WONK ALERT (Category 4): This post is very wonky and statistical and contains 15 charts; I can't even remember if I included any jokes at all!
1. This post presents extensive vote-share data for the ALP, Liberals, Greens and collective "Rest" since the 1989 election, including breakdowns of swing patterns.
2. At least in the case of the Greens and Liberals, there is evidence that swings are not uniform. For these parties, swings in either direction tend to be greatest both in electorates where they perform well, and where their vote share in a given electorate at the last election was high. For these parties it may be reasonable to make some adjustments when predicting vote share in given electorates.
3. For the Labor Party there is less variation in average results between electorates and it is not clear whether there is a reliable pattern. For the Labor Party it may be easiest to just assume uniform swing except if there is convincing polling evidence otherwise.
4. A major issue in predicting the Green vote at the 2014 election will be whether or not there is a Wilkie-style independent standing.
This post is a review of voting results in Tasmanian Lower House elections since 1989. The main inspiration behind this post is that since EMRS have stopped publishing seat-by-seat breakdowns, there is a lot of data suggesting that, unless things change dramatically, there will be a large swing at the next election to the Liberals and away from both Labor and the Greens. But in trying to predict the outcome of this swing with only a statewide breakdown of the vote to go on, the question arises: should I assume that the swing will be the same in every electorate? Or is there reason to assume some parties in some seats will swing more than others? By answering this question it may be possible to come up with a "drop-in" model for which I can take a given EMRS poll and the results of the last state election, and get a projected seat distribution.
I have chosen 1989 as the starting point because that is the beginning of an era in which there were three significant forces in all five Tasmanian seats - Labor, Liberal and the Greens (or initially, the Green Independents.) Green independent candidates and Australian Democrats had some successes prior to 1989 but never polled significantly in all five electorates at the same election. The 1989 election saw the Liberals lose their majority by one seat despite polling nearly 47% of the vote, and the question of majority government has hung over every Tasmanian state election since.
During this time fourth-force candidates have had relatively little impact. Three exceptions have been the victory by Independent Bruce Goodluck in Franklin in 1996 (Goodluck served a single term then retired for health reasons), the very narrow defeat of Independent Andrew Wilkie in Denison in 2010 (since elected to the same seat federally), and the relatively strong showing of the Tasmania First party in 1998, especially in Lyons.
This graph shows the changes in support for the three parties and the combined "others" in this time.
From 1992 to 2002 the Labor Party gained 23% over the course of three elections while the Liberals lost 27%. This appears to be in the process of reversing; recent polling has been producing very 1992-like results. Until a few years ago we could say (as Antony Green did) that swings like this just don't happen in other states. But now in Queensland the LNP (ex-Coalition) has gained 24 points in 11 years with Labor losing over 20 points in the same time. That is probably exceptional; in Tasmania there is reason to believe these kinds of swings are now systematic. The Hare-Clark system delivers proportional representation, but not proportional power - and every time the Greens win the balance of power they are as powerful as the other two parties despite not having nearly as many seats. Perhaps even more powerful (see The Compliant Coalitionists). Voters who do not like the Greens having the balance of power therefore "game the system" by voting for whichever major party looks like it might win a majority.
This is what the pattern looks like on an electorate by electorate basis:
A similar pattern but leaning Liberal compared to the state as a whole and with a slightly lower Green vote. The spike in "rest" for 1998 is mostly Tas First.
Braddon used to be extremely pro-Liberal, especially in 1992 when the Liberals beat Labor by the small matter of 45 points. However in the last few elections Labor has polled at about its state average in the electorate. The Green vote in Braddon has always been low. A relatively high figure for others happened in 1996 when the National Party (here not treated as part of the Coalition as preferences are not directed under Hare-Clark) polled 4.4%.
Denison is the Liberals' worst electorate and the Greens' best; in 2002 the Greens even outpolled the Liberals there. The independent candidacy of Andrew Wilkie in 2010 prevented the Greens from gaining the swing to them that occurred in the other seats.
Franklin has always had a high Green vote and in 2010, boosted by the popularity of Greens Leader Nick McKim (and also demographic change) the seat recorded a higher Green vote than Denison, and in fact the Greens' highest result in any electorate. (Had Andrew Wilkie not contested the latter the Denison vote would probably still have been slightly higher than the Franklin vote.) The electorate used to lean to Labor compared with the Liberals but this is no longer true, possibly as a result of the success of Opposition Leader Will Hodgman; it now leans the other way. In 1996 former federal Liberal MHR Bruce Goodluck was elected as an independent. There was also a rather high Democrat vote that year.
Lyons has gone the other way to Franklin. It was once a very strong seat for the Liberals as a result of the stupendously high personal vote for Robin Gray. Even the turmoil of 1989-1992, in which Gray lost the premiership and the leadership and was the subject of adverse findings during the Rouse bribery scandal, barely scratched the man. Since Gray left the electorate has returned average to slightly below average results for the Liberals compared to Labor. In 1998 the Tasmania First party polled 9.9% but failed to win a seat.
I now move on to electorate by electorate swings by party. These graphs show the change for each party from the previous election, at each election starting with 1992. Firstly, Labor:
First one who says that is a W for Will is a rotten egg! In 1998 the Liberals recorded strong results compared to 1996 in the two southern seats; in the case of Franklin, the return of votes lost to Bruce Goodluck in 1996 would have been a factor. However, in 2002 the impact of the Bob Cheek leadership and the disunity it caused was especially felt in these seats, while in 2010 some votes were lost to Andrew Wilkie.
Again, 1=1992, 2=1996, 3=1998, 4=2002, 5=2006, 6=2010
The Greens started at a high level and have now reached a record high. In the intervening time there have been two large swings to them and four small to moderate swings against. Generally, Denison (where the party's vote is highest) has been the most volatile electorate and Braddon (where it is lowest) has been the least. However in 2010 the party's vote in Denison was suppressed by Andrew Wilkie.
In trying to predict swings at future elections (using the state swing as a base), these are some of the questions that can be asked:
1. Are there electorates where each party has tended to have a vote that swings more or less than the state average?
2. When a party's vote in an electorate is high, is that vote likely to swing more or less than the state average?
3. When a party's swing in an electorate is higher than the state average, is that party's swing in the same electorate likely to be higher or lower next time?
I'll give the answer to 3. first: apparently neither. If there is a relationship at all, the size of it is so tiny (1-3% of variation explained depending on party) that I won't be bothering with it further.
Moving on to 2, these are the relationships by party for the primary vote compared to the amount by which the seat swing exceeds (or is less than) the state swing at the next election. Firstly, Labor:
A very weak relationship there; in fact, not significantly correlated. Moving right along: Liberal:
It's not a particularly strong relationship, but it's reasonable, with nearly 20% of variation explained. If the Liberal vote in an electorate is high, the swing at the next election will exceed the state average; and if it is lower, it will be smaller.
And finally, the Greens:
In this case the Denison Wilkie-related outlier is so extreme that I've booted it as a data point (it's marked X on the graph). For the rest, the relationship is modest, with 13% explained.
So for the Greens and Liberals there is some evidence that their vote swings more from high bases than from low bases. The relationships above, adjusted to consider the strength of each party's performance in 2010, are not strong, but do provide reason to expect that:
* All other things being equal, the Green vote would be expected to swing 1 point more than the state average in Franklin and 1 point less in Braddon.
* All other things being equal, the Liberal vote would be expected to swing 1 point more than the state average in Braddon and 1 point less in Denison.
However, in the first case, a great deal depends on whether or not there is a Wilkie-like independent contesting Denison. If there is not, then the Greens would be expected to recover the impact of Wilkie (estimated at over 4 points), meaning that a swing against them in Denison would be 4 points less than the state average. All the other swings would need to be adjusted for this to produce a zero-sum outcome (hence Denison swing -4 compared to state average, Braddon 0, Lyons and Bass +1, Franklin +2). So for a five-point statewide swing against the Greens, with no Wilkie-type candidate in Denison, it would be expected that Franklin would swing against the Greens by 7 points (down to 20) but Denison by only 1 point (to 23).
We can also look at the past relationship by party for different electorates. Firstly, Labor:
The Labor vote has been bounciest in their second-worst seat, Braddon, and least bouncy in their second-best seat, Denison. But the relationship between seat quality and how much the seat swings is not consistent and it is also notable that the party's average vote share doesn't vary much across the five seats.
Secondly, the Liberals:
The Liberal vote shows a clearer relationship - bounciest in their best seat, Braddon, and least so in their worst two seats, Denison and Franklin.
And finally, the Greens:
The Greens' vote also swings the most in their best electorate (I've excluded the 2010 Wilkie-affected result again) and swings the least in their worst.
In either a future post or an update to this one I will have a go at seeing whether the limited and mostly dated electorate sampling available thus far provides any useful insight into electorate swings. Given the huge amount of bouncing between 200-vote samples, and the problems with data from some pollsters, I'm not overjoyed about the prospects there, but if there are strong regional differences they may pop up.
(Update: And they did! The above task is conducted in Uneven Swing To Liberals in Tasmanian State Election Polling.)