drag phenomenon, and ask this question: does federal drag cause governments to lose seats even when it appeared in the same way at the previous state election? 

A good example is the NSW 2019 election: the Liberals were affected by federal drag at that election, but they were also affected in the same way in 2015.  So should it have still caused a swing against them, or would the only swing they should have expected come from the increasing age of their government?  And this also applies to Queensland: Labor is in opposition federally, but it was also in opposition federally in 2015 and 2017.  The Palaszczuk government sits on an uncomfortably small seat margin, so does it get a further dose of help from Canberra, or was that all baked into the massive swing to it in 2015, and no longer useful, leaving it on track to lose more seats because it has been there for longer?

Results Since The 2014 Article

Firstly, just a quick roundup of how the theme of the article has fared since it was written.  Since the 2014 election, there have been nine state elections (two each in Victoria, Queensland and NSW, and one each in WA, SA and Tasmania).  As in the previous article, I use seat share change (measured as the change in the percentage of seats held by the governing party, eg 50% of seats at one election to 55% at the next is a 5% gain) as a major measure of state government fortunes for statistical purposes.

Coalition governments went to the polls six times since mid-2014.  They lost office in Victoria 2014, Queensland 2015 and WA 2017.  They retained government in NSW 2015, NSW 2019 and Tasmania 2018.  In all six of these elections there was a swing against them in seat share terms of at least 6%, with exceptionally severe seat share swings in Queensland 2015 (38%) and WA 2017 (34% of the parliament), the second and fourth largest seat share losses by incumbent governments since 1969.  In NSW, the Coalition was able to absorb successive losses of 16% and 6.5% of seats because of the huge size of their 2011 victory.

Labor governments went to the polls three times.  The first-term Andrews government in Victoria 2018 was returned, increasing its share of the parliament by 9% (though this understates what a massive win it was in 2PP terms).  The first-term Palaszczuk government in Queensland 2017 was returned, increasing its share of the parliament, albeit only by 2% and with no two-party swing to speak of.  The Weatherill government in South Australia 2018 was finally defeated (with a 2PP swing to it which was negated by a major redistribution), but had been in office for 16 years and was therefore expected to lose according to the regressions in the 2014 article.  

The massive seat swing against the Newman government was an outlier because that government was so young; indeed the historic relationship at the time of the article suggested there was practically no chance it would lose.  But it did.  As a result of this, the relationship between age and seat share change has weakened, and age of government now explains only 21% of variation in seat share change since 1990 (down from 37%).  But the explanative power of "federal drag", and the average difference between governments benefiting from it and those not, have both slightly increased.  For state elections after 1989, "federal drag" explains 34% of variation in win-loss results, and governments with it on their side now do 16.7 points better in seat share change terms than those without.  To be specific, the average result for a party that goes to a state election while in power federally is a loss of seats equal to 15.4% of the parliament, while the average result for a party in opposition federally is a gain of 1.3% (the medians are 12.3% loss and no change, respectively.)

For state elections after 1989, the following is now the linear regression for these two terms:

Expected seat share change = .1577*(party in Opposition federally) - .0154*(age of government) -0.0418

(51% of variation explained. "party in Opposition federally" equals 1 if it is, and 0 if it isn't.)

According to this regression, a 3-year old government of the party in power federally can expect to lose about 9% of its seats at an election.  A government that is in opposition federally is expected to gain about 7% of seats at the same age, isn't predicted to lose seats at all unless it is at least 7.5 years old, and takes 13 years to be expected to do as badly in seat share change terms as the former.  

I recently also found that federal drag applies to the Northern Territory - the CLP does much better when Labor governs federally.  In the ACT with its proportional representation system, however, it has so far applied in seat terms only to Labor, and not to the Liberals.  

What Drives Federal Drag?

I wondered if the federal drag equation above was giving the full story, for the following reason.  Many of the worst results for state governments happen at elections where the same party has moved from opposition to government in Canberra since the last state election.  27% of state elections since 1969 have been of this kind, yet this category includes four of the worst five seat share loss results (Queensland 2015, NSW 2011, WA 2017 and SA 1997, though in the last listed case the government survived in minority).  

Maybe then it's the shifting of the direction of the drag that drives the federal drag effect?  So to look at this, here's a contingency table (click for clearer version):

The columns are as follows:

Federal: whether the state government is in government or opposition federally

Last: whether the party of the state government was in government or opposition federally at the last election

Change: the direction of the change in whether the party of the state government is in opposition federally or not.  (In theory, moving into opposition is an advantage, moving into government a disadvantage)

There are two results sections, one for all state elections from 1969, one from 1990.  For these:

N - number of elections

Age - average age of the state government

Wins, Lost - self-explanatory, except that Queensland 1995 counts as half a win and half a loss.

% - percentage of elections won

Ave - average change in seat share as a proportion of parliament (eg -0.1 = average loss of seats totalling 10% of the parliament)

So, what do we see here?  Well, not what I was hinting that we might!  The Gov/Gov section (in which we find NSW 2019 for example) has the same average seat loss as the Gov/Opp section (eg WA 2017), and a worse winning percentage (though not significantly worse).  The governments in the Gov/Gov section are slightly older on average, but not much.  Likewise, while it might seem that the Opp/Gov governments would have a bigger advantage than the Opp/Opp governments, if there is anything it's not enormous; the latter still nearly always win, and on average do not lose seats.  (A side-note: there are very few Opp/Gov governments because by the time a federal government loses office, it has usually also lost office in most of the states, as a result of federal drag effects.)

Rather, there is a major (and statistically significant) difference between the Opp/Opp governments and the Gov/Gov governments, even though in both those cases there has been no change in the direction of federal drag since the previous election.  State governments that are in power federally, and where the same party was also in power federally at the previous election, have a track record of losing slightly more often than not, and lose seats just as readily as if they had not been in power federally at the previous election.  State governments that were in opposition federally at both the current and the previous election, however, on average don't lose seats (despite having a few more years under their belt) and are very rarely beaten - especially since 1990.

What does it mean?

These results are surprising, but they are also stark.  The best model to explain them could be that at every state election where the defending government is of the same party as is in power federally, some portion of voters who previously supported that state government decide to shift their vote.  At each subsequent election, more do so, until either the state government loses or the federal government does.  The other side of the issue is that being in opposition federally may make voters who would otherwise be increasingly dissatisfied with their state governments more forgiving.  Hence the pattern in the Howard years of opposing Labor state governments repeatedly winning big.  

All this is "on average", of course.  The average 2% seat share gain for Opp/Opp governments since 1990 (exactly what the Palaszczuk government got in 2017) comes with a +/- of about 10%.  Losses of seat share that would be enough to put said government into minority have happened at over 40% of the Opp/Opp elections, and losses that would be enough to cause it to lose have happened about one-sixth of the time.  Every election is different and special factors such as COVID-19 responses, current economic circumstances or government scandals could well trump these historical patterns.  But, for what it's worth, unlike another historic pattern of questionable current relevance, federal drag is still on Queensland Labor's side, just as it was in 2017.

6 comments:

  1. The swing to Labor after the ousting of Newman had very little to do with federal drag but all to do with just how terrible the Newman government was. Just my 2c's worth.

    ReplyDelete
    Replies
    1. I think it had a lot to do with both how terrible the Newman government was and also how terrible the previous Labor state government had become by its end (which inflated the margin the LNP was sitting on). But I also think federal factors had a strong influence. In particular, the Queensland election came at a time when the Abbott government was polling badly, and there is a correlation between poor federal government polling and poor results for state governments of the same party (as graphed in the original article). Furthermore, the 2015 Queensland election day came three days after Tony Abbott's widely ridiculed decision to knight Prince Philip.

      Delete
    2. Given the closeness of the result, adding a knighthood to Prince Philip`s collection of knighthoods may have been the clincher.

      Previous state Coalition/Liberal state governments, starting with Greiner but including Borbage and Newman, not reintroducing knighthoods at state level and Howard not doing so at Commonwealth level did not help the popularity of knighthoods either.

      Delete
    3. The actual seat results for the ALP in East Coast state elections since your 2014 article have been slightly muted, compared with their 2PP seat results, mainly because of the Greens taking or holding seats with ALP 2PP margins. Melbourne 2014 (Vic) (Prahran was a narrow Liberal 2PP win), Newtown and Balina gain and Balmain hold (a 2011 gain) 2015 (NSW), Maiwar gain 2017 (Qld) and Melbourne no change Prahran Green hold ALP 2PP gain and Brunswick Green from ALP 2018.

      Ex-National independent Russel Northe held Morwell in 2018 but the ALP won 2PP.

      Queensland will be one to watch on this front next month.

      Delete
    4. Yes. There's no perfect measure of how close or not close an election is; I used seat share because it was the metric from the original article and because it is the primary determinant of at least forming majority government. Presumably federal drag affects a government's competitiveness against third parties as well. NSW 2019 with the Coalition losing four seats to Shooters and an indie since the previous election is another example.

      Delete
    5. The Greens have been able to take seats of the ALP at state level with the Commonwealth ALP both in government and opposition. However only from the Coalition at state level under a Coalition Commonwealth Government and usually Coalition state government. The Greens have a greater tendency to win vacant and/or significantly redistributed seats off the ALP than seats with a sitting ALP member (although choice of candidate in Richmond (Vic) 2010, 2014 and 2018 may not have helped on this front).

      There is some possibility that Commonwealth drag hurt the then Government supporting (on supply and confidence) Greens in Marrickville 2011 and the 2012 Melbourne by-election.

      Delete