Possums Pollytics

Politics, elections and piffle plinking

The informal vote

Posted by Possum Comitatus on January 24, 2008

Inspired by Mr Mumbles look into the nature and causes of informal voting at the last election (down at his January 15th entry), I thought we might break out the old stats box and have a little squiz ourselves.Peter got it dead right when he said:

Broadly, three things are likely to lead to, or are associated with, informal voting: optional preferential voting at the state level; high numbers of people from non English speaking backgrounds; and lots of candidates.”

For this, what we’ll do is regress the informal vote as a percentage of votes cast in each electorate against:

  1. The proportion of persons in the electorate who speak English not well or not at all as measured by the last census. We’ll call this variable NES.
  1. The number of candidates that stood in each electorate, we’ll call this variable, surprisingly, CANDIDATES.
  1. A dummy variable that has the value 1 for those States that have optional preferential voting at the State level, and a value of zero for those States that don’t. Qld and NSW have optional preferential, and SA has a strange ticket voting system that mimics the effects that OPV has in NSW and Qld on the informal vote (I tested it independently) – so we’ll classify SA as an OPV State as well. We’ll call this variable STATEOPV.

So running the regression on all 150 electorates we get:


The “C” is a constant that needs to be added into the equation (the value of which is calculated by the regression software mathematically) which tells us what the level of the informal vote would be should the other three variables equal zero. In this particular case, having zero candidates sounds a bit silly, but it’s a necessary procedure that gives us a statistical baseline to work from and only sometimes can the value of the Constant be taken and used literally. This is not one of those times.

So what this equation result tells us is that for every 1% increase in the proportion of people in the electorate that speak English not well or not at all, the informal vote goes up on average by 0.3%.

For every additional candidate on the ballot paper, the informal vote goes up by 0.2% and States with optional preferential voting at the State level experience, on average, an additional 1% increase in their informal vote compared to those states without OPV.

All these results are highly statistically significant, and collectively explain about 56% of the variation in the informal vote – which for cross sectional data like this is a pretty high level of real world explanatory power.

The NES variable is by far the most important, explaining about 34% of the variation of informal voting just by itself. If we run a scatter plot of the informal vote vs NES by electorate, we get:


To the naked eye that looks pretty linear, in that as NES increases, the informal vote proportionally increases – but for those of you that have stared at enough scatter plots, you might notice that there is actually a fair degree of non-linearity involved in the relationship.

As the proportion of people who speak English poorly or not at all in the electorate increases, there is a slightly disproportional increase in the informal vote.

After testing, if we use just NES as a variable against informal, NES explains about 34% of the variation in informal voting, but if we use the square of NES, that explanatory power jumps to 41%.

This seems to suggest that when an electorate contains a large community or communities of people with an NES background, rather than just clumps of people with NES background, informal voting really starts to jump. This is probably a good argument for the AEC to pester the new government for some extra money – electoral education programs in local community foreign language newspapers and media looks to be a good place to start if we want to lower the level of informal voting.

Similarly, after further testing, the number of candidates standing in an electorate doesn’t enjoy a linear relationship with the informal vote level either, it behaves similar to NES in that there is a disproportional increase in the informal vote with the more candidates you get standing in an electorate, although the statistical relationship is ever so slightly different in its shape.

So accounting for these slight non-linearities, our new equation becomes:


Which has increased the explanatory power up to over 60% simply by accounting for the disproportional effect that increasing numbers of candidates and increasing levels of NES have on the informal vote level.

Next post: some more on this informal vote, how decreasing the informal vote level would win the ALP more seats and where (so expect to see the AEC receive some new education program funding) and some funny relationships between party votes.


After some handy advice from Antony over SA state voting behaviour and a few other things, the above needs a bit of an adjustment. We’ll drop the CANDIDATES variable and simply use CANDIDATES^2 in isolation, as well as remove South Australia as a state that has an OPV effect polluting the informal vote in Federal elections in SA. There still seems to be something going on with the way the informal vote behaves as a function of the number of candidates – hopefully we can ferret out the causes of that underlying behaviour over the next few weeks. So the new working equation becomes:


AddThis Social Bookmark Button

add to kwoff


51 Responses to “The informal vote”

  1. kwoff.com said

    The informal vote « Possums Pollytics

    Inspired by Mr Mumbles look into the nature and causes of informal voting at the last election (down at his January 15th entry), I thought we might break out the old stats box and have a little squiz ourselves.Peter got it dead right when he said:


  2. Andos the Great said

    NSW gets 17 of the top 18 highest informal vote divisions. Nice work.

  3. steve_e said

    I tip that an AEC Education plan will get funding when other government spending is being cut. Further this education for voters in languages other than English can start well before the next election is due in 2011. Sampling of voting patterns at State Elections can be made to measure the success of the education program.

    While this will not be an Education Revolution it will get the support required because of the outcomes it will deliver (more seats).

  4. The Doctor said

    The other thing that may worth looking into is an NES*CANDIDATES variable.

  5. Possum Comitatus said

    Doc, I had a good look at using interacting variables like that, but was surprised to find that they had nothing significant to say when added to the variable list, and usually reduced overall explanatory power if substituted for NES and CANDIDATES variables (or groups thereof).

    I’m glad I wasn’t the only one that expected otherwise! It’s a bit of an odd result.

  6. dany le roux said

    Possum, thank you.

    One of the ad. genres which was flogged before the election was formally announced and which of course was funded by you and I was of the type which encouraged people to take up citizenship.It seems that many can become citizens without any English.

    Anticipating the conclusions of your next post, I think we all know why.

    Shouldn’t people be required to speak a bit of English as a requirement for citizenship?

  7. Will of Kooyong said


    Nice work as usual. I know it’s weird, but did you try to factor in the Tasmanian voting system? Obviously it could be too much to deal with, but it is a different form of voting for the Lower House in Tasmania and federally.

  8. Possum Comitatus said

    Will, just tested Tassie and OPV doesnt seem to affect their federal results. It’s a good point though with their HC voting system there. Maybe Tasmanians are just smarter than mainlanders and aren’t so easily confused :mrgreen:

  9. Will of Kooyong said

    Sweet. Keep up the good work. I should get around to doing some post election analysis with the Monte Carlo simulator I did about 10 days before the election. I kept running it while the votes were coming in, and it did come up with 83 seats. Of course I don’t have a blog to post it at. I want to try to factor in state swings with it all.

    As for Tassie, perhaps 2 heads are better than one. 😉

  10. Ron said

    Ask any ordinary aussie for Informal vote causes.

    He may say its simple:
    Some voters just don’t care , some are politically uneducated ,
    some are not smart , some are smart arses , some can not read.

    Mumbles chart shows 3 seats to support the ordinary aussie view
    ALL with average number of candidates:

    Lowe 5th highest non English voters -only 21st largest swing
    Gorton 4th highest non English voters -only 36th largest swing
    Bruce 7th highest non English voters -only 81st largest swing

  11. The number of candidates is often less important than the position of the main parties on the ballot paper and their how-to-vote order. The impact on a particular candidate’s informal vote can only be ascertained by an analysis of actual voting papers, based on the first voting preference where indicated by a number. Party scrutineers should have that information from close counts such as McKewen or Solomon.
    I scratch my head at decisions like Kalgoorlie where the ALP did not go straight down the ballot paper because of past concerns about giving One Nation or other far right parties high preferences even though they are never distributed.

  12. Antony Green said

    Just a correction on S.A. Their strange ticketing vote system does not mimic optional preferential voting. It is simply a savings provision for votes marked with only a 1. And we know it’s incidence, about 3% of all votes. That means the other 97% have full preferences. All it does is halve the informal vote compared to Federal voting. South Australia should definately NOT be included as an OPV state.

    All voting screens in SA have sample how-to-votes displayed. All have full lists of preferences. These displayed HTVs cannot have exhausted preferences. It is nothing like NSW or QLD.

    From ballot paper surveys, we know that well over 50% of ballot papers at NSW and QLD have incomplete sequences of preferences. In SA we also know the rate, about 3%. If your model is finding something special about SA when you include it, I am certain it is not finding an optional preferential voting factor.

  13. Antony Green said

    And by the way, I am sure you will find one of the reasons why informal vote goes up disproportionally with higher NES percentages is because they are normally ultra safe Labor seats. Try and find a Liberal or minor party voter handing out how-to votes in these seats. Even Labor doesn’t blanket these seats with how-to-votes. Less how-to-votes, higher informal voting. Australia is the only country in the world that insists you must number every square in an exact ordinal sequence, so it is no wonder that migrants can fail to understand this quaint Australian requirement.

  14. Antony Green said

    And I don’t like that model with the squared terms. It stuffs up you’re constant term. Any model like this should be accepting that there is a base level of informal voting that will always exist with the lowest values for independent variables and this should correspond to C. In your first model, that is about 1.2% which makes sense historically. But your second model gets a value 4.2%. And you get a negative term for candidates because you have a squared candidate term. Sure you get a slightly better fit for r-squared, but at the cost of a model with coeffecients that individually tell you less. Your negative Candidates coefficient has created a high value for the Constant which makes the model less useful as an analytical tool. Surely you shouldn’t put both Candidates and Candidates^2 in the same model. One or the other, but not both. Your model is trying to explain something, not just get a good fit.

  15. Antony Green said

    Actually, the problem is that no seats have 1,2 or 3 candidates, and few have 4. Until you get lots of candidates, the candidates term dominates over Candidates^2, which creates the high constant value as a nonsense high informal vote is created for non-existent seats with small number of candidates. If you are going to use that sort of model, uses (Candidates-3) or (Candidates-4), so you get a sensible constant. Again, you’ve created a second model that fits better, but it’s a model that I think produces nonsense values at the lower range. The second model r-squared means the interpolation of data has improved, but at the loss of an ability to extrapolate.

  16. Ron said

    Anthony Green has supplied the main cause.

    From ballot paper surveys, we know that well over 50% of ballot papers at NSW and QLD have incomplete sequences of preferences

    Presumably most of these did not have a HTV at all and assume a minority mis-transcribed the HTV.

    So to address over 50% of errors, why would not there be HTV’s in the AEC voting booth or should there be optional preferential voting with perhaps the Partys optional preernce code listed inside the polling booth

  17. Antony Green said

    Ron, you’ve taken that quote out of context, or perhaps I just didn’t make the point clear enough. At NSW and Queensland STATEe elections, we know an overwhelmeing majority of voters use optional preferential voting, so the majority of voters do not fill in all preferences. So understandably, when they come to Federal elections, some will make the mistake of voting as if it was also optional preferences. I made this point to argue that the South Australian ticket voting system is not OPV, because clearly from published data, South Australian voters fill in all preferences at state elections and it can’t be compared with the OPV experience in NSW and QLD.

  18. Does anyone have any thoughts on why the informal vote dropped in every state in 2007 compared with 2004?

  19. Should have included this link: http://vtr.aec.gov.au/HouseInformalByState-13745.htm

  20. apostrophe said

    Could it be that a lot of the explanatory power of the model depends on those four electorates with NES > 10% ?

  21. Possum Comitatus said


    (nice to see the Askimet anti-spam filter didnt stick you in the spam bin this time)

    I dont mean the SA system state voting system mimics OPV, just that it seems to mimic the effect in SA that OPV has on the informal Federal vote in QLD and NSW – although to a slightly lesser degree. If only 3% of the state SA ballots get effected, and the electoral commission advertising in SA state elections push a normal full preference vote – then it would clearly be something other than state ballot pollution affecting the federal informal vote level in SA.

    Thanks for that – I’ll fix it up. Having a quick squiz taking out SA in STATEOPV, it only marginally changes the coefficients on all variables except for STATEOPV which drops by about 0.1.

    On those safe seats, you’re right – not only does the informal go up in seats with high ALP primary votes, but also seems to statistically increase in seats with a high Liberal primary (although to a lesser extent).

    On the constant here – the constant in this equation has no literal meaning. You cant have a seat with zero candidates for instance, so C isnt included in the equation for any interpretive value it may have – it’s simply there to make the in-sample errors unbiased and allows the regression line to find its own level (if we removed the constant, the variance of the errors wouldnt change, but the mean of the errors would change, creating bias)

    On the candidates and candidates^2 issue, using CANDIDATES by itself doesnt quite wash out right, and although CANDIDATES^2 is much closer to what seems to be happening – there’s just something a little funny about it too (well to my eye there is anyway). Combining them together doesnt seem to work right either.

    I take your point – I’ll use Candidates^2 in isolation and ferret around a bit more to see if something else is playing out there that interacts with candidate numbers in terms of the informal vote level, or if something else altogether is going on.

    Many thanks.

  22. Possum Comitatus said

    Apostrophe – it also happens further down the NES scale, just not so pronounced. Those 4 big NES electorates do make it stand out to the naked eye though.

  23. mate said

    wow, what a excellent series of comments, didn’t understand a word of it but enjoyed reading all that anyway 😉

    As for Tassie, perhaps 2 heads are better than one.

    classic 😆

  24. Possum Comitatus said

    Peter – the drop is a strange one.

    I just ran a regression on the size of the change in the ALP TPP vote(04 to 07) against the change in the informal vote (04 to 07) by electorate, and there is a significant relationship there, in that the bigger the swing in 07, the greater the drop in the informal vote in that electorate in 07. However, its a pretty small number, only explains a few percent of the variation and the informal vote in 04 was based on the old 04 electoral boundaries.

    Do you have a list of the number of candidates that stood in each electorate for the 04 election (like you produced for the 07 election on January 15th)?

  25. Andos the Great said

    Informal votes cast as a protest in 2004 less than 2007? (Clearly there’s no empirical way of checking this)

  26. Andos the Great said

    Err, less in 2007 than 2004 is what I meant.

  27. Possum Comitatus said

    Andos – that’s what I was wondering. There’s a bit of evidence to support it but the measurable effect isnt very large . What effect there is (and we cant be sure anyway because of the electoral redistribution, so we are comparing 04 apples with 07 oranges in a number of seats) might also be explained by the drop in the size of NES effects in 07 compared to 04 – especially since there is a correlation between the ALP vote and NES over both periods.

  28. Possum: no, I don’t have those numbers at this stage alas.

    Re ‘protest’ informals, the AEC did study them last time, and attempted to separate the ‘deliberate’ ones from accidental. Hopefully will do so again.

  29. Russ said


    I voted in a booth with a very high NES population last year. I have never seen a line move so slowly, mostly because booth staff were having to explain to every person in the line how to vote, and what the forms were. You could save both staff and the expensive print/tv advertising campaigns if the AEC ran how-to-vote tutorials (or made a 10 minute how-to-vote video) at booths over the course of the day.

    [sorry Russ, you were in the spam bin….Poss ]

  30. Ron said

    Anthony Green I did quote your full sentence and assumed the sentence related to NSW & Q’ld FEDERAL results.Sorry mate

    However your stat of 50% of NSW & Q’ld ballot papers AEC surveyed having incorrect sequences is a critical stat 2 work on

    Alot of otherwise intelligent people have trouble with “NUMBERS”
    whether they are English speaking or not

    The problem seems to be NOT whether people are English speaking
    at all but the difficulty for the 4% whether English speaking or not of inserting numbers in sequence.
    This would need a System solution not an education solution

  31. Antony Green said


    According to the AEC’s informal voting research in 2004, the category with informals that increased most in 2004 was marks and writing, what is believed to be deliberate informals. When we see research for 2007, maybe this category came down and might explain this problem. Maybe an alienated part of the electorate felt less alienated this time.

    Another possibility that is difficult to measure is the simplification of how-to-votes. When One Nation was on the ballot papers, the major parties and the Greens went to some effort to put One Nation last. With very few One Nation candidates, many parties start to revert to sequences of numbers that were easier to copy.

    Parties are aware that complex how-to-votes increase the informal vote. In both 2004 and 2007, Warren Snowden in Lingiari had One Nation before the CLP as he opted for a straight down the ballot list of preferences. Lingiari has won of the highest rates of illiteracy in the country so a simple how-to-vote is viewed as a disadvantage.

  32. Antony Green said

    I’ve got to say Possum, I see no reason why you would use the model with the squared terms in preference to the first model. The difference in r-squared is minimal, and I don’t see a theoretical justification for using a squared term. Informal voting goes up with the number of candidates and it goes up with the %NES. But on the NES graph, I don’t see a justification for using NES^2. It’s the outliers at top right that start to drift away from the linear line. If the straight line provides a good explanation, you can speculate on what hasn’t been explained. But to say you get a better fit by using the square of the value means you are just soaking up variance that might be caused by something else.

    If you are going to use a squared variable, you have to say why the rise in informal vote is better explained by the square of the number of candidates rather than simply by the number of candidates. All you are doing with the squared terms is soaking up variance which needs to be explained, and saying it is related to a squared term isn’t an explanation.

  33. lurker said

    At my voting booth in SA at the last Fed election, they did NOT have the HTVs pinned up in any of the booths. Don’t know why.

  34. Antony Green said

    Lurker, the display of how to vote material only applies at SA state elections.

  35. The Doctor said

    does the census data give NES background information – like you’d expect a Dutch speaker to handle things better than an Arabic speaker for example?

  36. Ron said

    literacy is not the main cause of Informals based on the actual 2007 results

    Lowe , Gorton & Bruce were in the top 7 non English speaking electorates per Mumble.
    Yet the informal votes in these 3 seats were ONLY respectively 21st , 36th and 81st highest rate of informals of the 150 seats

    This suggests Language is not the problem

  37. Antony Green said

    I can’t agree Ron. The AEC has undertaken Possum’s analysis at every election since 1984, though without the squared variables. Everyone of them has shown that the census variable with the strongest relationship to informal voting is one related to non-english speaking background. Just because you can point to some electorates that deviate from that relationship doesn’t mean the relationship doesn’t exist. It’s a regression model, some electorates deviate but ordinary least square models such as Possum’s produce a straight line that minimizes the variance from the model.

    A simple model such as Possum’s, including the number of candidates and a variable representing non-English speaking background, and more recently a dummy variable for optional preferential voting in states, always explains over half of the variance in informal voting. The relationship is long standing and consistent and isn’t disproved because you name three seats that don’t seem to match the relationship.

  38. Antony Green said

    Just to add some extra information. In recent years, it is consistent that half of the Federal informal vote has been single ‘1’ votes, and in recent years this has been slightly higher in NSW and QLD. When you look at state election results, you find a similar pattern in all states that use the above and below the line upper house ballot paper. Where an election is held that has the Senate style upper house ballot paper, the informal vote is higher in the lower house than the upper house, and it does appear that the increase in lower house informal voting is caused by single’1′ votes, the form of voting used by many voters in the upper house.

    The one exception is SA, where the lower house ‘1’ votes are saved by the state’s funny registered ticket voting system. If it were not for this system, South Australia’s lower house informal vote would be as high as at Federal elections.

    Two of the seats Ron named were in Victoria. Until the last state election, Victoria was the only state that did not have the Senate style ballot paper for the upper house. And for quite a few Federal elections, Victoria has had a generally lower informal vote.

    Check my submission to the last Joint Standing Committee on Electoral Matters which looked at the differences in state and federal informal voting.


    It is in Part 2, pp.7-35, the last 8 pages of that range including all the tables of comparative informal voting.

  39. fred said

    I understood SFA of that.
    So what I went and did was check out the % of informal votes cast in ’07 in 4 random SA seats.
    1. 4.25
    2. 3.85
    4. 4.66
    4. 4.96
    Those are for the House of Reps.
    Looking at the % informal votes for the same 4 seats, same order, for the Senate I found:
    1. 2.45 ie 1.8 % less, same seat, same voters, same time etc.
    2. 2.52 ie 1.3% less, ” ” ”
    3. 2.58 ie 2.08% less, ditto
    4. 3.31 ie 1.65% less, ditto


  40. Antony Green said

    Exactly. Using the SA state data as an guide, and past surveys of House informals, the House informal vote is doubled by the Senate ballot paper which induces people to use a ‘1’ only vote in the House. It comes down to the basic fact that a ‘1’ only vote is formal in the Senate but informal in the House. We have a high informal vote not because voters are dumb, but because the formality criteria in the legislation is dumb.

  41. Ron said

    Anthony ,

    The ballot system reuires recording ‘number’preference sequences

    Voters who vote unintentionally informal do so due to numeracy
    not language.

    I suggest the solution lies in the voting system and not in language education

  42. Ron said

    2 further points ,

    1/ I suggest alot of the highest informal votes are in safe Labor seats where HTV’s are not strongly serviced by Labor volunteers.

    YET many of these seats have high NES’s.

    To suggest the informals are due to the high NES’s rather than either HTV servicing or a combination is drawing a long bow.

    The question is is the informals high in swinging Labor seats where there is a high NES AND in safe Liberal seats where there is a low NES

    2/ the 50% incidence of 1’s in informals supports the suggestion there is a numeracy problem for voters of 2 different Houses ballot papers systems rather than literacy

    Common ballot paper methods for both houses may reduce informals

  43. fred said

    On an aside but possibly related, I noted on election day, when I was handing out HTV cards, that the card itself was larger than the HTV cards of previous elections and this one was only printed on one side in contrast to those previous which were printed back and front. Usually they had HoR on the front and Senate on the back, but this time they had both on the front.
    I remarked on this to a rival at the polling place and he said theirs was the same and he had heard that the reason for going single sided was that his party believed that voters failed to turn the card over and so their Senate vote had missed out.
    Was this the same with HTVs all over or just ALP and Lib and confined to SA?
    How correct was the bloke’s explanation?
    I wonder what impact this change had and how it could be detirmined.

  44. Ron said

    Fred , there’s far too much commonsense it what you asked.

    Why would we want to simply HTV’s or the 2Houses voting systems
    and thereby may reduce informals

    The ‘experts’ have failed to reduce informals

  45. My take was here. In 2004, the interaction of NESH and ballot length did ok too.

  46. Antony Green said

    Ron, the incorrect sequence erros are quite small, so its no numeracy. The biggest groups of informals are ‘1’ only votes, blanks and words and marks. Literacy and language come in because presumably there are some people who can’t read the instructions. As for the just vote ‘1’, all the state informal figures show that at elections where you don’t have the upper house system allowing a single ‘1’ vote, the ‘1’ only votes plummet. Anyone who filled in the upper house ballot paper first has some likelihood of using the same voting method then on the lower house ballot. As South Australia has shown, if you introduce a provision to deal with people using a ‘1’ only vote in the lower house, you promply halve the informal vote.

    Fred, from my past plowing through HTV cards, and I do it for most seats at most elections, the major parties have almost always used single sided HTVs.

  47. fred said

    Thanks Antony, my past experience has generally been with one particular minor party [you know the one that uses recycled paper].

  48. feral sparrowhawk said

    Fred are you seriously telling me that the SA Greens used a card prior to this election with the House on one side and the Senate on the back? If so that alone would have easily accounted for their losing margin in 2004. I’ve seen this used once and it was a disaster.

  49. Marrickville Mauler said

    Possum, Antony, Peter and other esteemed contributors – has anyone done any analysis of informal rates when people are using electronically assisted voting as per trial at last Federal poll and I think a couple prior at Territory and State level? although process takes longer I think federal electronic trial gave voter feedback before lodging on whether formal requirements satisfied?

  50. The Doctor said

    when I’ve voted electronically at an ACT election it forced a formal vote. I don’t know about other states.

  51. Thanks for discussing your ideas right here. The other issue is that if a problem comes up with a pc motherboard, people should not go ahead and take risk connected with repairing the item themselves because if it is not done right it can lead to permanent damage to an entire laptop. It is almost always safe to approach the dealer of a laptop for your repair of that motherboard. They’ve already technicians that have an competence in dealing with pc motherboard issues and can make right diagnosis and undertake repairs.

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 )

Google+ photo

You are commenting using your Google+ 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 )


Connecting to %s

%d bloggers like this: