Possums Pollytics

Politics, elections and piffle plinking

Newspoll, Pollytrack, Satisfaction Ratings and the Greens.

Posted by Possum Comitatus on July 15, 2008

Another Newspoll Tuesday, another bunch of charts. Todays poll has the TPP the same as last time at 55/44 with the ALP and the Coalition both down 1% to 43% and 38% respectively in the primary vote.

First up, we’ll take a look at where Pollytrack stands at the moment on both primaries and TPPs. This is the three pollster rolling aggregation weighted by sample size of Morgan Phone Polls, ACN and Newspoll. Current MoE from sampling error stands at 1.56% from a sample size of 3931.

Next up, the bastard cousin of Pollytrack, Loess Allpolls for primaries and TPP. This runs a Loess regression through every poll of those three pollsters, including Morgan Face to Face.

Next up, let us have a little squiz at how the Newspoll satisfaction ratings have been playing out this year.

That kind of speaks for itself.

And finally, we always follow the ALP and the Coalition, but this time it might be worth adding the Greens to this to see how their support levels have been changing over time. Newspoll goes back to 1997 for the Greens, so we’ll just start from there. Note, I made this yesterday before todays Newpsoll came out. The greens are currently on 11% in todays poll.

Who else thinks something dodgy was going on in the sampling for the Greens to jump from around 4% in mid to late 2007 to 12% last month?

The alternative is that many Greens voters played tootsie with Labor last year but couldnt quite bring themselves to actually go through with it on the primary vote, resorting to a kind of preferential love instead.

UPDATE:

In comments, Sir Ian Upton, apart from cracking a funny, made a very good point about the Greens vote in regard to Gippsland.

So let’s look at the Greens vote at the last election in terms of which kinds of seats the Greens do better or worse in. We’ll use the AEC socio-demographic seat categories here of Inner Metro, Outer Metro, Provincial and Rural.

If we turn those into dummy variables and run a regression using the Greens primary vote at the last election, by seat, and those variables – we get the following results.

We didn’t include the Inner Metro seats as a variable as that’s our baseline. So what this tells us is that in Inner Metro seats the Greens got an average of 9.79% of the vote (the Constant C value for this regression).

In Outer Metro seats the average vote is 2.32% less than the Inner Metro vote, in the Provincial seats the vote was on average 1.81% less than the Inner Metro Green vote, and in Rural seats the Greens scored an average of 3.58% less than their Inner Metro seats.

If we look at Gippsland, at the election the Greens scored 5.54% of the vote, or 4.25% less than their Inner Metro vote which is around what we would have expected.

If we run that regression again, but this time just against the dummy variable of the Rural seats, we get a result that tells us the Green vote in Rural electorates is 2.29% below their average national vote.

During the recent by-election, the Greens scored 7.04% of the vote which, ignoring any by-election dynamics for a second, would suggest the current Greens national polling should be around the 9-10% mark (7.04% +2.29% ) to be consistent with the Gippsland result.

And that’s about where it is at the moment. So it would seem that the Greens have probably had a bit of a boost in their vote since the election, which is coming out in the polling.

54 Responses to “Newspoll, Pollytrack, Satisfaction Ratings and the Greens.”

  1. gandhi said

    Does this Greens result vindicate my recent comments on the Possum Box or what?

  2. Rx said

    I don’t know if such figures exist, but it would be interesting to see a breakdown of support for the parties of the Coalition, and how they compare with the Greens at 11%. Is it likely, do you think, Possum, that the Greens now have a larger base than the Nationals?

  3. Rx said

    Further to the theme of the demographic trainwreck unfolding for the conservatives, Michelle Grattan in The Age reports that “demographers [have looked] at the [National] party’s membership prospects, and it faces a “critical problem in the next decade”.

    http://www.theage.com.au/national/nats-warned-do-nothing-and-you-will-die-20080714-3f3z.html?page=-1

  4. Possum Comitatus said

    RX, the Nats generally sit on about 5 to 6% primary vote in real terms lately, even though the polling tends to have them a few points below that. This comes from a mix of rural areas being a little harder to sample correctly than metro areas (well, some pollsters find anyway) as well as the Liberal/National voter substitution effect whereby some rural and regional voters will tell the pollster they intend voting for the Liberals even though their local Coalition candidate will actually be a Nat. That can add a point or two to the Liberal vote and reduce the Nat vote by the same amount – but once you add them together it all comes out in the wash.

    The Greens are the third largest party by voteshare in the country, having outpolled the Nats at the last two elections. But the big difference between the two is that the Nats have all their voters clustered in a few seats while the Greens have them spread across the entire country.

    So the Greens certainly have a larger voting base – even of it’s not in the right geographical places for them to gain lower house seats (Although Adam Brandt did well in Tanners seat).

    It would be interesting to know which of the Nats and Greens had a larger party membership.

    Anyone know?

  5. gandhi said

    Possum,

    Can’t see hard figures on membership numbers at either party site, but worth noting the Greens’ claim to be the fastest growing party in Australia.

  6. Aspirational Aspirationalist said

    I saw Christine Milne on Q and A last week do a very good job. The only point i’d argue with her was her assertion Dr Mills’s Solar Thermal is baseload power generation. Several days of thick cloud will still cause the plant a bit of grief, but nothing calling a couple of fossil-fueled backup stations online couldnt fix and its operational output is still much easier to predict than wind.

    But I’d really like to hear the government say how we are actually going to get more renewable generation onto the grid and reduce our emissions and know that we actually can achieve our targets.

  7. Andos said

    No ‘r’ in Bandt, Possum. (I think there should be an ‘i’ in there, though… but that’s just my juvenile sense of humour)

  8. Possum Comitatus said

    Oops – many apologies to Adam.

  9. Zaf said

    If you go the Newspoll’s website and look at the ‘Geographic and Demographic Analysis’ they published on 3 July 08, it gives you the age breakdown of Coalition, Labor, Green and other voters. (Also capital cities/non-capital cities.)

    Unsurprisingly the figures illustrate the demographic abyss awaiting the Coalition (especially the Nationals.) What looks interesting, however, is the movement between parties, and how that breaks down demographically, from the November 07 election onwards.

    The Coalition lost support from all age groups after its November defeat. If I understand the table correctly (admittedly arguable) 8% fewer 18-34 yos, 10% fewer 35-49 yos and 7% fewer 50+ yos supported the Coalition in the Jan-March survey than had in the November election. Where did they go?

    18-34 yos
    – 8% Coalition
    0% – no change; Labor
    +7% Greens
    +1% Others

    35-49 yos
    -10%Coalition
    +5% Labor
    +5% Greens
    0% – no change; Others

    50+ yos:
    -7% Coalition
    +4% Labor
    +2% Greens
    +1% Others

    Bearing in mind that these figures just show net changes (vote gains + vote losses), and that there’s no evidence (here) to support any other conclusions about voter movement, it looks like the youngest cohort Coalition voter attrition either was directly mostly to the Greens, or filled a gap in Labor support that was caused by a movement of Labor supporters to the Greens. In the next age cohort (35-49 yos) Coalition voter attrition benefited Labor and the Greens equally (net). In the oldest group, Coalition voter attrition benefited Labor twice as much as it did the Greens (net).

    Was this due to actual voter movement between the Coalition and the Greens, or rather due to Labor lurching convincingly rightward enough to pick up some Coalition voters on the roundabouts only to lose an equivalent number to the Greens on the swings? Is there any way of finding out?

    The decline in Coalition support seems fairly uniform (7-10%) across all age groups, yet the results for Labor and the Greens vary quite a bit (from mostly benefiting the Greens, to 50:50, to benefiting Labor twice as much as the Greens) from age group to age group. Does this imply a direct movement of voters or is that just fuzzy thinking on my part?

    And any forecasts about how the increasing Green vote will play out in State elections – for example in NSW, where State Labor is on the nose, the Opposition lacks charisma and some inner city demographics give the Greens a chance at lower house seats?

  10. Possum Comitatus said

    Zaf (and others interested) – that Newspoll quarterly stuff can also be seen graphically by pressing the “Newspoll Quarterly” button at the top of the site here, or by clicking this:

    https://possumcomitatus.wordpress.com/newspoll-quarterly-breakdown/

    There’s no real way to tell the composition of the movement from polls like this, but we can see the net movement that you’ve described. Undoubtedly people moved from the Coalition to the ALP, from the ALP to the Greens and some straight from the Coalition to the Greens – but we cant figure out the numbers of each unless we get either longitudinal polling or pollsters start asking the question “Who did you vote for last election”.

    One would think that the Greens have better chances at State elections than Fed elections for grabbing Lower House seats, if for no other reason than the seats are smaller and tend to be more focused around particular communities than do their Federal peers, but it comes back to that big problem of herding enough green voters into a given geographic area.

    Preference allocations of the major parties and exhausted preferences by voters in States with Optional Preferential Voting are also an issue as the Greens might not even be able to win seats if they come second. If HTV cards of the majors are running against them in a given seat it basically knocks them out of the game.There’s also the question of whether people who voted for the major party that came third (assuming the Greens came second) would actually distribute their preferences if its OPV.

  11. Sir Ian Upton said

    The 2008 Gippsland By-election final figures don’t seem to support the Green swing. According to the AEC, Dr Malcolm McKelvie for The Greens secured about 7%. McKelvie was a star candidate with an impeccable pedigree. Dr McKelvie listed an interest in China as well as studying Mandarin (someone for Kev 747 to converse with) and he plans to import electric-assisted bikes from China. Dr McKelvie was pitted against one candidate who was listed as a roué, thespian, writer, producer, director and co-ordinator of cultural events. One other candidate had an interesting CV with the high point being his role with ‘Champions of the Bush’. Rounding out the field of hopefuls was a candidate whose activities included being a professional volunteer as well as a basketball coach and player.

    Notwithstanding this woebegone collection of life’s detritus the good doctor only attracted about 7%. The question must be: Just how many times can Bob Brown respond to surveys before he is ruled ineligible?

  12. Yaz said

    Hey, Sir Upton, I think the Liberal candidate only got slightly more than 7% in the seat of Brunswick in the last State election. I don’t really know if they’ll survive as a party…
    Then again, perhaps it’s all about context. Nonetheless, I am glad you appreciated the stirling qualities of the Greens candidate!

  13. Possum Comitatus said

    You raise a very good point Sir Ian – I’ve made an Update to crunch some numbers through it.

  14. Eratosthanes said

    “Who else thinks something dodgy was going on in the sampling for the Greens to jump from around 4% in mid to late 2007 to 12% last month?”

    I don’t think this is dodgy at all. Antihowardism was unifying the left under the ALP brand. I made a prediction in mid 2007 that the green vote would decrease leading into the election and then blow out afterwards. There are an enormous number of people who were happy to vote for the ALP to make sure Howard was removed. But once he was gone they jumped straight to the Greens who better reflect their views.

  15. fred said

    Hey Possum, could you give us your comments on the methodology/accuracy/credibility whatever of Essential Research/Source Watch?
    Pretty please….with sugar on top of fruit.

  16. gandhi said

    the Green vote in Rural electorates is 2.29% below their average national vote.

    I don’t think that is at all suprising, given that the rural voters are most likely to identify Greens as dirty tree-huggers etc. And I suspect that could be down to very deliberate Coalition “special education” classes in the bush.

    BTW I am NOT a Green!

    I just find myself arguing their points quite voraciously in the current climate (and I voted for them at the last 4 elections – was that naughty?).

  17. Possum Comitatus said

    Fred, glad you asked!

    Essential Research use online surveys, but have recruited a survey pool of over 100,000 people with most apparently recruited using non-online means, which I’d imagine were mail outs and phone calls to get them to join that pool. They send out a weekly survey to about 8000 people out of this pool and then weight or otherwise filter the answers they receive by age, sex, income and the usual demographic categories to obtain a bunch of survey answers that are representative of the Australian population as a whole.

    However, they have a major flaw when it comes to political polling, or at least had one at the end of June which was the last time I looked.

    If you head over to Poll Bludger and download the Essential Media release of 23rd June:

    http://www.pollbludger.com/EssentialReport_230608.pdf

    And then turn to page 9 of that document, you can see their demographic breakdown by age.

    The big problem here is that there were no people over the age of 65 in their sample (probably because they’ve found it difficult to recruit that age group for online surveys like most organisations). But you also might notice that they’ve “overweighted” the 50-65 year age group to make up for the lack of the over 65’s.

    The big, big, big problem with doing this is that the over 65’s are the Coalitions strongest demographic of support and the only demographic that the Coalition can be said to dominate politically. The 50-65 are Boomers and vote Labor far more than the over 65 group, yet Essential Research simply takes those Boomers, overweights them and uses them to make up for the shortfall of the over 65s in the sample, skewing the political polling results toward Labor by about 2% on the primary vote and 3% on the TPP by my calculations.

    Boomers arent pre-boomers, they vote differently – you cant pretend that they’re the same because it’s convenient, which is what seems to be happening here.

    They need to stand in the naughty corner for a bit.

  18. fred said

    Ta Poss, what do you reckon are the implications for the responses they give to less party orientated questions, the ‘need for climate change action’ stuff for example.
    Also a little distorted or pretty close to the mark?

  19. Antony Green said

    Oh boy Possum, if you’ve been as involved as I have over the years in how the AEC classifies electorates into those four categories, you wouldn’t use them as input to an equation. The method they use is way way dodgy, and it’s a categorisation that goes back to explaining Whitlam’s victory in 1972. David Kemp did an entire book on this issue of the old inner-city working class seats versus the outer suburbs.

    I mean, the AEC classify both Newcastle and Moncrieff as Provincial on the basis they are not Capital Cities and they are not Rural. Another example, as an analyser of elections, I see no reason to differentiate between Robertson and Dobell, which the AEC calls provincial, and Lindsay which they call Outer Metropolitan. Both get all their media from Sydney, and both are largely dormitary suburbs of Sydney and both are about the same distance and they have the same demography. I’ve been doing elections for 15 years completely ignoring the AEC’s classification because they can’t give you a sensible reason for why they classify electorates as they do.

    If I were you, I’d just split electorate into being within 10kms of the GPO versus the rest and you’ll explain the Green vote. You’ll explain even more if you add Newcastle and Fremantle to the inner city category. And if you split rural seats into those along the coast and those inland, the Green vote is entirely different, especially in NSW and Victoria.

  20. Antony Green said

    Another e.g., the Queensland seat of Fisher is still classed as rural. Yes, it was rural in the later 1980s, and it still has a rural component, but it’s based on Caloundra and Maroochydore. Not exactly rural. Leichhardt get’s classed as rural. It’s basically Cairns and a lot of Indigenous communities. so why is it rural?

  21. Ningaui said

    I hadn’t known the green responses in the polls were up but had assumed they would be. There are some likely reasons that the Green vote would be up since the last fed election and will hold that way for a while.
    (1) As labour maintains itself as a rightish-centrist party then there will be a bit more room on the left for the greens to fill. (2) There has to be some homeless ex dems who wander over to the greens. (3) There would be a certain number of people previously not green enough to vote green but who are sufficiently alarmed to shift their vote for climate change reasons. These are a new lot on the block. Not sure how many there are but I would count them as shifting their voting paradigms. (4) There are probably a few discombobulated coalition voters who just don’t get Blind Eye and his careerist cronies, (or any of the state lib rabbles) and who would never land on labour. So there are some homeless coalition voters, some of whom would drift to the greens, probably only temporarily. (5) There are also some disenchanted, small but specialist groups who had great hopes of Labour in the budget and elsewhere. These have been disappointed by getting the reverse (ie cost cuts in their favourite program bits, whale harvesting continues – shock horror, losing solar heater rebates) of what they had hoped for. (6) the environment is simply getting worse – and the MDB water deal would have disenchanted a few more. As a general assumption I suppose I would say that the worse the environment gets the more likely it is that people will vote green, particularly if they get to thinking that the major parties are part of the problem. I suspect that babyboomer children would quite naturally go that way, coz they know that boomers are selfish and have stuffed it up for them (7) the greens have not been able to exercise power so they have not been able to piss anybody off.
    The libs as a rightish-centrist party had the same sort of tension with more conservative elements wanting somewhere else to go, like to Pauline H. There are no current pressing reasons why people who have actually voted green would stop doing so. If the Greens mis-exert some balance of power in the Senate that might change. So, all-in-all, it is fairly logical that the green vote in the polls has gone up.
    On the classification of seats, Gippsland is rural if you count the lots of small towns, farmers, sawmillers and commercial fishers; provincial if you count Sale and maybe Bairnsdale; tourist/service delivery/sea changers and treechangers if you count the lakes tourist centres, Lakes Entrance and some of the more pleasant forested hill country; and heavy industrial brown coal urban if you count the bits in the Latrobe Valley as well as the oily and gassy bits from Bass Strait that happen in the electorate. If it was a listed company it would be a conglomorate that makes no commercial sense. Mutual incomprehension is the main thing the denizens have in common.

  22. Harry "Snapper" Organs said

    Many thanks, Possum and Antony. You guys! I’m particularly pleased with Possum’s explanations about the Essential mob’s sampling. Antony, can you apply any light on how the AEC decide on categories of electorate and whether they have any processes in place to revise these? Is there a website I should visit?

  23. Possum Comitatus said

    Antony – that was the first time I think I’ve ever used the AEC classifications.From what you’ve just said I think it will be my last!

    Thanks for pointing that out.

    Those classifications seem to generally hold for explaining some of the Greens vote because as you say, it’s basically a half assed substitute for distance from civilisation, but I just tested them on the primary votes of the majors over the last 5 elections and it’s pretty much useless.

    After building quick correlation and covariance matrices for those seat classifications on that primary vote of the majors through time, there’s more unexplained movement within a seat classification group than there usually is between seat classification groups.

    Have you ever tried to build a more realistic seat classification system or know of anyone that has?

    I knew there was a reason I stuck to the census data – ecological fallacy is a far easier problem to deal with than completely meaningless categories:mrgreen:

  24. Possum Comitatus said

    Fred, since Essential Research look to have a systemic bias in their sampling frame, any of their other political questions would probably have a higher level of support for positions more generally held by Labor voters than Coalition voters by about the same margin. So those climate change ones are probably a few percent more favorable on the “need for climate change” compared to where the Australian population actually sits. It’s not by great chunks, just a few percent.

    It mostly matters with the voting intention questions as a few percent makes all the difference, well, unless we are talking about preferred PM or something:mrgreen:

  25. Antony Green said

    Possum, the AEC seat classifications are greatly loved by lobby groups, especially rural lobby groups, who lump all the rural and provincial seats together and argue how important their industry group is. The fact their managing to include Gosford, Newcastle, Geelong the Gold and Sunshine Coasts in the category seems to be something they prefer to ignore.

    The biggest problem in any analysis of voting based on demography is that the problem is multivariate, but there is so much colinearity between the variables that why you get a model that explains a lot of the variance, but the explanatory power of each individual variable becomes negligible. People like Andrew Leigh find data sets where they can use one variable in the model on a reasonable assumption that the rest of the variance is randomly distributed in the error, but this only ever lets you explain a little bit of the problem. Everyone gets excited because we get a number for the donkey vote, but that only explains a small though constant part of the variability in swing.

    For me still the most useful paper I have seen exlaining Australian voting patterns was published in the late 70s, maybe early 80s, and I think done by Ian McAllster and Jonathan Kelly. (I’d have to dig up my uni thesis to remember the exact reference.) It used the old-fashioned technique of factor analysis to group all the demographic variables together into factors.

    My stats knowledge is not deep on the technical side, but it seems no one likes factor analysis in the social sciences these days, and I don’t think its just because it was invented by eugenecists. But that paper came up with three factors that best explained voting patterns in Australia, and they were a class variable, a city/country variable, and another which was an amalgam of family structure and religosity. It’s still a model that in my mind best explains traditional swing patterns in Australia, thought he rise of the Greens and inner-city gentrification has altered they way the class characteristics of an electorate can be assessed.

    But it is interesting how one variable can explain one part of a political equation but not another. Inner city explains the Greens, but tells you nothing about Labor versus Liberal. So for instance, inner Sydney seats like Wentworth, Sydney, Grayndler and North Sydney all record very high Green votes. Distance from the GPO does not explain the Green vote, but it is a de-facto variable for rates of tertiary education, participation in the knowledge economy, rate of female participation in the workforce, low numbers of couples with children and willingness to live in flats or town houses and to use public transport.

    But those same four electorate have nothing in common when it comes to Labor versus Libeal voting. Ahhhh, such are the joys of modelling complex real world variables.

  26. Antony Green said

    Ughh! I’ve got a their instead of a they’re in the first paragraph. Minus 10 points!

  27. Antony Green said

    Harry, the AEC used to decide on classification by the local Returning Officer being given the list of four categories and asked which of the four their electorate fitted in to. About 10 years ago they centrally re-categorised them all (I know the person who did it) but it looks like they have let them drift again.

    The only categories that ever made sense in explaining traditional voting patterns were Rural and Outer Metropolitan. You can classify electorate by a cut off point of how much agriculture takes place. You can look at the proportion of an electorate that is buying new housing stock. But the problem is Federal electorates are so large and complex that any simple classification is by definition going to over-simplify the complexity of electorates.

  28. Sir Ian Upton said

    Possum I still think someone has his finger on the scales with regard to The Greens polling 11%. Kev747 is the one who has been busy announcing the big ticket items (Kyoto, Sorry, Petrol Commissioner, whale meat ban at all official gov’t bar-b-qs) so a spike in The Greens popularity is hard to understand. I would think that The Greens are as popular as a pigskin wallet in a synagogue.

  29. Labor Outsider said

    Possum

    I agree with Antony that it is rather difficult to interpret the regression you ran with AEC classification. It would be a much larger task, but you would be much better off making use of data from the census. The big problem there is allocating data to electorates, but you could make a first pass using the data that George Megalogenis put together for the last election. That data set included information in income, debt burdens, ethnic composition, age distribution, education – if you added to that simple stuff like distance from cbd, or the ABS remoteness and accessibility index, as well as state and coastal dummies, and perhaps average house prices, you would have a more satisfactory model….

    I ran some simple regressions last year, with the coalition 2pp margin at the 2004 election as the LHS variable and was able to explain around 60 per cent of the variation in the coalition’s 2pp margin….highly significant variables included: average household income, averge mortgage burden and proportion of homes that were “McMansions” (all positively related to the coalition vote) – and proportion of the electorate aged 18-35, proportion of the electorate from a NESB (negatively correlated with the coalition vote). As I say, pretty simple stuff at this stage, but you could easily make make the analysis more sophisticated…

  30. Antony Green said

    Labor outsider, I’m not surprised that throwing that many variables at the problem explained 60% of the variance. After all, your chucking at the problem a whole array of variables that we know define social status in Australian society, and if there is no relationship between social class and voting patterns, I think I’ll give up my job. But your analysing the wrong independent variable. You should be trying to explain 2PP swing, not 2PP margin. You don’t have to do flash statistical analysis to explain 2PP margin in Australia, just go for a drive through a few seats. But why at a particular election did the swing vary between electorates? In the long run, 2PP margin reflects the underlying social margin, but it is the pattern of swing that measures change.

    The other problem is, with so many variables, how do you know which one is more important? If you analyse the 1998 election, the one electorate demographic catagory that was critical to Howard being re-elected was the mortgage belt. National 5% swing, much higher in many seats, but the Lindsays, Makins etc of this world didn’t swing. If you have a model with too many variables, trimming the result down to the key variables becomes difficult. It is true, if one variable becomes more important, the statistical tests on other variables will point to them being less significant to the point where they can almost be excluded.

    As an example, the Australian Election Study surveys of the 1996 election were interesting because the occupational class variable became irrelevant. Hence a few end of class analyses of that election. But occupational class came roaring back in 1998.

    Being an r-squared fetishist may be useful in modelling exhange rates, bond yields and stock market indicies, because you are just trying to pick the next measure, but in the social sciences, you have to understand what your model means and its limitations, not just come up with a model that fits to past data.

  31. Labor Outsider said

    Just a few more points:

    – industry employment share is also useful – agriculture is a good predictor of the coalition vote – manufacturing a good predictor of the ALP vote

    – i’ve seen comments that income isn’t related to voting patterns because the country’s poorest electorates vote coalition (usually national). If one uses a simple rural dummy, or agricultural share of employment variable, income becomes a very good predictor of the labor/coalition split

    – the variables I identified in the previous post do a better job of explaining the 2pp margins in 2004 than they do the 2pp swings in 2004

  32. Possum Comitatus said

    Antony,

    You can see why the lobby groups love it, one only has to take a squiz at Table 2:

    http://www.aph.gov.au/library/pubs/RB/2006-07/07rb01.htm
    (Distribution of businesses by industry and Commonwealth electoral division from 2004)

    …to see some fortuitous stats for for lobbying firms.Adjust some of those industry distributions by removing the Central Coast, Newcastle, the Gold Coast etc and the cupboard becomes a bit bare for a lot of those industries and their “doin’ it for the bush” spiel.

    If that piece of analysis is ever redone using 2006/7/8 data – a panel data approach might yield some interesting Workchoices effects since we now know the proportion of AWA’s deployed by a lot of those industry types and where.

    Thanks for the tip on the McAllster and Kelly paper, I’ll try and hunt it down.

    I might start using more panel data, since it allows for trasformations of the data combined with differencing to remove a lot of the collinearity among variables. Npt all, but at least enough that the data sets become more usable.

    LO,
    I’ve actually built a database of all of the 2006 Census data, most of the 2001 data (and a fair bit of other stuff) that’s been released and which can be aggregated into Commonwealth divisions. Ideally I’d like to eventually have a database that I can do something like nearest neighbour analysis on based on a whole heap of demographic variables, which would be handy when we get polling that focuses on specific seats or groups of seats.

    Unfortunate as it is, the most useful purpose for this type of stuff isn’t actually predicting or even fully explaining voter movement, but simply debunking narratives and demonstrating what isn’t or cannot be happening.

  33. Possum Comitatus said

    Another big problem with trying to analyse who did what in the 98 election is that an awful lot of stuff just got washed out and simply overwhelmed by the One Nation effects.

    One Nation voteshare correlated enormously with things like Year 10 being the highest level of schooling, welfare outlays by electorate as a proportion of total electorate income and the number of males in blue collar work in each electorate apart from the obvious things like geography.

    But the preference flows from One Nation to the majors were all over the place, to the point where they were almost random in some seats that had a history of behaving similarly, and have behaved similarly since.

    The 1998 election completely screws up nearly every piece of long term time series analysis you use, from polling to election results. The only way around it for polling is to whack a great big dummy variable over the period, but for election results it’s just a nightmare.

    The least Pauline Hanson could have done was to apologise to the nerdy types for making their analytical life difficult:mrgreen:

  34. Antony Green said

    One Nation in 1998 shows the huge problem in using the wrong assumptions to analyse data. The Coalition (and especially the media) misinterpeted One Nation’s support because of the demographics you referred to. In survey data, One Nation voters looked like Labor voters. Survey data in 1997 and 1998 always showed One Nation gains at the expense of the Coalition with little impact on 2PP. But there was too much focus on the 2PP result.

    But urban One Nation supporters were what the English call ‘working class tories’. These people probably weren’t Labor voters, as the 1998 Queensland election proved. That result caused the Liberal Party to switch from avoiding talking about Hanson to directly attacking her, and switch to talking about an economic issue in tax reform rather than banging on endlessly about the Wik bill. Queensland Premier Rob Borbidge payed the sacrifice but taught a lesson that the Howard government learnt very quickly.

    At the 1998 Queensland election, One Nation polled well in all the safe Labor seats where the National Party had finished second to Labor ahead of the Liberals in the 1980s. At the 1999 NSW election, the best One Nation urban vote was in safe Labor seats in Western Sydney, but at the expense of the Liberal Party not Labor. In both cases, Coalition advance against Labor in safe Labor seats brought few rewards. In Queensland, those rewards were of little use when it resulted in the National’s base vote in its own safe seats being shredded.

    Murray Goot has written the best analysis of One Nation. In the survey work, One Nation voters look like Labor’s working class base. Except on one question, attititude to migrants. Labor surveying at the time also noted a difference based on traditional voter allegiance. Neither of these issues show up in census data, which is why your census analysis gets stuffed up by One Nation.

    One Nation is a classic example of what happens when a new cleavage opens up in politics. Australia has three traditional cleavages that interplay. Two still exist, Class and City versus Country. The third traditional one which is greatly diminished is Catholic versus Protestant, though you can still see shadows of it when you look closely enough at the data.

    One Nation cleaved across these divides by opening immigration to political debate, but the major parties are wary of touching that issue because it is so divisive. It’s like Canadian politics, where the Anglo-French divide is clearly the greatest cleavage in society, but one that risks splitting the nation if brought to the centre of debate. The debate becomes one of how you debate the difficult issue, not the issue itself.

    The 2001 Australian election was interesting because the immigration cleavage was opened as a defence issue and was therefore seen as more legitimate. But given the Howard government was only narrowly re-elected in 2001, no one should assume that the way the issue was raised that year was anything other than a high risk strategy by the goverment. A great ‘if’ question is, would the boat-people issue have played as well and as long as it did if the great externality of September 11 had not occurred? But it did happen, so all you can do is speculate.

  35. Labor Outsider said

    Antony

    I don’t agree that the 2pp margin isn’t interesting. Its analagous to a stock, and I always find that it is useful to understand the stock before moving on to explaining the flow. It is a complement to analysis of the swings, not a substitute.

    I also don’t agree that statistical analysis of the 2pp margin can be substituted for by just driving through a few seats. If a particular phenomenon (voting behaviour) has multiple causes, statistical analaysis helps to tease out their relative contributions. I used the r2 just to give the impression that the model didn’t do a bad job, not because i have an r2 fetish. Modelling voting behaviour is, I think inherently difficult – and all attempts to model it will have its limitations. But there is no reason to think that a model with a small number of variables will be more enlightening than a model with a large number of variables. The idea is to be parsimonius – that could be 5 variables, or 10 variabls – but if all 10 variables are highly statistically significant, with only a weak collinearity, I wouldn’t exclude some of them just to reduce the size of the model. That leads to an omitted variable bias. The model that explained 60 per cent of the variation had only 10 variables – I would’t have thought that was particularly large.

    There are lots of ways to gauge which variables are most important. First, the size of the cofficient will tell you whether you need large or changes in the RHS variable to explain marginal changes in the LHS variable. That is the economic or social significance of the variable. Tests of the variable’s signifcance is useful, as long as you don’t have a multi-collinearity problems. If you do, then joint tests are better. You can also drop each variable from the regression and examine what happens to the overall fit, and whether the coefficients on other variables change much when the variable is excluded. Exmanining covariance matrices also helps. I’ve undertaken quite a lot analysis of micro household data sets, so I have a fair idea of how to go about it. Economists take identification very seriously – which is why we make use of things like IV, randomised trials, regression discontinuity, etc. I certainly wasn’t claiming that a model I whacked together in 30 minutes of the 2004 2pp margins was particularly rigorous!!!

    Effective statistical modelling is as much an art as a science. Practictioners have to understand that the variables they are using are often not picking up exogenous variation in that variable, and that the modelling can be sensitive to alternative specifications. That is why peer-reviewed papers should undertake rigorous sensitivity analysis.

    I certainly don’t doubt that some factors become more or less important from election to election. But is that so surprising when the relative important of the underlying issues can change so much from election to election?

    If I were studying elections in as deep a way as you do, I would never use cross-sectional results from one election to make strong claims about long-run changes in the variables that affect voting behaviour.

  36. Labor Outsider said

    Just in response to the observations about one-nation…

    Another way of framing Antony’s comments – is that analyses of the 1998 election suffered from a large omitted variable bias – support for one-nation was correlated with demographic variables, but a more important underlying factor was attitude to migrants. Another interesting question is whether attitude to migrants can itself be explained by other variables.

    Unfortunately, without a database that tracks individuals over time – their votes, their demographic characteristics, their attitudes, parental variables, etc – our understanding voting behaviour will always be limited.

  37. Possum Comitatus said

    heh!

    LO at 35,

    This is all very deja vu-ish! It’s not irregular that Antony and I descend into exactly the same debate!

    The econometric approach and the social science statistical approach are quite different in the way each group attacks a problem. We focus more on how, the social science side more on why. We also treat a lot of little things vastly different – constants in a regression being a case in point. For us it’s not a big issue to treat it as pretty much irrelevant if it’s nonsensical for the independent variables to ever have a value of zero, but constants are often a big thing in the social science stats field. So we end up speccing out models that accommodate those two positions and which end up entirely different as a result. We both treat uncertainty differently, we treat parsimony differently, econometricians are willing to be a fair bit more experimental with transforming the data to find nuance in relationships (which comes back to our differing expectations and treatment of uncertainty)

    Neither side is right or wrong here – we just come at the same problems from vastly different angles.

    And let’s face it, social scientists just don’t like econometricians!

    They think we’re ostentatious and overpaid!:mrgreen:

  38. Possum Comitatus said

    Another thing on One Nation was their unusual approach with HTV cards, what they directed their preferences to, where, and how many people followed it. Like so many things with One Nation, it too was a bit all over the shop.

    One Nation effects were so complicated that I dont think we’ll ever truly get to the bottom of it. It was like the party was just a vehicle for a whole grab-bag of grievences that were often incompatible with each other.

    For instance, there were quite a few ordinarily Greens voters on the NSW North Coast that switched to One Nation, but similarly One Nation carved a big chunk out of the conservative Nats in the same seats – two groups that on any given issue usually want to beat each other to death with lumps of 4×2’s.

    There were libertarians siding with agrarian socialists, racist pigs siding with very polite and social justice oriented little old ladies and the drown a dolebludger crowd snuggling up with rural unemployment support networks!

    Fark me dead!

  39. Antony Green said

    Labor outsider, I was being slightly facetious about driving through electorates. It was more a comment that there are papers going back to the 1960s comparing voting patterns to census data so nothing you wrote surprises me. I was going to do an honours thesis doing exactly that so read most of the papers. I didn’t do it in the end as it was the 1980s and getting the data sets into a format you could analyse was vastly more complex than it is today and I would have spent half the year just organising the data rather than fitting it into an analytical framework.

    Thankfully models are now a little more statistically sophisticated than the first British papers that found the two census variables with greatest explanatory power for Labour voting were the proportion of an electorate with access to to a private rather than a shared bathroom, and the proportion with access to an indoor as opposed to outdoor toilet. Such is the way that obscure census questions reveal the nature of class. Also interesting to note how rising levels of wealth have changed housing stock.

    As for your other comment, the problem is trying to get variables that don’t have a lot of collinearity. That gets harder the more variables you feed in. And I’m aware of the statistical techniques where you test the robustness of a model by inserting and removing variables. But we know average electorate level of income is related to proportion of people employed in agriculture, proportion aged over 65, proportion employed in mining, proportion with a university degree, proportion with a trade, but if you fed them all in to standardise the importance of income on its own in determining vote, you’d get completely snowed by the colinearity.

    I’m glad Possum raised the problem of ecological fallacy in relation to census data, because too many people with access to stats packages completely ignore it. I’m one who believes that students should, as I was, be forced to read the papers from 1950s that first explained the problem.

  40. Antony Green said

    Possum, on One nation preferences, who ever knew what was handed out? Just because they published a how to vote doesn’t mean anybody ever got it. Locals may even have published something entirely different. Other parties may have also polluted the preferences flows by issuing a competing how-to-vote cards. I’ve never seen anything that worked as a pattern for One nation preferences, so my view is that something was going on at the coal face of polling booths to which we’re not privy. I prefer that to the alternative hypothesis that One Nation voters had no idea what they were doing. But that’s my small ‘l’ liberal bias that if individuals can’t be trusted to be the judge of their own self-interest, who can be?

    Read about Rutherford’s Geiger-Marsden experiment that proved statistically that atoms had only a tiny nucleus. If only the social sciences produced such clear cut results.

  41. Possum Comitatus said

    I voted in Cowper during that election, and the booth I went to early in the morning, a relatively smaller booth, had about 4 One Nation people handing out HTVs. Come lunch time I took my better half to vote at a different booth, a much larger booth, and there wasnt a One Nation HTV card to be seen!

    I thought it was pretty funny at the time.

    Not as funny as the skin head, built like a brick shithouse complete with various anti-certain-people tattoos on his arm, handing out Greens HTV cards at a State election in a rural booth in NSW once that I saw, but pretty close!

    The bloke was actually a nice guy – a reformed miscreant that kept his tats like a weird penance, but must have been a shock to others that didn’t know him.

    Like you, I think a lot of the One Nation preference flows were just down to organisational incompetence and miscellaneous funny business, but also because of the density and distribution of various grievance groups that voted One Nation for all sorts of different reasons.

    Booths that had normally strong Coalition votes because of a large ageing demographic, but got a largish One Nation vote would be expected to push their preferences back to the Coalition – especially if their One Nation vote was driven by an anxiety about the rapid change that the country had just been through and seemed to be continuing to go through. But on the other hand, those blue-collar “Howard battler” type over cliched booths that voted One Nation in good numbers would have been expected to deliver some preferences back to Labor, if for nothing else than over the GST and as a reaction to Reith’s workplace relations reform.

    But for every booth where that happened, there was another where an entirely different set of local circumstances were occurring, on top of the HTV card fiasco.

    Another thing worth considering is that there were quite a few One Nation candidates that had a fairly large local community profile and in some places, particularly regional areas, would have attracted a personal vote that came from both sides of politics. Those preferences would have flowed in all sorts of ways.

    It’s all very messy.

  42. Antony Green said

    You may be interested to know Possum that I am appearing before the Joint Standing Committee on Electoral Matters next week, mainly to talk about informal voting. My preference would be to bring in Optional Preferential Voting, but I don’t think they’ll buy that. So my next preferred position is to argue for a change to the formality rules so that votes with incomplete or incorrect preferences be allowed to remain in the count so long as the preferences on the ballot paper do not need to be counted. Doesn’t help the minor parties much as all their votes will still need to have complete preferences, but it at least puts a big dent in the informal vote, down 30-50% based on data published by the South Australian Electoral Office.

  43. Labor Outsider said

    Antony

    Perhaps those variable would be affected by collinearity, but perhaps not. In the labour market research I’ve seen, many of the variables you mention are independently and statistically significantly related to regional labour market variables. I see no reason why the same might not be true for voting behaviour. The idea of a model is to control for the effect of correlated variables – ie, the effect of the age distribution on the margin/swing once controlling for income, etc. Collinearity has to be pretty high to make that problematic and should be assessed on a case by case basis.

    All that said, I am a BIG fan of your approach to electoral analysis – election night wouldn’t be the same without you!!

  44. Antony Green said

    Labor Outsider, you’re making a really fundemental error. Your census based model explains level of Labor or Coaltion voting in electorates, NOT voter behaviour. With a well constructed ecological model, you can make an assumption about the individual based on aggregate statistics, but you have to be very careful.

    Your census data is modelling the % Labor vote against a range of % variables, and for various clustering reasons, you get a stronger correlation than you would get with survey data. If you do the same things with survey data, you’re lucky to ever get an r-squared above 0.3, because you either have numeric data such as income, or ordinal data like living in a McMansion or not, and your trying to explain whether an individual votes Labor, Liberal or Green. You always get a much weaker relationship when you explain individual voter behaviour than when you try and explain the ratios of voting for electorates.

  45. Antony Green said

    Possum, the article I was referring to was “Contextual Characteristics of Australian Federal Electorates”, Ian MacAllister and Jonathan Kelley, A&NZ Journal of Sociology Vol 19 No 1, March 1983. It builds three factors from Census variables, Socio-economic status, Familism, and Rural-Urban. At a quick glance now I’m home, it’s a bit dated and less relevant than I thought, but but I still find that three factor way thinking of electorate behaviour useful.

    The paper also adds three migrant variables, Mediteranean, Eastern Europe and Northern Europe. Australia was much less ethnically mixed then, and historically there was a big difference between pre-1970 non-Commonwealth migrants (the ‘Balts’ as all the strongly anti-Communist Eastern Europeans who came to Australia were called) and the more mixed migrants that followed. You’d approach migrant variables very differently now.

    It’s still funny to dig up studies from around 1970 that highlighted the Labor Party’s failure to have any appeal to migrants. They’ve acted to address that problem since.

    I like the four census variables used to define Familism. % dwellings with sole use of a kitchen, % dwellings with sole use of bathroom, % of population purchasing dwelling, % dwellings with one or more homes. That was the 1976 census. Not sure the first two are even collected these days, and the cars variable would be two or more these days.

  46. While I would love to believe all the rise in the Greens polling is a result of the factors mentioned above, part of it is also a change in the way Newspoll asked the question. This occurred two or three polls before the federal election and contributed to the very sharp jump you see around that time.

    Its not the whole story however, possibly not even half of it. We see a smaller rise in Morgan and the Fairfax polling, and I’d attribute that to the factors people have discussed above, particularly the absence of the feeling that the most direct method to kick Howard out was to vote 1 Labor.

    Regarding One Nation preferences, I heard a rumour that in some seats their cards were printed on both sides of the paper, with preferences to Labor on one side and Coalition on the other. There are stories of Labor HTVers trying to persuade (possibly even bribe) One Nation people to hand out the cards with the pro-Labor side face up. I don’t know if these are true, but if they are you could pretty much put the variation down to that alone, particularly since some booths might have gone one way in the morning and one in the afternoon.

  47. Harry "Snapper" Organs said

    Many thanks, Antony, for you’re response to my question. Doesn’t sound very impartial. This has been a most informative thread, Possum. You deserve nice fruit, say, a nice red and a cigar!

  48. Possum Comitatus said

    heh!

  49. Possum Comitatus said

    Antony,

    Looking at that three factor approach, it’s basically an income/culture/geography process.

    Unfortunately we dont get those cracker questions like sole use of the bathroom (thinking about it, I cant remember any quirky questions in the last Census that would even come close to that – amazing how times change so relatively quickly), and with culture and family architecture (for lack of a better word) sort of splintering into more granular groups over the last 30 years, something representing that cultural approach would be difficult to construct using the current census info.

    The geography side gets tricky because of the way home the Commonwealth electoral division footprints overlap both areas of high home ownership and areas of rental accommodation in the inner cities – which are very different groups of people.

    In the outer suburbs, there’s a bit of a rough socioeconomic split along house price lines to boot.

    Coastal and non-coastal would probably work for the regions. Life seems to have become a little more complicated and diverse since 1976!

    Thanks for the details on the paper, I’ll look it up next time I’m in the city.It reminds me of some of the early market research done with the old Gallup polls in Australia. I’ve actually got some old transcripts of mid 1960’s product testing done in focus groups in Australia somewhere – some of that stuff is an absolute hoot looking at it through today’s prism.

    Vance Packard and the likes must be rolling around in their graves.:mrgreen:

    Bring back the quirky questions I say!

  50. Possum Comitatus said

    I also like your idea on informality only applying before preference exhaustion. A 30-50% drop in the informal vote can only be described as a Good Thing. Too many people seem to treat the franchise as a blase thing these days, it would be nice to see the balance move back towards the voters right to be heard in the general case, rather than a system that seems to look for excuses to knock that right on the head at the earliest opportunity.

  51. Labor Outsider said

    Antony

    I wasn’t making an error, I know exactly what the regression is doing. I was using the term “voting behaviour”, because it was a simple way of describing the regression. Of course electoral vote shares are not the same thing as individual voting behaviour, except under very strict conditions. Apologies for speaking loosely.

    I don’t doubt that the r2 would fall if one had individual level survey data, combined with their primary votes. However, in cross-sectional regressions on individual or household level survey data, an r2 of 0.3 is usually considered quite good. Further, I would imagine the a survey based model’s ability to explain the variation in voting behaviour would depend on how comprehensive the survey questionnaire was. If, for example, it was as comprehensive as HILDA, you may explain more variation.

    Regardless, as has been said before, r2 is not the best judge of such a model. It is difficult to explain individual level behaviour. However, if one had a survey of say 10000 people, and HILDA data combined with actual voting behaviour, I would expect to find quite a large number of highly significant variables.

    Btw, with Commonwealth divisions so large, and hence masking a great deal of within electorate variation, has anyone made use of booth level information? If you pay the ABS enough, you can get census data at very low levels of aggregation. Instead of aggregating the census data up to electorate level, you could take booth level vote shares and aggregate up to SLAs, or CCDs, or something else that was more homogenous than the Commonwealth divisions.

    The anecdote about migrant voting patterns prior to the 1970s is an amusing one. I think it serves to remind us that the parties’ bases can change considerably over time. The best example of that I can think of is the blue state / red state divide in the US. Sometimes that is discussed as though it was fixed in time, when in truth there has been a great deal of flux as political parties responded differently to changing attitudes to things like civil rights. The transformation of the Democrats from a party strongest in the South prior to the 1960s to a party strongest in the north-east in recent decades is the most remarkable example. Did you know that FDR received over 95 per cent of the vote in southern states like Georgia in the 1930s?

  52. Labor Outsider said

    By the way Possum, is it possible to start a new thread on the politics of emissions trading? Over at LP, most of the bloggers seem to be working on the assumption that the goverment has got both the policy and politics of the ETS wrong. However, it strikes me that the Green Paper is a pretty clever document. It identifies the issues that will be difficult for the goverment – petrol, compensation, jobs in regional areas, carbon leakage – and steers a middle course in each area. It allows them to paint both the greens (they don’t care about your wallet or your jobs) and the opposition (they don’t care about the environment) as being more extreme. If worse came to worse and they couldn’t secure the passage of the legislation through the Senate – I think they would be in a pretty good position for a double dissolution, though I think the green share in the senate would increase…

    Thoughts?

  53. Antony Green said

    Labor Outsider, both the Labor and liberal parties use ABS data at CCD level against polling place results, having made reasonable assumptions about where people vote. They use mapping tools to display the data, as does the Parliamentary Library. When the parties combine the census data with access to the electoral rolls and past polls of every voter plus follow up polls and letters to all newly enrolled voters, every marginal seat holder has a very detailed picture of their electorate far in excess of anything you get out of the census.

    Its why I say if your going to do anything with census data, it’s more interesting to do it looking at swing, not 2-party preferred vote. I don’t know what a HILDA survey is, but there are several survey data sets around which investigate voter behaviour which have been matched up against census data in the past, none of which have produced anything hugely revelatory that we didn’t already know.

    The problem is, once you match up survey data with census data, all you discover is that elections tend to be decided by people who aren’t particularly committed to once side of politics or the other. Safe labor seats are full of committed Labor voters, Liberal seats are full of committed Liberal voters, and marginal seats are where the balance of committed Liberal and Labor voters is equal and so the swinging voters are more important.

    I’ll add a point to that, there are situations and also certain electorates where the pool of swinging voters may increase. The most obvious is the mortgage belt. Young families buying a house are at the point in the life cycle where established political partisanship is most likely to be mugged by present economic reality. Look at most elections where a change of government took place, and you’ll find the swing is greatest in seats that can be classed as the mortgage belt. Labor’s defeat in 1996 is a classic. Look at 1998 and 2004 and you see similar interesting patterns in the mortgage belt. To me, 2007 includes a correction to the 2004 result, as seats that swung strongly against Labor in 2004, especially around Brisbane, came savagely back to Labor in 2007.

    One former Labor Party secretary once said to me that he never thought anything useful could ever come out of the Australian Election Study surveys. In a sense he’s right, as the survey is looking at the whole Australian electorate, including all the committed voters. It’s a sociology exercise in explaining social structure. And I’d argue that census analysis of 2PP margin is doing the same thing. It looks at the stability of Australian voting patterns, not the change in voting patterns.

    The major partiers concentrate their survey work on people who may consider changing their vote. Their interested in a different problem, not explaing social structure, but trying to figure out what causes people to change or not change their vote. So if you’re examining voter behaviour, there are two different levels to examine the problem, one that essentially examines all voters and explains social structure, and the second which parties are interested in and examines why voters who change their vote decide why to change their vote.

    If a party moves into an electorate it has never previously worked in, it always starts with census profiles by CCD, and based on past knowledge of the way people vote, concentrates on areas where it is thought the number of swinging voters will be high. But if you want to know how a voter behaves, in the end the best way is always to ask them rather than ask a series of questions about age, occupation, salary and housing status, which is the limitation of census data. We have one set of data of election results, and another of census data, and while we can make assumptions about behaviour from those data sets, nothing ever beats surveys that actually ask about voting behaviour.

  54. web design company California…

    […]Newspoll, Pollytrack, Satisfaction Ratings and the Greens. « Possums Pollytics[…]…

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

 
%d bloggers like this: