Monday, February 12, 2018

Tasmania 2018: More Internal Polling Games

The 2018 Tasmanian state election (see my guide) is three weeks away, but still we have had no neutral public polling in the state for over two months.  In contrast, by this stage in 2014 we had had a recent public EMRS, a recent public ReachTEL, and also a Liberal-commissioned ReachTEL of which detailed results were released.  All we've had this year is a dribble of internal polling snippets and one commissioned seat sample by the Australia Institute.

This makes it rather hard to tell what's going on.  The Hodgman Liberal Government, which appeared to be asleep or in a holding pattern for most of 2017, has been quick out of the blocks with a large number of policy announcements covering most (perhaps even almost all) of the many issues in play at this election.  Insiders claim what they're doing (especially claiming there could be a return to a Labor-Green government) is working.  Could it be a repeat of 2006, when a seemingly ailing government that was widely considered doomed to lose its majority in fact surged through the campaign and ended up winning very easily?  While there are some differences between 2006 and 2018 in polling terms (especially the severe volatility of the late 2005 polls) majority government is a major issue for some voters, and a party that can establish that only it can win in majority has an important advantage.



In the absence of a public poll, we are seeing a high level of reporting of party internal polls in the media.  In general, parties only release internal polling to the extent it suits their purposes, but whether it suits a party to say it is winning easily, winning narrowly, trailing narrowly or losing hopelessly all depends on the scenario.

Before I go any further, a comment about the language journalists use in reporting internal polls parties have given them.  It is common to see party polling described as "leaked".  But a poll isn't leaked if someone on the inside has given it to a reporter with the tacit approval of the party's campaign team or leadership, and if indeed the option of feeding it to the media was part of the plan all along.  Both parties in Tasmania have a long history of releasing internal polling whenever it suits them, and this is several times commoner than unwanted releases of party polling information.  Leaks are the stuff the commissioners of sensitive information don't want the public to know.

In this election, the internal-polling strategy is pretty simple.  Each party wants to release results that "show" that it, and better still only it, can win majority government.   What's been interesting is that so far we have seen just the Liberals supplying quite a lot of detailed data that supposedly shows them either capable of winning a majority or actually set to do so.

We have not seen any such figures from Labor, though they have said (paywalled) their polling shows Rebecca White ahead as Preferred Premier (44.2% to 42.8%), and have also made various claims (not linked to any specific figures) about "strong dissatisfaction with the record of the Hodgman Government".  State secretary Stuart Benson has also claimed that "Labor is very competitive and Labor's response to the health crisis is resonating in the community. Labor can win majority government."

I suspect Labor don't have numbers showing a serious primary vote lead over the government.  Even if Labor has numbers pointing to a clearly hung parliament if the election were held right now, to release such numbers would just invite the Liberals to say that "even Labor's own internal polling says they cannot win a majority".  So it's really just the Liberals' polling we are getting lots of detail from at this stage.

 The other, less likely, possibility is that Labor might be putting data out there but having trouble getting reporters to pick up on it.  I list that one only for completeness.  Print media love free data, however dubious.

MediaReach: An Unknown Player With Some Very Long Polls

As mentioned in my previous article, MediaReach, which the Liberals have been using for regular polling since mid-January, is a pollster of unknown quality in the Tasmanian context.  Poll quality is not just about the questions the poll asks, the length of its interviews and so on, but also about the secret herbs and spices - how well the pollster samples an electorate, and how well they adjust for the fact that response rates to phone polling these days are so low that any raw sample won't be representative.

The length and content of the MediaReach polls has varied slightly, but some of them have been very long indeed.  One sent to me by a reader on 30 January had 22 questions.  After three introductory questions it then moves into:

* Voting intention (Labor/Liberal/Green/One Nation/"Jacqui Lambie's Network"/"some other party"/undecided)

* Likelihood of changing vote

* Whether the voter's (i) local area and (ii) state are moving into the right direction or the wrong direction

* Whether the voter firstly likes the leaders as people and secondly thinks they are good at their job.  The canvassed leaders are Will Hodgman, Rebecca White and Jacqui Lambie.

* Preferred Premier Hodgman vs White

* Whether the Liberal Government is doing a good job of running the state

* The importance (or not) of majority government for Tasmania's future

* Who will win the election (Hodgman Liberal majority/White Labor majority/"some form of minority government")

* Poker machines

* Most important issues for voter and their family from a set list:
- health and hospital funding
- cost of living pressures
- protecting our local identity
- fixing water and sewerage services
- major roads in all regions
- encouraging businesses to grow and employ more people
- tackling illicit drugs
- upgrading infrastructure, services and amenities
- some other issue/undecided

* Most important issues for the state (ditto)
- strong economy and jobs
- stable majority government
- investing in health services
- investing in education services
- funding major infrastructure projects
- keeping Tasmanians safe
- some other issue/undecided

* Regardless of the voter's voting intention and expected result does the Hodgman Government deserves re-election

* Regardless of [everything under the sun] which party is the voter normally closest to?

My source took six minutes and ten seconds to get through all of this, though I've seen comments from one industry professional that anything over two minutes is risky in poll design.  (On the other hand, polls with dozens of questions seem quite common in the US.) The poll also promised "one final question about the state" with eight questions to go, then "almost done" with six left, suggesting some questions may have been added on to the end.

Another oddity is that at the end of the poll it refers to a privacy policy available at MediaReach's website, but I can't actually find any website for MediaReach.  There's "Reach Media" but it doesn't seem to be the same thing.

Some people have tried to call these MediaReach polls push-polling, but they are quite simply not even close to it.  The poll is clearly a case of a party asking genuine questions because it wants to know and track the answers.  The question designs generally aren't skewed and don't seek to train the respondent into responding a certain way (and in most cases aren't capable of doing so), though there's one possible exception which I discuss further below.  The questions don't even engage in any dodgy "message-testing" in which the voter is asked if some questionable statement makes them more likely to vote one way or the other.  Neither of these things would make a poll push-polling anyway (as discussed last election and so many times before.)

The results

While the question designs of these MediaReach efforts appear to have been OK apart from the length issue, the shortage of past benchmarking for this pollster in Australia (let alone Tasmania) means we can have no idea how accurate it is.  It could be correct, it could all be a total load of rubbish.

The most recent reports in the Mercury say the Government polled a 46-29 primary vote lead over Labor on Monday last week and a 47-31 lead on Thursday, with the Greens on 12% and the Lambie Network on 5% across the two nights.  It's a little hard to credit Labor would only have gained 2.7 points from the 2014 election disaster, especially given the drag effect of the none-too-popular Turnbull federal government, so I wonder if this is yet another poll that has the Labor vote too low (an issue for most pollsters who have polled within the state in the last decade).

If the Government beats Labor by 16 points on primaries it will retain its majority easily.  The Liberals failed to win a majority with a 12.2 point lead in 1989 but that was with a rather high Green vote.  Labor won a majority in 1998 with a primary vote lead of just 6.7 points.

Given that the Government has three incumbents to Labor's none in Franklin, it would have some chance on these numbers (if accurate) of retaining its 14th seat in Franklin, but it is very hard to credit that it could win 15 seats on such numbers.  David Beniuk's report says "The party also expects to hang on to three members in Lyons, with Labor to win one, and the fifth to be fought out by the major parties, with high-profile JLN candidate Michael Kent an outside chance."  Four Liberals in Lyons, Rebecca White's home seat and which the Liberals lost in last year's federal election?

(There's a long-term problem with Tasmanian seat samples of a few hundred voters per seat producing nonsense results.  In theory the margin of error for a 50% result and a sample of 300 is 5.7%, which is enough to make nonsense of a lot of Hare-Clark projections by itself, but polls are not true random samples anyway, and more importantly when you do several seat samples at a time there's an increased chance that one of them will be outside the margin of error by chance.)

Last week the Liberals also sought to counter the Rebecca White phenomenon by claiming Will Hodgman had a 57-27 lead as Preferred Premier over White in Braddon.   It is possible that the structure of the MediaReach polls - which asks voters to distinguish between whether they like a leader as a person and whether they are good at their job before asking them which leader they prefer - is producing some skew to Hodgman compared to if voters were just asked straight out who their preferred premier was.  It's also the case that Braddon is the electorate in which one would expect Hodgman to do best.  The Liberals also referred to a 48-40 verdict against Labor's anti-pokies policy, which given that it's Braddon isn't that crash hot for them, but the most important thing with this policy is not what voters think of it generally, but whether it works the same way as forestry in driving the votes of a sector of crucial and swinging voters.  On that, we have no evidence.

Hopefully we will soon get some public polling from a pollster with a public track record so we can get a more reliable idea of how the parties might be travelling!  It's getting tempting to commission some myself!

Update - 13 Feb

Today the Liberals released further figures from their internal polling - a five-sample aggregate since late January that has them on 44.2, Labor 29.9, Green 13.6, JLN 5.8, One Nation 2.1 (not confirmed as running candidates), other 4.4.  On those numbers if accurate the Liberals would be likely to win 13 seats.  The same cautions apply as for the rest of their polling.  Incidentally I believe the start date of the tracking is Jan 30 not Jan 20.

Update - 14 Feb

More of this in The Australian with the Liberals claiming their polling shows the Lambie Network is tanking (paywalled).  The Liberals claim their tracking has JLN down from 15.5% to 6.1% in Braddon, 11.7% to 6.1% in Bass and 12.4% to 5.7% in Lyons.  The Lambie Network's Glynn Williams has objected to the polling "relying on landline calls".  While major pollsters no longer risk landline-only polling, it is not necessarily severely distorting since scaling can be used to get around issues with it to a degree.  Thus the old Newspoll recorded an outstanding result at the 2013 federal election using only landline polling.

Assuming these are the five polls used in the aggregate, it seems the poll includes JLN in the readout for Denison and Franklin (where they are not running) but that the party only averages about 2% in those seats.

Update - 15 Feb

More of this in the Mercury today with Labor claiming White has a 46-41 lead as preferred Premier in their internals but again not releasing voting intention figures.  Meanwhile the Liberals have also released results showing 72% of respondents believe majority government to be important.

Two things might be hypothesised here: Labor aren't releasing their voting intention figures because they are ugly, and the Liberals are continuing to release vast amounts of detail because if they give the media free polling data the media will be discouraged from spending money on its own polls.

8 comments:

  1. Kevin what do you make of the disparity in the amount of advertising between Labor and Liberal? Liberal party ads seem to outnumber Labor by 5 to 1 on TV.

    ReplyDelete
    Replies
    1. I expect Labor's door knocking is outdoing the Liberals by a similar margin. TV advertising is not the force it was because of disaggregated media and home entertainment and a similar barrage didn't save the Reps seats at the federal election. All the same it seems to say something about which side has money to burn, and the question being asked is where some of that is coming from.

      Delete
  2. Observation - this election appears to be a referendum on Pokies and the Libs are suffering massively. In my part of the Lyons electorate, we have seen a Labor Federal Member, and an ALP upper house member get both up against incumbent Liberals in the last 2 trips to the ballot box. The Libs are travelling badly on all issues in this part of the state and should be very grateful that the Hare Clark system buffers them against wipe-outs.

    ReplyDelete
  3. On the Labor Bus David? http://www.themercury.com.au/news/opinion/blair-richards-unloved-seats-ticket-to-ride/news-story/b41e3e12cd78e7a0e758373f869389e2

    ReplyDelete
    Replies
    1. We are on the same bus and have the same name but there are a few of us about! He's the good looking one.

      Delete
  4. Alas...I have been in Tasmania for almost 50 years now and ...the tenor of debate and political comments has not improved...in fact, it seems less. More than ever the parties appear to worry about their personal futures and how they will be re-elected or elected. Those who run for offices and have the best chance seem to be stacked by those who already are working in some pollies' office with the result of, The same old things!. I have even quit watching Insiders because of the numbing sameness of it all. We need some brave leaders...preferably, females.

    ReplyDelete
  5. Kevin
    In the absence of EMRS polls I had a go at trying to interpret what the Sportsbet.com markets are saying. I don't think Sportbet.com odds are much different to the odds implied by the Nov 17 EMRS poll. I posted a blog at http://tasfintalk.blogspot.com.au/2018/02/election-odds.html#more

    ReplyDelete
  6. This kind of exercise is very tricky. There are a few things that are important to keep in mind when looking at election odds as a sign of what the markets "think":

    1. Longshot bias. The bookie's "rake" isn't evenly distributed. Events at longer odds happen less often even relative to those odds than events at short odds. The election could be run in 10,000 parallel universes and the Greens would not win two seats in Braddon in any of them, yet it is offered at only 12.00, implying that it could actually happen. One adjustment I've seen used is to cull everything at double figures but that doesn't completely solve the problem as longshot bias can also occur in at least the high single figures too.

    2. Non-independent probabilities. So, for instance, the Liberal chance of winning three in Lyons and Franklin are supposedly 25% and 28%. However if the Liberals do win three in either of these, then that suggests their state vote is at the high end of expectations, which increases their chance of not only winning the other one, but also winning in both Bass and Braddon. (On the other hand if they don't win three in Lyons then that may well mean they're failing somewhere else.) I don't know if this actually makes a big difference in this case, but one of the first bookmakers to field on Australian federal elections didn't realise this problem and was taken to the cleaners by punters on multiplying multi-bets on different seats. Every federal election I see a nonsense model where someone runs Monte Carlos of all the individual seat probabilities and concludes that one side has a 100% chance and the other side has 0%. In fact, if there is a polling error in one direction or another, or a campaign event that causes a swing, then that affects the chances across multiple seats at once.

    Also, specific to this election:

    1. These individual seat odds would have a shallow money pool - they wouldn't be taking many bets on them, and for small amounts of money at that. They would mostly reflect the intuition of a single bookmaker rather than the opinion of a market of (hopefully) informed punters. Sportsbet isn't a totaliser and has stated that this is how it does its odds - they have a modeller set them, and every now and then that person looks at the money coming in and considers whether to make changes.

    2. They don't have any odds on JLN, but that doesn't necessarily mean they don't think JLN can win any seats. (That would increase the chance of a hung parliament somewhat, whereas the point about probabilities not being independent reduces it.)

    I'll be very interested to see how those odds go. As I mentioned on my main guide page, bookies odds have a poor record in Tasmanian elections.

    ReplyDelete

The comment system is unreliable. If you cannot submit comments you can email me a comment (via email link in profile) - email must be entitled: Comment for publication, followed by the name of the article you wish to comment on. Comments are accepted in full or not at all. Comments will be published under the name the email is sent from unless an alias is clearly requested and stated. If you submit a comment which is not accepted within a few days you can also email me and I will check if it has been received.