Tuesday, July 23, 2013

Poll Roundup and Seat Betting Watch July 23

2PP Aggregate (Tuesday 23 July): 50.3 TO COALITION (+0.3 SINCE LAST WEEK)
Individual Seat Betting: Labor favourites in 63 seats (no change)
Seat Total Market: Labor 66 seats (-2)

This is week four in a regular weekly series in the leadup to the federal election.  Week three was here and through it you can click back to the previous weeks.  Or just click the "betting" label at the bottom.  As stated before, the aim of this exercise is not to claim that seat betting markets have predictive value, but to test whether they do, and to see which of the markets and the aggregated polls see the ultimate outcome of the election first.

This Week's Polls

So far this week we have had three polls: a 52-48 to Labor (by last election preferences) from Morgan Multi-Mode, a 52-48 to Coalition from Newspoll, and a  51-49 (It is alive! It moves!) to Coalition from Essential. The last represented Essential's best reading for Labor since February 2011.

Aggregators out so far (Mark the Ballot, Pottinger) have shown not much overall movement as a result of this and project the most likely outcome as a very narrow Coalition win (in MTB's case if an election was held now, in Pottinger's assuming Aug 31 as the date).  I suspect Bludgertrack will show similar, or perhaps a little more to the Coalition. 

My interpretation is that Labor would probably very narrowly lose an election held right now.  I say this not because of the 2PP picture but because the state picture is not lining up well enough - Labor is increasingly struggling to get the swing required for substantial gains in Queensland, but is apparently still in trouble in Victoria, NSW and Tasmania.  However, there is plenty of time for this to change.




Morgan MM has now had Labor in the lead four times when no other pollster has done so even once except for the single sample from AMR (who are not regular players).  Given Morgan's very large sample size I can't ignore this behaviour any more and have assumed a house effect and slugged its reading for Labor one point in my aggregate, though there is a strong case for slugging it two.  I was going to apply a house effect correction to Essential as well, but after this week's reading I've given it the benefit of the doubt on the possibility that it's just very slow to respond to recent changes.   The case for adjusting it a point in one direction  probably now close to matches the case for slugging Morgan two in the other, so hopefully my (over?)conservative responses to both weekly polls will cancel out. 

A Newspoll question about which party was best to handle asylum seeker issues, one of the hot issues of the week following Labor's big policy shift on the matter, saw Labor rated best by 26% of voters to the Coalition's 33% (full results here).  This compared with a 20-47 reading last time the question was asked, but many people did not realise that the question was last asked in February, when Gillard was leader and the Coalition lead had just blown out to a 56-44 lead.  Therefore, the "counterintuitive" nature of the findings (Labor 2PP down, but asylum seeker rating much improved) was explained by the asylum seeker rating comparison being with a Gillard rating, not a Rudd one.

I can't find the full support-by-party breakdown for the February figures but some aspects have been widely reported.  Among Coalition supporters, their own party is down nine points as the best party to handle asylum seeker issues, from 80% in February to 71%, while Labor is up three from 4% to 7%.  Among ALP supporters, Labor is up three from 53% to 56% and the Coalition is down 16 from 21% to 5%. But while it might be expected that Labor's move to the right has resulted in more of its own voters preferring the Greens' policies, this actually hasn't happened.  In Newspoll the support for the policies of "others" on the matter among all voters is stable at 12%, and in Essential support for Green policies on asylum seekers is up only one point on a month ago.  The big winners apart from Labor in Newspoll are the None and Uncommitted categories (+4 apiece) while in Essential the movement is more or less straight from the Coalition to the ALP.

These polling patterns may represent more or less direct movement from voters preferring the Coalition on the issue to voters either preferring Labor on the issue or not caring, or they may represent more complex churn patterns between the different categories.  But the most striking thing out of both Essential and Newspoll is that the Greens' policies on asylum seekers are not becoming more popular as a result.  If Labor voters are disgusted with the new regime, they're not embracing the Greens' alternative either.  There was, however, a small move to the Greens on primary vote in the readings of two of the three pollsters.

Another notable thing about the breakdowns for the Newspoll asylum-seeker question is this: they actually need help from rounding to add up.  71% of 45% (Coalition supporters preferring the Coalition on asylum seekers) plus  5% of 37% (Labor supporters preferring the Coalition) is 33.8%, but the Coalition's support on the question as a total is 33%.  Making the most generous allowances for rounding possible, the 33.8% can be reduced to 33.0% and the 33% raised to 33.5%.  That means that at most 6%, and probably not even that, of non-major-party supporters prefer the Coalition's asylum-seeker policies over all others, suggesting that support for those policies from Ind/Others voters is very small.  It may be that supporters of Independent and Others candidates tend strongly to give giving uncommitted/none of the above/someone else answers to the question.  The support for Labor asylum-seeker policies over all others from Green/Ind/Others voters is probably about 18%.  Most of that would be coming from Ind/Others.

The Newspoll saw Kevin Rudd with a net satisfaction of +1, down from +7 last time.  I previously noted that Rudd's improvement in net satisfaction from his first to his second Newspoll as restored PM was one of the best for a new leader.  It is normal for new leaders to experience a rise in dissatisfaction (as Rudd did, from 36 to 41) in their third Newspoll.   His net change from first to third Newspoll (+1) is a little below the median of +3.  (Most leaders do better in their third Newspoll than their first, but Nelson and Gillard did very much worse.)

Tony Abbott recorded a satisfaction rating of 35 and a dissatisfaction of 56, giving him a netsat of -21 and meaning that he has recorded identical ratings for three consecutive Newspolls.  There is one previous case of this - in 1997 Kim Beazley recorded 48-34 three times in a row.  Cases of a leader having one score or the other the same three times in a row are fairly common (there have been 29 in all) and cases of a leader polling identical ratings twice in a row are also reasonably common (31 in total).  There have been four cases so far of a leader having one score or the other (but not both) the same four times in a row. 

Newspoll Nonsense Ahoy!

All up we've had one poll move 0.5 points in Labor's favour, one poll which uses a rolling average move 1 point in Labor's favour and one poll move 2 points in the Coalition's favour.  But Newspoll generates vastly more attention than Essential and Morgan in all of the media, the betting markets and the political classes, and so the response to a two-point shift in Newspoll has been completely over-the-top.  Let's keep in mind here that two points is about the normal move in poll to poll from Newspoll, as a result of sample error, although the normal move in actual voting intention between Newspolls is a fraction of a point.  Irrational responses to this Newspoll fell into three categories:

1. That it was definite proof the surge was over and Labor will now lose.
2. That it was movement within the MOE, or cancelled out by Morgan, and should be ignored.
3. Complete lunacy.

In category 3 I include the genre of Newspoll conspiracy theories, as promoted by Bob Ellis and many others, in which virtually any aspect of a Newspoll result other than it showing Labor leading, and virtually any timing of release, may be argued as evidence of corruption and fiddling.  This time the supposed irregularity was that Newspoll was late, when actually it wasn't.  A Tuesday release with the first results out Monday night has been normal for several months except when circumstances (like new Prime Ministers and other such trivia) have justified surveying and releasing earlier.

As for category 2, a 52-48 from Morgan doesn't cancel out a 48-52 from Newspoll if Morgan has a house effect, as seems very likely.  And the widespread awareness of the concept of margin of error (MOE) among the pollgazing public has unfortunately shown again that a little knowledge is a dangerous thing.  The more significant problem is that a move from poll to poll that is not in itself statistically significant does not prove that nothing has changed.  It just means we can't be very confident there has been a change.

But if a party's polling position worsens by an amount that is not statistically significant, its modelled chances of victory will nonetheless go down, because while the two-point shift in Newspoll is probably mostly noise, it still makes it more likely there's been some move away from the ALP than some move to it.

To demonstrate the same thing on a bigger scale, suppose one week an aggregate of 5000 votes from different pollsters shows Labor at 50:50.  Then the next week the same aggregate shows the Coalition leading 50.3:49.7 based on the same sample size.  The movement is nowhere near statistically significant.  But in the first case the chance that Labor is actually ahead is 50% and in the second it is only 33.6%.  In a sound election simulation, this would flow on to a substantial difference (perhaps, say, 10%) in Labor's chances of winning if an election was to be held at that time. And it would flow on to some difference in Labor's chances of winning in a month or two.

(An extra minor quibble is that the concept of margin of error compares a given poll with the underlying real distribution, and that it is not designed for comparing a poll with another poll (no matter how much ABC policy on reporting polls wrongly thinks it is).  In practice, this distinction doesn't usually make much difference.)

As for category 1, this is another case of one swallow not yet making a summer.  The two-point drop in Newspoll is countered by rises in both Morgan and Essential and comes from a poll that's quite bouncy.  The likelihood is that actual movement in voter sentiment this week, if any, has been just the usual fraction of a point - for all the noise about carbon taxes, refugee policy, Fringe Benefits Tax and more. The most significant thing about the poll is that it fuels a narrative that the bounce is over and that Labor's support will now subside, and we can expect that narrative to have a lot of legs for the next few days, even if it turns out not to be true.

Frankly, the poll finding in the last week that made me cagiest about Labor's chances wasn't the Newspoll at all.  Rather, the Queensland state breakdown in the ReachTEL results, which wasn't publicly released but showed Labor getting hardly any swing compared with 2010 off a sample of 448.  That's an outlying value compared to other pollsters, and it's not a big sample, but it does continue a trend of Labor's lead being slowly whittled down in that state as more data comes in.  Morgan also delivered bad news in Queensland, with a Labor 2PP of only 48.5% (respondent-allocated, four points below the national average by that method).  Given the sophomore effect problems Labor faces in winning seats in that state (and Mumble has now written the deluxe version on that issue here) it seems that the battle to win a lot of seats in Queensland is steadily becoming harder.  And if Labor can't win multiple seats in Queensland and can't win the national 2PP substantially, they've got problems.

When Newspoll Sneezes The Market Gets Pneumonia

Newspoll has a massive impact on betting markets, as noted by Simon Jackman here.  Labor headline odds, which dipped to below $3 last week before ending the week just above $3, have been moving into the high $3 range through the day.

The Newspoll result has also seen a great firming in Tony Abbott's perceived chances of remaining Opposition Leader.  On the weekend he was at $1.08 and $1.12 on the two main exchanges; now it's $1.02 and $1.07.  Betfair's interesting "Next Elected Prime Minister" market has seen Abbott come down from about $1.48 to $1.32 while Turnbull has blown out from $14-ish to $22.

Some of the moves late last week may have been driven by belief that the replacement of Abbott with Turnbull was becoming less unlikely, and that this actually increased the Coalition's chances because Turnbull would be more likely to beat Rudd.  But this week the markets seem to be taking a two-point move in Newspoll as meaning more than it actually does.  Either that, or the markets think that the change Kevin Rudd made to asylum seeker policy should have resulted in an increase in Labor support.

On to the seat betting roundup.  Firstly the "Correct Election Result" market on Centrebet has reversed last week's gains and taken Labor back to an expected 66 seats.  A major cause of this is a plunge on the idea that Labor will be soundly beaten, with Coalition 91-100 seats shortening to $4.20.

As for individual seats, here's the chart as at about 5pm Tuesday.  This is mainly based on Centrebet/Sportingbet as Sportsbet has taken many of the interesting seats offline today.


The colour-coding again:

Medium blue: A seat in which the Coalition is favoured in all betting markets.
Pale blue (none this week): Coalition favoured in some markets, level in others.
Grey (none this week): All markets tied or both parties ahead in some markets and behind in others.
Orange (none this week): Labor favoured in some markets, level in others.
White: Labor favoured in all markets.

Bold shows a seat that has changed in colour on the graph from last week.  There are only two of those this week, and both are seats that were tied in one market last week, but only available in one market this week.

So far, no change in terms of seat favourites following Newspoll, with Labor still favourite in 63 seats, and expected to lose the same eleven of its own seats while gaining Brisbane as well as Melbourne from the Greens.

However the close seat picture is where things have changed.  The number of seats where Labor are close favourites is up three to thirteen including Melbourne.  The number of seats where the Coalition are close favourites is down one to twelve.  Surprisingly, the "close" seats held by Labor now include Barton (6.9%) and Richmond (7%).  The Coalition's "close" seats are again mostly either seats it wants to hold in Queensland, or seats it is looking to win in NSW. 

There may well be more significant movements tonight or tomorrow; I may update when Sportsbet again shows the full seat list.

Bludgertrack Updated (24 July): As expected the current Bludgertrack update shows the Coalition leading, 50.5-49.5.  The Queensland recovery on Bludgertrack has subsided to a gain of only 3 seats, and what BT is currently projecting is effectively the status quo (nationwide) allowing for Lyne and New England reverting to the Coalition.  Bludgertrack is also now back to projecting 2 seats for Labor in Tasmania after a period of projecting 3.

Recommended Reading (25 July): Simon Jackman's latest Guardian article , also dealing with the over-emphasis on relatively slight movements in single Newspolls, is excellent.  And yet another aggregator gets more or less the same result; his has the Coalition leading 50.4-49.6.



8 comments:

  1. Kevin,

    Can you please explain your comment further about polls not being designed to compared to other polls? I appreciate that there is post-sampling stratification that goes on, but why cannot you just calculate the difference of values and the appropriate standard error? (I guess the trick comes down to working out what the standard error is properly).

    ReplyDelete
    Replies
    1. I take it that this relates to my comment that commences "An extra minor quibble". The maximum MOE figures given by pollsters at the 95% confidence level mean that if the result is 50% then the real figure should have a 95% chance of being within the MOE of that 50%.

      Problems arise when people compare from poll to poll and say that the difference between them is outside the MOE of the poll and therefore the change is statistically significant. But when you're comparing two samples that each have their own random error, the difference required for statistical significance is slightly higher than when comparing one random sample with a hypothetical actual distribution that has no sampling error. So yes, it's just a question of obtaining the correct threshhold given both sample sizes, which is not the same thing as the MOE for a single sample as published by the pollster.

      An extra quibble is that pollsters often publish just the maximum MOE for their sample and fail to note that when the %age result is much lower or higher than 50 then the MOE decreases. This often results in changes in the Green vote being dismissed as within MOE when they may be statistically meaningful.

      Delete
    2. Thanks - this is what I thought. I thought that there might be some fundamental reason I was missing which related to sampling methods etc.

      Agreed re standard error - assuming similar sample sizes and similar values the standard error of the difference will be about 40% larger.

      I don't bother with the stated MOEs because they are stated as maxima. I use the binomial formula and sample size where available. This of course is wrong anyway because of the stratification, but it has the advantage of being self consistent.

      Thanks for your explanation.

      Delete
  2. Hi Kevin,

    I am a long term reader of this blog and I wish to thank you for continuing to contribute such fascinating and insightful information for us all, it's really appreciated!!

    If you have the time would you mind letting us know what the most recent Reach Tel Poll State Breakdowns were, as I was especially interested in your comments regarding the lack of a swing to the ALP in QLD. Also how do you get these state breakdowns from Reach Tel? Can you just email them and ask for them since they don't publish them?

    Thanks

    Andrew

    ReplyDelete
    Replies
    1. Not sure I can publish the full ReachTEL breakdowns - how I got them was that someone who emailed ReachTEL for them passed them on to me for my interest. Can email on on condition of non-publication; unless I'm asked not to.

      Delete
    2. Thanks for the information Kevin

      Delete
  3. Kevin,

    i am highlighting all your federal political comment/analysis each week every friday.

    I hope you gain more readers. you should. This is a high quality blog.

    Comments ,ecept for mine of course , only add to the quality

    ReplyDelete
  4. Ta. Pleased to report readership here is continuing to grow and doing so at a steep and apparently linear rate. Some of this may reflect increasing interest in pseph sites generally as the federal election approaches.

    ReplyDelete