Life of Bryan, death of Labor?

Bryan Palmer, who maintains the excellent political science site Palmer’s Oz Politics, has a post noting the bookmakers’ latest odds on the federal election race. Bryan also analyses the recently-published marginal seat polling, and concludes that Labor is likely to fall around 4 seats short of the 11 net gains it needs to form government. Unfortunately Bryan doesn’t yet have a permalink feature on his blog. The post I’m talking about is dated Friday 17 September.

Today (Saturday 18 September) Bryan analyses the totally conflicting ACNeilsen and Morgan polls published today, the first showing the Coalition slightly in front and the second showing Labor with a massive lead! If you want to consider these results reasonably carefully and objectively rather than just highlighting the one whose result you prefer, go read Bryan’s post. This bloke knows his psephology and he’s a good writer and political scientist to boot. A great new addition to the blogosphere!

In essence, and sadly for Labor barrackers, Bryan concludes that the ACNeilsen poll is more likely to be correct. His reason? Morgan’s polls are conducted face-to-face, and people are embarrassed to confess to anyone’s face that they’re intending to vote for Howard. Confessing to supporting Howard is a bit like masturbating in public, Bryan thinks:

[E]verybody does it, but not everybody admits to it, especially in face-to-face situations.

Given the trends in all the major polls (showing the Coalition pulling level with Labor by last weekend) except Morgan’s face-to-face ones, and now the Neilsen poll showing the Coalition going ahead, it’s beginning to look like the “opting for the devil you know in prosperous but dangerous times” trend I anticipated in the wake of the Jakarta bombing may indeed be occurring. But there are contradictory indications as well.

About Ken Parish

Ken Parish is a legal academic, with research areas in public law (constitutional and administrative law), civil procedure and teaching & learning theory and practice. He has been a legal academic for almost 20 years. Before that he ran a legal practice in Darwin for 15 years and was a Member of the NT Legislative Assembly for almost 4 years in the early 1990s.
This entry was posted in Politics - national. Bookmark the permalink.
27 Comments
Oldest
Newest Most Voted
Inline Feedbacks
View all comments
Brett
Brett
2022 years ago

The Morgan poll published today is actually a phone poll.

Ken Parish
Ken Parish
2022 years ago

Brett

Sorry, I hadn’t noticed that, and I’m alightly misrepresenting Bryan Palmer. He does suggest that either Neilsen or Morgan must be wrong, and does suggest that Neilsen is the more likely to be correct. But he does so on the basis that Neilsen exhibits less apparent systemic bias over time, in that both Neilsen and Newspoll fluctuate in similar ways. Morgan, on the other hand, doesn’t and Bryan attributes this to its face-to-face methodology.

However, as you rightly point out, today’s Morgan poll was a phone poll, so Bryan’s hypothesis doesn’t really explain the divergence of these two specific polls at all. Note that he also mentions other theories, including one that hypothesises that the pollsters each use different alogrithms for calculating the two-party-preferred result, and that this may introduce systemic bias. That might help to explain the large divergence between the two polls, although it’s difficult to make a direct comparison between them in that respect because Morgan’s press release doesn’t disclose primary votes and two party preferred separately.

Finally, statistical margin of error could partly but probably not wholly explain the discrepancy between the two polls, as Brian discusses (in a rather more complex way than my discussion below). Morgan’s poll has a sample size of 1,000 (for an error margin of about 3%), while Neilsen’s is (I think) 1400 or thereabouts (in any event it claims an error margin of 2.6%). The Morgan poll shows the Coalition on 45.5 and Labor on 54.5. Adjusting for margin of error, the Coalition could be as high as 48.5 and Labor as low as 51.5. Doing a similar exercise on the Neilsen results, the Coalition (which is shown at 51) could be as low as 48.4 and Labor (which is shown at 49) as high as 51.6. As you can see, both polls overlap at the edge of the margin of error adjustment, but only just.

cs
cs
2022 years ago

His reason? Morgan’s polls are conducted face-to-face, and people are embarrassed to confess to anyone’s face that they’re intending to vote for Howard. Confessing to supporting Howard is a bit like masturbating in public, Bryan thinks

As discussed, it was a phone poll. Also, Morgan doesn’t involve any public confession. People don’t tell the pollster, but mark their preference on a secret ballot.

Incidentally, there is some discussion of Morgan’s methodology over at Morgan, posted yesterday. There are interesting arguments re face-to-face, but, in any event, no-one is asked to ‘masturbate in public’ under this method.

FWIW, my guess is that Morgan’s too high and Nielsen’s too low (for Labor, that is), on the basis that Nielsen’s primary for the Coalition at 48 doesn’t look credible (Howard only got 47 on the primary in his ’96 landslide, and was below this in ’98, ’01).

Ken Parish
Ken Parish
2022 years ago

cs

If Morgan uses a secret ballot as part of its normal face-to-face polling, that would tend to negate Bryan Palmer’s hypothesised explanation for Morgan’s divergent trends (compared with both Newspoll and Neilsen).

I notice that Morgan argues that phone polls bias results in favour of the party that’s received the best publicity on the last week, and seems to argue therefore that its standard methodology is better. But again, that doesn’t explain the divergence between today’s Neilsen and Morgan polls, because both were phone polls.

I remain interested in the phenomenon of why Newspoll seems generally and consistently over a long period to have a slight pro-Coalition bias, while Morgan seems to have a long-term rather larger bias in favour of Labor. Individual polls can be explained on margin of error grounds, but a long-term apparent bias in a poll must have some other explanation. Any suggestions? I doubt that either is deliberately acting as a shill for one party, so how come the seeming inherent bias? Neilsen hasn’t really been around long enough to be able to make long-term bias assessments about its polling.

cs
cs
2022 years ago

Got me Ken. You’re right about the apparent bias. Morgan and Newspoll have been level pegging in their rogues of late, each as unreliable as the other. Why? Buggerred if I know. Their money polls are, of course, the election eve polls, where if they get it wrong it goes straight to their national reputations. They don’t have much riding on these mid campaign polls, but their interest remains squarely on getting the things right, not skewing them away. One theory is the mobile phone hypothesis. Increasingly, young people only have mobile phones, and these are missed by Newspoll, skewing it to older people who would favour Jack.

Incidentaly, I once did some face-to-face polling over a summer when I was a student. I still well remember an occasion when it was pouring rain and the poll was due, and I sat around with my supervisor and a variety of pens filling the things in all afternooon! I guess anything can happen.

Ron Brunton
Ron Brunton
2022 years ago

If the Morgan face-to-face polls include a secret ballot for voting intention, how can Morgan then do cross tabs of voting intention against other characteristics without compromising the secrecy, unless all the other characteristics are also obtained by secret ballot? Or aren’t any cross tabs provided for the face-to-face polls?

Homer Paxton
Homer Paxton
2022 years ago

Ron , given the margin of error for such characteristics all three polls are gilding the lilly or howarding when they speak about such things.

They did a secret ballot on thew two occasions when I participated.

The most likely explanation is that morgan has a rogue poll. Newspoll has had two at least this year.

The odds of this happening increases when you go from fortnightly polls to weekly polls as morgan had done ( and Newspoll ).

This election is quite bizaare. It is the first where it is clear the punters would like a change but their conservative nature ( They voted for Keating in 93 remember and he was hopeless)is making them somewhat schizoprenic.

I still like the way Iron Mark’s campaigning now. Why is howard so off song and seemingly nervous?

too close to call howver for all those people who are putting their all their hopes on Centrbet odds I give you two words. cowboys and Storm.
both were at greater odds than the ALp last weekend and got up.

that fat sorry extra kilo lady has not yet sung

Bryan Palmer
2022 years ago

There are two issues I considered in my post. The first is the two telephone polls published today. On the basis of a statistical z-test, I conclude that at least one is wrong. I favour ACNielsen as correct on the basis that it is consistent with Newspoll. This is essentially a two out of three argument. “Masturbating” did not come into my reasoning on this issue.

The second issue is the long term systemic bias between Morgan (pro-Labor) and Newspoll (pro-Coalition). Here I suggest one possible explanation might be embarrassment in admitting to masturbation. Even if Morgan uses a secret ballot (which I understand they do), that does not get past the fact that the interviewer is in the room when the form is completed, and the possibility of respondent suspicion that the interviewer will look at the form when he leaves the house. But this is only one of a number of possible hypotheses. I also mention a couple of others in today’s post.

What intrigues me more is the fact that there is clearly a problem somewhere – one or both of Newspoll and Morgan are doing something (probably unconsciously) that skews the results. No one benefits from this.

Link
2022 years ago

In the unlikely event that I was polled – I would be tempted to lie. I would be very tempted to lie, not sure why.

cs
cs
2022 years ago

The Newspoll actually trends to Labor (by 1.5 points), which may be read as rendering Nielsen as the rogue. Nielsen’s unlikely primary (higher than the LNP has got in an actual election in over 25 years) also suggests it is the rogue.

Still, the discussion is fascinating

James
James
2022 years ago

“Neilsen hasn’t really been around long enough to be able to make long-term bias assessments about its polling. ”

That’s not really the case. Neilsen acquired AGB McNair in the early 1990s – the methodology hasn’t changed, so there is a long history.

I’m interested in the debate about Morgan’s “secret ballot” in its face-to-face interviewing. As Ron points out, if it were really secret, how could they link the results to the respondent demographics. In fact, they got busted for this many years ago. The “secret” ballot form had a hidden code that enabled the interviewer to recover the vote and link it back to the respondent. This, of course, was unethical. I don’t know what method they use now.

As a professional market researcher, I would trust the phone polls over the face-to-face. I also think that Neilsen would be better than Newspoll, although I concede I don’t have the information that would be necessary to conclude that that was the case.

cs
cs
2022 years ago

Do you know the Niesen/McNair record then James? In a recent Herald op ed, out of the last 6 federal elections, the writer scored 4 to Newspoll and three to Morgan, biut only referred to the last poll for Nielsen (which it picked). If its beeen around since the early 1990s, it should have a record across at least 4.

On the face to face issue, I don’t know, but imagine a smaller envelope goes into a larger envelope, with identifiers recorded on a sheet elsewhere – allowing anonymous cross tabbing by the counters.

Shaun
Shaun
2022 years ago

Why the big move in SA – something like 5-7% move to Labor according the latest poll?

Ken Parish
Ken Parish
2022 years ago

Shaun

When you’re dealing with a total national sample of 1000-1500, the nuber of respondents from South Australia is probably around 200 or so. The error margin with a sample of that size is huge, so it really isn’t especially useful to take much notice of those sorts of apparent movements in a single state in a single poll. If we see it reflected in the normal Newspoll and Morgan polls that will be published early this week, then it would be worth taking seriously.

Ken Parish
Ken Parish
2022 years ago

Here’s a copy of a comment I posted on Bryan Palmer’s blog, which also relates to some of the comments here:

Bryan

If you read the sequence of comments under the post on my blog that refers to this post, you’ll see that I realised I had misunderstood your argument as soon as someone pointed out that the latest Neilsen poll was a phone poll.

I thought about your point that maybe the physical presence of the interviewer is an inhibiting factor even if the actual recording of preferences is by secret ballot. However, with a phone poll there’s no element of secrecy from the interviewer at all, even though the interviewer isn’t physically present. Is it more inhibiting to confess a vice to a stranger on the phone, or to do so on paper where you think (although you may not be absolutely certain) that the stranger won’t know who among his many subjects made the confession? I have no idea, and yours is clearly one possibility, although the existence of a secret ballot procedure would seem to reduce somewhat the probability that it’s a credible explanation.

BTW Do the normal fortnightly Morgan polls record primary vote preference separately from two party preferred? I know I could look for myself, and I probably will in the morning, but you probably know anyway. If they do, it should be realtively easy to see whether they do something anomalous with their algorithm by comparing the primary and two party preferred results for Newspoll and Morgan over a series of polls. If Morgan’s adjustment consistently moves the two party preferred result further in the direction of the ALP than Newspoll, that would favour Mumble’s explanation.

Andrew Leigh’s hypothesis that “Labor voters are more likely to be at home or more disposed to talking to a face-to-face interviewer in their house” would be much harder to test and seems to my mind somewhat far-fetched.

On Newspoll’s apparent long-term slight pro-Coalition bias, you might have noticed Christopher Sheil’s suggestion on my blog that it might be caused by the fact that many younger voters only own mobile phones and are therefore not available for phone polling, leading to the sample possibly being biased towards older voters who may be somewhat more inclined to vote for the Coalition. On the face of it that sounds plausible. Are the Newspoll and Neilsen polls completely random samples, or do they attempt to achieve some measure of demographically representative sample by asking questions about age etc? Clearly they DO ask questions about age, because they usually give percentages of party support by age group as well as overall. But do they trim the total overall sample to make it roughly representative of the age makeup of the general community? If not, then Chris’s explanation seems quite likely. That might mean that Newspoll is consistently biased towards the LNP for Chris’s posited reason, and Neilsen exhibits a similar bias for the same reason, while Morgan’s face-to-face polls are slightly biased towards Labor because of your “masturbating in public” theory.

However, even if all that is right, it doesn’t explain the discrepancy between today’s Neilsen and Morgan polls. I suspect Christopher Sheil is probably correct in observing that the LNP primary vote of 47% recorded in today’s Neilsen poll is suspiciously (even unbelievably) high, given that the party has never achieved a primary vote that high over the last decade, not even in the 1996 anti-Keating landslide. That rather suggests that the parties are probably still level-pegging (as they were last week). Why the Morgan phone poll should show such a divergently high pro-Labor result, however, has got me completely stumped.

Bryan Palmer
2022 years ago

Ken

Okay, so the masturbation thesis is not the most compelling. But something is going on. I would love to know what it is. Like you, I am stumped.

To your questions. The pollsters have different approaches to recording TPP, and (I understand) that approach changes during an election period. Mumble (www.mumble.com.au), has the following quote on his site from Antony Green, Remember that Newspoll and Morgan have always asked primary vote questions differently. Morgan offers a list of minor parties, Newspoll ask Labor, Coalition and other and then ask a second question as required. Traditionally Newspoll has had a lower minor party vote as a result.

The young-person mobile-phone hypothesis may explain why Newspoll over-estimated the Coalition by two points last election. But this is still within the error margin. What I cannot work out is how Morgan got it so wrong last election (and I do not buy Morgan’s last minute volatility thesis). And it was not just a once-off blip that happens from time to time. Morgan had an increasing Labor majority in its three polls leading up to the 10 November 2001 election [evidence

Homer  Paxton
Homer Paxton
2022 years ago

if memory serves me right then morgan’s poll was one week out ans therefore di not reflect what happened on election day.

Having said that It was out of kilter with what Newspoll and Neilsen had on the same weekend.

no-on in either of the three organisations will admit it but they get rogue samples.
This is more likely to happen the more frequently you do them such as in election camapaigns when there is pressure to do them weekly.

One major problem Morgan has since his metod is moore labour intensive is combining weekends to get a result. however in this case this didn’t happen.

the one thing that shocked me was the lack of impact of the ALP’s education policy. I thought they would either go up or down a few points.
They didn’t.

Zoe
Zoe
2022 years ago

Much to my disappointment, the Morganistas came around yesterday while I was out. They asked a questions ranging well beyond politics and left booklets to be filled out by my husbang and a pen which says:

Roy Morgan
– Research –
http://www.roymorgan.com
Your Opinion Counts!

All offers will be considered. In fact, any offers will be considered.

observa
observa
2022 years ago

The large number of waverers haven’t decided yet and won’t until they’re in the booth faced with the ballot paper. They’ll vote the way they feel then. Anything can happen now.

Homer Paxton
Homer Paxton
2022 years ago

two further points.

why does Morgan have a bias in federal elections but not State elections where its record is impeccable.

If Newspoll has a bias to the Libs then why doesn’t Neilsen as its technique is identical?

Don Munro
Don Munro
2022 years ago

All this reminds me of the Frank and Ernest cartoon in which Ernest (holding a clipboard) asks “Do you always tell the truth to pollsters?” and Frank answers “No.”

Homer Paxton
Homer Paxton
2022 years ago

It looks like that it is Neilsen that has the problem,
sorry Gary

Ken Parish
Ken Parish
2022 years ago

Homer

Yes it does look like that. But interestingly Newspoll has Labor up by about 1.5 points on a 2pp basis compared with their last poll, whereas Morgan has Labor down by a similar margin. I wonder why? Nevertheless, they both seem to have converged from opposite directions on a position that puts Labor a fraction ahead (thankfully – I was starting to feel very pessimistic).

Homer Paxton
Homer Paxton
2022 years ago

no explanation from me however As I commented at BP it is annoying that firdt Morgan and now Sol is putting decimal points in their results.

The ABS survey of the labour force has a margin of eror of around .075% and they round the numbers so should Sol and Gaz.

Geoff Honnor
Geoff Honnor
2022 years ago

Malcolm Mackerras predicted a Coalition win by 10 seats on “Australia Talks Back” last night. It quite put me off my dinner preparation until I recalled that Malcolm is more famous for his electoral pendulum than he is for the accuracy of his predictions. His point seemed to be that a hung parliament isn’t on the cards, despite what some pundits might be surmising. He also opined about a Senate outcome of 38 or maybe 37 to the Coalition, 24ish to the ALP, 7 0r 8 to the Greens and about 4 Dems.

Homer Paxton
Homer Paxton
2022 years ago

that is proabably unlikely now.
We know most people vote above the line in the Senate thus the preference deals become vastly important.

My guess is that it means 34-35 seats now for the Government

trackback
2022 years ago

Fear, loathing, etc #11

With Mark Latham surging in the polls, here is an open-election post for folks to raise or report or discuss anything they wish about Australia’s ’04 date with destiny. We seem to be getting through these posts at an increasingly…