Monday, January 16, 2017

GetUp! National And Centrelink Poll Reporting Is A Trainwreck

GetUp! ReachTEL (undecided redistributed) Coalition 37.1 ALP 35 Green 9.8 PHON 10.6 
Published 2PP 54-46 to Labor
2PP by 2016 preferences 52.1 to Labor
Verdict: Go back to sleep

[Updated 24 Jan with fresh Essential polling on Centrelink issues]

My little eyes lit up when I saw in my Twitter stream that I had somehow missed the release of a national ReachTEL at about midnight last night.  There has not been a national ReachTEL since before the July election. Given the relative paucity of polling data since then, and the Centrelink and ministerial "entitlements" issues currently affecting the Turnbull Government, new data concerning where the government was standing could be quite interesting.

Unfortunately it turned out that this was not a new Seven or Fairfax ReachTEL, but rather one commissioned by the lobby group GetUp!  Moreover, the level of immediate publication of the poll's details has been abysmal.  Rather than it being promptly released with full details either on GetUp!'s website or the ReachTEL site, what seems to have happened is that it has been sent (in part or full, who knows) to a range of media agencies who have then presented us with a partially digested dog's breakfast of the findings.

Thus for instance from the SMH we learn that:

"While primary support for the Liberal party and Labor was deadlocked at 32 per cent, the opposition captured the majority of the "undecided" support 33.3 per cent to 19.4, with 58.5 per cent of those still making up their mind answering they would preference Labor higher."

For the Turnbull Government to be behind the Labor Opposition on the primary vote (after redistributing the undecided) would be startling news indeed, given that the Government has held a primary vote lead - albeit generally a feeble one - in all polls since the election but for one Newspoll and five Essentials that showed ties.  However, from the figures published by Poll Bludger it becomes clear that the Government is ahead on primaries in this poll too, because the figures published by the SMH are, as stated, a reference to the Liberal Party only, and ignore their Coalition partner the Nationals.

Poll Bludger publishes the primaries as 37.1 Coalition, 35.0 Labor, 9.8 Green, 10.6 One Nation.  GhostWhoVotes on Twitter additionally gives Other as 7.5, which doesn't quite sum to 100 but the loss of a few tenths here and there in rounding will make little difference to the picture.  The published 2PP is 54-46 to Labor.  However, by last-election preferences, this poll comes to only 52.1% to Labor.  So, unless One Nation voters have suddenly developed a massive preference for the Labor Opposition over the Coalition (which is possible, but would fly in the face of the entire history of One Nation preferencing to date) then we are probably looking at the normal problems with respondent preferencing: it skews to Labor overall, but it also makes the estimate of the 2PP vote in specific polls more volatile.

If the primary votes are really as stated in this poll then the conclusion to be drawn would be that the Government has suffered no damage on a 2PP front from the Centrelink debt recovery atrocity or the Sussan Ley scandal.  Compared to the BludgerTrack aggregated primaries as at the end of last year, which produce exactly the same aggregated 2PP as my own estimate, the only real action to see is a drift of about a point each from the Government and Opposition to the seemingly rampant One Nation.  The One Nation surge in this ReachTEL could be a product of sample noise or sampling method rather than a genuine shift, but you shouldn't bet your bottom dollar on either of those at the moment.

Even if the poll did show a primary vote shift that matched the reported 54:46, this would not amount to evidence that the government had taken any hit at all (it could just be sample noise), and even if there was such a hit it would not show it would be caused by anything in particular.  So GetUp! has got the story it wanted about the Centrelink debt recovery atrocity damaging the Government's standings, but the evidence behind the claim is flimsy.

This is no way to conduct national polling releases.  If GetUp! want to be taken remotely seriously they should release their polling properly rather than only feeding it (in an unknown form) to the media.

Centrelink: A Personal View (Rant Warning!)

Readers may have noticed that I describe the Centrelink debt recovery system as an "atrocity" above. From what I have been able to glean from the large amount of often confused coverage of this matter, Centrelink recipients present and past are sent letters advising that data matching has suggested possible discrepancies between what they were paid and what they should have been paid, and asking them to verify their details online.  I would have no problem with this whatsoever if it provided an easy road to resolution for those who were in fact paid the correct amount.

However, recipients who earned income for a part of the year while they were not on benefits are in some cases confirming that their annual income is correct, and then for having the temerity to naively make a true statement, are being plunged into a nightmare in which a bogus "debt" is raised against them and they have to challenge it and prove they don't owe money.

Challenging is onerous because of the difficulties of collecting information from several years ago and severe problems with being able to contact Centrelink in a timely fashion without getting wrapped up in red tape and incompetence.  Also, fortnightly pay slips often mismatch fortnightly reporting to Centrelink, in part because of different fortnightly schedules.  Centrelink's bizarre practice of asking recipients to report earnings in advance (what work you did for which you will be paid) rather than as they are received is a further complication here. It is a classic case of trying to use data because you have it, without considering whether the data are fit for purpose.

For those having to challenge a false debt, this can become what in a busy world is a nightmare: an open-ended time-sink in which one must waste an unknown amount of time for which one is not compensated, in order to (maybe) prove that one is not guilty of a debt for which one has been presumed guilty on insufficient evidence.  It doesn't help that in some cases if the original payments were incorrect, this was in fact as a result of Centrelink incompetence with their own rules rather than any wrongdoing by the recipient, and yet the recipient will be made to do the donkey work.

This should be a national disgrace as it is totally contrary to supposed Australian values of a fair go. But it is not an isolated case; it is just the latest ramping-up. The administration of welfare payments, in which jobseekers who are financially and psychologically vulnerable are humiliated and dehumanised by being compelled to apply for jobs they have no chance of getting, has been a disaster zone for at least 20 years, with bipartisan support from the major parties and little meaningful opposition from anywhere else.  Even lobbying from the Greens, for instance, has far too often focused on the financial levels of payment (which is an issue) and not on how recipients are treated.

Centrelink: The Polling

Much as I would like to report that GetUp!'s polling shows that Australians share my views of this appallingly clumsy and nasty debt recovery scheme and Centrelink in general, the GetUp! polling on Centrelink isn't very useful.  As related by Poll Bludger (the exact wording of the questions being elusive in all the mainstream media coverage) recipients were asked:

“The Turnbull Government recently started using an automated system issuing tens of thousands of letters to Australians about alleged debts from Centrelink overpayments. The Government admits that at least 20% of these letters are incorrect, but the burden is on Centrelink clients to correct the information or pay the debt. Do you support or oppose the Government stopping the automated debt collection system?”

In undeserved fairness to the Government on this point, the Government in fact disputes that 20% of the initial letters are "incorrect" (let alone that it has said they are) on the grounds that the initial letters merely raise a potential issue and ask for confirmation of income.  (The sleight of hand in this claim concerns what happens when a person agrees that their annual income was as stated without realising this will then result in an incorrect debt.)

In any case even if the claims in the preamble are true, the placement of claims favouring one side of the debate and not the other (for instance, that the scheme is successfully recovering some amount of money) could potentially bias the poll in favour of a negative result for the scheme.  But the double negative in the final question (asking recipients whether they support or oppose stopping the scheme) might confuse some respondents.  Moreover, some respondents might oppose stopping the scheme but might support serious changes.  So the net finding of 46.2% supporting stopping and 31.8% opposing stopping is simply number salad - you can make of it anything you want.

A further question reported is:

"The Turnbull Government has acknowledged significant errors in the Centrelink automated debt collection system. Where there are potential errors, do you think the burden should be on Centrelink to verify their claims against information they already have on file or on the individual to defend themselves, which may include accessing pay slips and employment records from up to five years ago?”

Here 78.6% agreed with the former option while 21.4% disagreed, and it seems that we have another problem here, as it seems forced choice was used for an issues question, which is inappropriate as it forces respondents who have no opinion to concoct one or abandon the survey.

It is also questionable that the Turnbull Government has acknowledged significant errors - Alan Tudge still doesn't. But that aside, the problem with this question is that it is likely to be primed by the previous questions and therefore the result is unreliable.  This highlights the need for full detail to be released in such polls immediately including the order of all questions asked and their full wording.

There is also the usual question about whether the scheme has made people more or less likely to vote for the government (49.8% less likely, 14.4% more likely.)  The effect size is a little larger here than the 30% that a question of this kind will produce for pretty much anything, but the level of priming from the other questions in the poll contaminates it.

The Centrelink debt recovery scheme is a very serious matter, and the letters sent out so far are only the tip of an intended iceberg.  The impact of the issue on the ability of the agency to even function could be very severe, with consequences beyond those who are even affected (the numbers of which alone stand to become very large.)  The issue deserves much better than to be used as a tool in GetUp!'s usual game in which scientific polling is something you muck about with in order to convince the media that the electorate feels a certain way.  I can only hope that the Centrelink debt recovery matter will feature in properly designed national polling commissioned by neutral sources, so we can get a better idea of what voters actually think.

A nearly final word on this belongs to the GetUp! campaigns director Mark Connolly, as reported by Sky News:

"Treating Australians like crap is going to get you crap poll results."

No, in this case it has only got us a crap poll.

Update (24 Jan): Essential

Essential has also polled the Centrelink question, but has done so using a much better wording than GetUp!  This is what Essential asked:

"Centrelink is currently conducting a debt recovery program in which welfare recipients are being automatically sent notifications regarding possible over payments. From what you have read or heard do you approve of disapprove of the way this program has been conducted?"

There are issues with Essential's issue polling, but at least here we have a well-designed question that doesn't spruik for either side of the argument or confuse the respondent.  The result is not wildly different from GetUp!'s, with 36% approving and 48% disapproving.  Typically, Coalition voters tend to approve while voters for other parties tend not to.

I'm inclined to downplay the following question:

"Which of the following issues are you more concerned about - over payments to welfare recipients or how politicians use their travel expenses?"

...because the Centrelink question prior to it might have affected the outcome.  But for the record 8% said welfare, 46% said polly perks, and 40% said both equally.  Again party preference had an impact with Coalition voters more likely to say both equally.

Essential has also recorded its worst 2PP reading for the Coalition under Malcolm Turnbull (46:54).  This was as low as the Coalition got with Tony Abbott as PM, as Essential had them at 46% twice in mid-2014 and six times in early 2015.  That said all of Newspoll, Nielsen, Galaxy and ReachTEL at some time had the Coalition lower.  The current reading seems to come off a woeful one-week sample since the Coalition had a primary of 38 in the first half of the sample but finished up on 35.  This suggests the Coalition's primary this week was in the range 30.5 to 33.5.

After taking into account the primaries, I have aggregated this Essential at 45.8, which is actually lower than anything in the Coalition's first term.  My aggregate is on holidays until there are enough polls in it for a meaningful result, which I suspect will be the case sometime in the next two weeks.