Wednesday, July 13, 2016

2016 Tasmania Senate: A Model Of What Might Occur

Button press imminent - results and discussion will be posted on a new thread.  Thanks all for the comments - this has been the most commented thread in this site's history by a long way!

----------------------------------------------------------------------------------------------------------------

Tasmanian Senate model
Outcome of model based on current data: 
Likely outcome 5 Labor 4 Liberal 2 Green 1 Lambie
Next most likely outcome is 5-5-1-1.
Nick McKim and Richard Colbeck main contenders for final seat. McKim appears somewhat better placed on my sampling but contest is too close to call.
Significant chance of Colbeck not making final two and final contest being McKim vs McCulloch
Outside chance of McCulloch (One Nation) win, other micro-parties don't appear competitive.

Note: updates added through count including second model run from Thursday 21 July  - scroll to bottom.

-----------------------------------------------------------------------------------------------------------------------



(Original introduction - see updates below)

Labor's Catryna Bilyk and Lisa Singh are likely to be re-elected to the Senate while Richard Colbeck is struggling to beat his ticketmate David Bushby or the Greens' Nick McKim for one of the final two seats.  We probably won't know who will win for sure until all the votes are entered and "the button" is pressed.

These are the provisional findings of my first attempt to model the Senate race using actual scrutineering data as well as current estimates of the vote totals.  There are many uncertain points in the model which can be refined with further sampling if anyone has the time to do it (I'm not really sure that I do), but I post this as a first attempt.  I warn that it is predictably wonky (again about 4/5 on the Wonk Factor scale).

For more background to this count see my original Tasmania Senate postcount thread.

Most of the text below is the original text of this article written on Tuesday 12th, when Richard Colbeck's position was somewhat better than now.

Tracking the count

Tracking the Senate count has been extremely difficult because of the nature of booth information available.  Votes counted on the night (or in post-counting) are initially recorded as "unapportioned" (meaning that they could be either above or below the line) and later segregated out as above-the-line "ticket votes" for parties and below-the-line votes for candidates.  However some booths temporarily appear in the results in a partly separated-out form, so that the ticket votes for the candidates are shown but the below-the-line votes are not split up by candidate.  As a result, it is necessary to check which booths at any given time have all the below-the-line votes split out to particular candidates.

The leading parties in the overall state count at the moment are:

Labor 4.442 quotas
Liberal 4.217 quotas
Green 1.412 quotas
Jacqui Lambie Network 1.081 quotas

A quota is 1/13th of the total vote (plus one vote then rounded down).  There may still be significant shifts in these votes as more votes are included.  In particular, postal votes - of which only a few thousand have been counted - should slightly improve the Liberals' position.

Large below-the-line votes for Labor's Lisa Singh and Liberal Richard Colbeck mean that the race for the final spots is about candidates not just parties.  As noted on the original thread, I currently project Singh to receive 0.77 quotas (5.9%) statewide, and Colbeck to receive 0.63 quotas (4.88%).  (Tally Room has slightly lower figures for both, but I believe those are statewide totals, not projections weighted by electorate).  These numbers are dependent on the selection of booths included in the count and are likely to move about in the post-count, but I'll go with them for now.

When the preference distribution starts, the top three Labor and Liberal candidates, the Greens' Peter Whish-Wilson, and on current numbers Jacqui Lambie, will all be elected either on primaries or on the cascading surpluses arising from the election of the top ALP and Liberal candidates.  (It's actually not at all clear which of Anne Urquhart (ALP) and Eric Abetz (Liberal) will be elected first, but it is purely of academic interest).  It's possible Lambie will fall short of a quota in her own right, but if so she will be elected quickly anyway.

With eight candidates elected, the remaining incumbent Senators (Bilyk, Singh, Bushby, Colbeck and McKim) have a small problem - five into four doesn't go, so someone is going to lose.   It might seem that it is just a matter of subtracting the quota totals for Singh and Colbeck from their party totals to see what Bilyk and Bushby will have left over, and on that basis that Bilyk and Bushby are in clover.  After all they start roughly level with Singh and Colbeck, but they have access to above-the-line preferences from other parties, while Singh and Colbeck can only get below-the-lines.  However there's an important complication:

Within-ticket leakage

Some votes within each major party's haul are below-the-line votes for candidates other than Singh and Colbeck.  For example my tracking has Eric Abetz on target for a below-the-line vote of 2.35% statewide.  Many of these below-the-lines go straight down the Liberal ticket, for reasons including:

* the voter wants to re-order candidates from another party they have preferenced, but not within the Liberal ticket
* the voter wants to use their 12 below-the-line votes to just preference four parties instead of six while complying with the instructions

However, some do not, and 1 Abetz 2 Colbeck votes are relatively common (so far I have found about 20% to be like this).  This makes sense since Abetz and Colbeck are very high-profile (a former Government Leader in the Senate and a current Minister, respectively) while Bushby, Parry and Duniam are not nearly so well known. With below-the-line votes close to 10% of Abetz's total, having 2% of his 2.4-quota surplus "leak" to Colbeck is no small matter.

Similarly there is some flow from Anne Urquhart to Lisa Singh (although most Urquhart BTLs are straight down the ticket, and Urquhart has fewer of them than Abetz).  Then there are voters who have voted 1 for a major party candidate who is not top of the ticket, but whose preferences will be distributed early in the count (Polley, Brown or Short for Labor; Parry, Duniam and Tucker for the Liberals).  These preferences are much less likely to follow the ticket in choosing between Bilyk and Singh on the one hand and Bushby and Colbeck on the other. From what I've seen so far, those splits aren't far from even.

There are also some outright leaks from the party tickets to other parties.  Taking all the within ticket-leaks into account (on some of them I have very limited data) Singh might have about 0.825 quotas, Colbeck 0.748, Bilyk 0.600, Bushby 0.430 and McKim 0.422 (picking up the odd leak from the majors).  This is just an artificial starting point once all other major party candidates are eliminated or excluded, although in fact some micro-parties will be cut out before Short and Tucker.

Then there are a pack of micro-party preferences, which look something like this:




This chart gives the approximate percentage of the overall votes, the portion of a quota, the quota portion split between above and below the line, and the percentage of votes that are below the line for each group.

The preferences of these micro-parties with a combined 15% will largely determine the election once the major party primaries are clear and leakage against the ticket orders processed.

Estimating preference flow

Over the past several days I have sampled the preference flow of 327 micro-party voters, including 58 below-the-line votes. There have also been many other votes I've watched without recording full sample data. BTL votes for micro-parties are rarely seen and difficult to follow on the screen, and JLN is the only micro-party for which I caught any real number of them.  The votes sampled come from electorates around the state, but are not necessarily an even or reliable mix.  

Initially I thought that four micro-parties should be considered in contention for a seat as well: One Nation, Family First, Shooters Fishers and Farmers, and Nick Xenophon Team.  These are the four highest polling micro-parties and also they would seem to have more chance based on the right-wing lean of the micro-parties generally (compared to the left-wing Sex Party).

The preference flows seen support something I had been saying all along in the Senate reform debate: micro-party voters do not just preference other micro-parties; many micro-party votes go to the "big three" (Labor, Liberal, Green).  The fact that micro-parties with really trivial vote shares (well below 1%) cannot win in the new system has little to do with some votes exhausting and everything to do with the fact that all preferences splatter.  The preference flows between any party and any other are too weak for any party starting from a very small share of votes to work its way through the field to victory, and this would be the case even if no votes exhausted.  The strongest preference flow I saw from any more than a handful of votes was 50% from Shooters Fishers and Farmers to One Nation.  

There are some very surprising preferences too.  In the first minutes of my main sample I saw a vote 1 Sex Party 2 Family First, even though these two parties are so strongly opposed to each other that the Sex Party is often seen as a deliberately anti-FF party.  Not too long after I saw the reverse and by the end of the day I had seen six votes with these two as the top two parties.  I tried to think of the most illogical 1-2 vote that could possibly exist and decided that 1 Shooters Fishers and Farmers 2 Animal Justice Party would be quite bizarre since the AJP believe that all of shooting, fishing and animal farming should be either banned immediately or at least phased out over time.  Within ten minutes I had seen such a vote, and later I saw the same thing the other way round.  

The following are some general themes I've picked up from these rather small samples and general vote-watching so far:

* Hardly any micro-party voters follow their party's how-to-vote cards.  Hooray!

* Virtually no-one votes for just one party above the line.  In thousands of votes I've watched I've seen three just-1s for the Liberals, two for Labor and one for the Animal Justice Party.  I've also seen one vote for just two parties.

* Lambie Network above-the-line voters often preference One Nation, but Lambie Network below-the-line voters more heavily preference prominent major party candidates.

* None of the left-wing micros that could have served as preference feeders for the Greens under the old system (Animal Justice, Renewable Energy, Sex Party perhaps, Science, Flux) reliably preference them when it is a matter of voter choice.  Voters for such micros may be nearly as likely to preference Labor.

* NXT voters mainly preference the "big three" rather than the right-wing micros.

* Religious right parties mainly preference each other and the majors.  There's only a small flow to SFF and One Nation.  

* The Family First vote has a large semi-donkey component. Quite often voters have picked six parties including FF and then put those six in the order they appear across the page.  This among other things seems to lead to quite a high FF to Labor flow. Let's not have a double dissolution in the future without a serious look at some basic ballot paper rotation first, thanks. 

(Any of the above impressions might be wrong based on the small sample sizes involved, but none of it is all that surprising.)

I used the data collected to do a mock preference distribution on the assumption that all micros other than the top four would be uncompetitive. The small sample sizes are such that there will always be zero preference flows and other such nonsense in the model, but again, these should on the whole cancel out.

The following (numbers in quotas) are the final stages of this crude mock-up of the election.



At the above point in the simulation Colbeck is narrowly eliminated and the election stops with McKim and Bushby elected short of quota.  But that does not mean this will necessarily happen! In particular, if the Liberal vote lifts further in post-counting, McKim may fall behind in the projection.

Some comments:

* SF+F might get over NXT but it doesn't seem to matter because they do not get enough preferences from either NXT or Family First to put them near One Nation.  Their problems include that the helpful Lambie Network ATL surplus is so small and that One Nation, a potential rich source of preferences, is too far ahead.

* NXT also just can't catch the party ahead of them.  Family First have strong flows from the Christian Democrats and similar flows to NXT from other micros generally.

* Family First might beat One Nation, but their flow from One Nation is weak, so I would not expect it to change the outcome if they did.

* It is possible given the various potential errors in the estimates that one of Bushby, McKim or Colbeck would be actually excluded before Kate McCulloch.  If this happens, McCulloch probably still wouldn't win, but she would still have a chance, if a distinctly long-shot one.  Probably she would rely on performing well compared to the model and having two of the above perform really badly relative to it, then having so many of the first beaten candidate's preferences exhaust that she beat both McKim and one of the Liberals.

* The final model outcome is very much too close to rely on.  In this run McKim and Bushby win and Colbeck misses out, but even changing the right 2-3 votes in the original sampling could cause Colbeck to win.  (Furthermore the flows from Singh to the remaining candidates at the end are just guesstimates since I haven't sampled that question much - but note that if Singh quotas after Bilyk, then Singh's surplus is swamped by Urquhart preferences and so all the 1 Singh 2 Colbeck BTL votes will count for little.  This is a result of the repulsive Inclusive Gregory system for surplus transfers, in case you were wondering.)

* Both Labor Senators crossed quota rather easily in this mockup.  Generally I found that Labor was getting stronger or as strong preference flows from nearly all parties than the Liberals or Greens were, which is not surprising in the context of the overall election result and the Singh factor.  However it is still possible for the current statewide projected totals to shift, or for the model to have errors based on the relatively small preferencing samples, such that either might still lose.  

I hope this model serves as a useful base for speculation about this contest, and if I learn more about some of the preference flows it can be further refined and rerun with a larger sample.  Meanwhile I do have to spend some time on other work!

Update: Thursday 14 July.  A large volume of Senate votes have just been uploaded.  Based only on the booth counts (not the postals, absents and declaration prepolls) David Barry has Singh on 3.2% of 29503 in Bass, 2.7% of 5827 in Braddon, 11.4% of 19536 in Denison, 8.7% of 10929 in Franklin, 4% of 20688 in Lyons, for a projected state total of 6.01% (0.782 quotas).  Colbeck is on 3.1%, 8.1%, 2.9%, 4.4% and 3.6% for a projected state total of 4.41% (0.573 quotas).  That is a gain of .01 quotas for Singh on the figures in my model, and a loss of .06 quotas for Colbeck.  However Colbeck's vote might be reasonably adjusted upwards, say to 0.58 quotas, to account for the Liberals' superior performance in post-counting.

(Note that the AEC totals which imply much lower below-the-line vote rates are not reliable because some booths and some vote categories are being sorted only as Ticket Votes vs Unapportioned, with below-the-line votes not separated out.  David Barry is running an automated version of the manual sorting of fully counted and partly counted booths that has been taking me hours every time a new set of figures comes out.  It's the same method I was using, except with much less potential for errors!)

It is difficult to know what is going on with the postal votes, which have been at least partly sorted into above the line and ticket votes in some divisions, because the rates of below-the-line voting in the postal votes appear to be massively lower than in the booth votes.  For this reason I believe it is only possible to use the booth totals at the moment in estimating the Singh and Colbeck votes.  If there is a real difference in below-the-line votes in postal voting this would make Colbeck's position more difficult, and McKim's easier.

There have also been shifts in the overall total with the Liberals up slightly to 4.23 quotas and Labor down to 4.40.

Using the new booth figures as a base for the model, the results at the end would become something like .92 quotas for Bushby and .75 quotas for Colbeck.  This means that with these revised figures the model now predicts 5-4-2-1 with Bushby defeating Colbeck for the fourth Liberal seat as the most likely outcome.  However more sampling is needed on the preference flows to the Greens from minor parties, and Labor vs Liberal from One Nation, especially to be more confident that this will be the case.

Update Friday: Again thanks to David Barry (and a free plug for his election statistics website which has all kinds of excellent goodies on it).  There is a further improvement in Singh's position and a further decline in Colbeck's.  In ordinary (booth) voting Singh is now on 3.3% of 32634 in Bass, 2.7% of 9061 Braddon, 11.7% of 23829 Denison, 9.1% of 14314 Franklin and 3.9% of 23287 Lyons, for a statewide average of 6.14% (0.8 quotas).  Colbeck is on 3.2%, 7.0%, 3.0%, 4.5% and 3.5% by electorate for a statewide average of 4.23% (0.55 quotas).  The problem for Colbeck is that his vote has been declining with more counting in the one electorate he is strong in, while Singh is strong in two electorates.

It will be interesting to see what happens with Singh's vote in the rest of Denison.  So far she has done exceptionally well in the left-wing inner-city and bush-fringe booths that the Greens usually dominate, but struggled to break 10% in the Glenorchy booths. There are two types of booths we have not seen action from yet and they are the North Hobart area (which has a fairly high Labor vote but not so many Green votes to plunder) and the Sandy Bay area (a Liberal enclave). Her vote might go down a little once these are included.

Further Colbeck/Singh Projection (Saturday):  I thought it might be interesting to try projecting the final ordinary booth votes for Singh and Colbeck on the assumption that their share of their party's below-the-line vote remains the same in each electorate through the rest of the count as it has done so far.  This method projects Singh's final ordinary booth vote to fall very slightly to 6.03% (0.78 quotas) but Colbeck's to fall significantly to 3.91% (0.48 quotas).  The major issue for Colbeck in this projection is that it seems the very early booths counted in Braddon were unrepresentative and the projection suggests he will continue to go backwards there, to about 5.5% of ordinary votes.  He would need to score a higher percentage of the Braddon Liberal BTL vote to beat this projection than has happened so far.   This projection seems to add more to the evidence that Colbeck is in a difficult position.

On the assumption that Colbeck is not beating Bushby, he beats McKim if the preference flow to the Liberals from micro-parties generally is much stronger (or that to the Greens much weaker) than in my sampling.  This is still a realistic chance given the small size of the sample.  If this were the case, Bushby would be put over quota earlier in the count, allowing all ATL votes reaching the Liberal ticket to go to Colbeck and allowing him to beat McKim.

However that would be a straight race between the Green and Liberal tickets in which the Liberal ticket must beat the Green ticket by (on current figures) .1806 quotas out of 1.881 quotas (a gain rate of .096 votes/vote).  This in a contest in which the Liberals are more exposed to BTL leakage out of their ticket than the Greens, and in which votes are exhausting in significant numbers, and votes are being soaked up by Labor, and also some of the parties will break much more to the Greens than Liberals.

The other complication here is that there's now a much more serious chance that Colbeck leaves the count before McCulloch.  That ticks one of the boxes for One Nation to win (the other two being that they do much better on preferences than expected and McKim does much worse).  So the chance of a One Nation upset, while still apparently low, is higher than it looked a few days ago.

Update Monday 18th: 

Revised figures for actually processed booths posted by David Barry show Singh on 6.2% (.806 quotas) based on actual counted totals from 154/310 booths (Bass 3.2 Braddon 2.5 Denison 11.7 Franklin 9.6 Lyons 3.9).  The main difference with my projection above is that it expected Singh to finish on 8.7% of ordinary votes in Franklin; after the new data this has increased to 8.9%.

Colbeck has continued to fall significantly in Braddon as expected and is now on 4.11% of the counted ordinary votes (.534 quotas) in counted booths (3.2, 6.3, 2.9, 4.6, 3.5). My projection for him to fall to 5.5% of ordinary Braddon votes has not changed.

The other news is that Labor's total has come down to 4.38 quotas since I started this article, with the Liberals up to 4.24.  In my sample, Labor was doing much better than the Liberals on micro-party preferences, and on that basis I'm projecting their ticket lead over the Liberals to increase to about 0.35 quotas at the point at which Family First have been excluded.  But even if my sample is rubbish and the Liberals are actually doing much better than Labor, this would place Bilyk at risk rather than Singh.

I've also noticed that while Singh performs slightly worse as a share of Labor's BTL postals than as a share of Labor's BTL ordinaries, it's not by much.  So unless it turns out that the proportion of Labor voters who vote BTL in the postcount is very much lower than those who do so in ordinary votes, it's getting hard to find a way that Singh could lose.

Monday 8 pm: Figures are now up to 181 booths and there has been not much change, mainly because not much more has been counted in Braddon.

Wednesday 1:30 pm: Further revised ordinary booth figures show not much has changed: Singh on 6.17% and Colbeck on 4.14%.  Colbeck had improved substantially in Denison (he polled 16.6% at Sandy Bay Beach, a Liberal stronghold) and I now project him to improve to 4% of ordinary booth votes in Denison, which means he probably won't fall below 4% statewide as earlier projected.

The other interesting thing is that as postal votes have filled out, we can see that the BTL candidates are doing well on postal votes in Denison and Franklin.  In Denison Singh has 12% of postals, which is exactly what she is getting in booth votes, and Colbeck has 4.2%, which is slightly more than he is getting in booth votes.  In Franklin Singh has 8.7% of postals, slightly below her booth vote, and Colbeck has 4.4%, slightly above his booth vote.  The suggestion is that once postals fill out for all electorates Colbeck's vote will improve slightly while Singh's will not decline much.  So there is more confidence that the booth votes are a good guide to where these two will finish up in the primary count.

Thursday 1 pm: A big upload for Braddon and as projected Colbeck has dropped back to 5.5% there and is now barely above 4% of ordinary votes.  However Singh continues to rise and is now on 6.23% with a new best booth of 20% at Mt Stuart.  I am doing more scrutineering today but if anyone has a scenario for how Singh can realistically lose from here I would be interested to see it, 'cause I don't.

Thursday 4 pm: I will post a new projection overnight.  In my sample so far today I am probably seeing slightly more positive signs for Colbeck and also McCulloch but whether it's enough to shift the picture will have to wait til I process the full sample.

Thursday 21 July: Model Second Run Based On More Data

I have done another day of scrutineering and beefed my sample of micro-party votes up to a total of 534 above-the-lines and 173 below-the-lines.  I have also looked at leakage in the Greens ticket.

The button press is expected around Tuesday or Wednesday next week and I don't believe we can be super-confident about the last seat until it happens.

The following are my general comments:

* The data entry staff are just astonishingly fast.  It was tough enough even for someone with my ferocious speed and good eyesight to keep up with them early on in the process but they have slightly more than doubled their average speed since the count began and have a graph on the wall to prove it! There is always the possibility that a preference sample like mine contains biases based on which votes are easiest to read data from before it disappears off the screen!

* Today's sample showed better preference flows to both the Liberals and One Nation than the previous sample, and weaker flows to the Greens.  Especially the flow to the Greens was weaker from the Sex Party/HEMP ticket than I earlier had it.

* The preference flow from micro parties to Labor continues to be strong.  Normally I would expect a lot of these micros to preference the Liberals more than Labor, but in the context of a 56-44ish 2PP drubbing at Reps level in the state, it seems that quite a few right-wing voters exhausted their preferences rather than preference the Liberals.

* Today's sample showed a stronger BTL "leak" to Richard Colbeck from Eric Abetz, pushing my estimate of the share of Abetz votes that reach Colbeck ahead of Bushby up to almost 30%.

* About 11% of Greens BTLs leave the ticket instead of going to McKim.  There are no prizes for guessing who gets most of the escapees.

* When a micro-party BTL reached the major parties, I originally had Singh getting 69% of the Labor ones and Colbeck 41% of the Liberals.  But after today I'm seeing more that just hit a major ticket and go down the line, so these numbers are down to 56% and 32% respectively.

* Sex Party-Family First, Family First-Sex Party.  Aaargh! I have now seen this stupid "vote for whatever micro-parties you have heard of" vote ten times, and saw a few more Animal Justice/SF+F votes today too.

Here's the closing stages of my revised preference distribution model:



In this model Singh picks up about .04 of a quota off BTLs within the Labor ticket and leaks from the Liberal and Green tickets and by the start point of the above already has enough votes to be safe.  ATL preference flows from the micro-parties to the Liberal and Labor tickets ensure the election of Bilyk and Bushby.  The micro-parties other than One Nation all again do not get anywhere.

The crunch point comes after the expected exclusion of Family First when someone (probably Colbeck or McCulloch, with an outside chance of McKim if the Greens' preference flows are much worse than my model) is excluded.  In the model it is McCulloch, and her preferences go all over the place with a lot exhausting and create a cascade of surpluses, leading to the end result where McKim defeats Colbeck by a modest margin.  However it could almost as easily be Colbeck who is excluded at this point.

If Colbeck is excluded, Bushby gets a large surplus dominated by Liberal ticket votes which then puts Bilyk over quota, and Bilyk's surplus puts Singh over quota.  This creates a Singh surplus which by value consists mostly of Labor ticket votes (Inclusive Gregory strikes again!), which will flow to McKim more strongly than McCulloch and would presumably ensure McKim wins.

If McKim is knocked out before either McCulloch or Colbeck (which on my sampling is rather unlikely) then that also creates Labor surpluses dominated by Labor and to a lesser extent Green votes which presumably (but I haven't checked yet) help Colbeck more than McCulloch - this seems to be McCulloch's best chance, though it seems rather unlikely McKim does quite badly enough to be behind both at the key point.

If my sample (and sampling) is representative then despite the fairly close margins the level of doubt about the result is not that high.  However even though it now includes preferences taken in three different sessions, it's still always possible the real preferences could be markedly different.  For this reason the Greens should not be that confident yet, and Colbeck still has a realistic chance.

Monday 25 July 12:30 pm: Primary booth figures are almost done and dusted with only a few booths in Lyons missing and we have Singh on 6.25% from ordinary booths and Colbeck on 4.00%. In all the primary count (including informals) has now reached 94.05% of enrolment and there must be very few votes left to add to the primary tally indeed, but 676 booth BTL votes are yet to be apportioned and there may be more postcount BTLs to be tidied up as well.  Singh and Colbeck are currently up to 6.09% and 3.94% in the overall totals so should go very close to matching their booth votes when all votes are included.

The significant development is in the totals with the Greens up from 1.427 quotas when I did the projection above to 1.45 quotas and the Liberals down from 4.234 to 4.228, a net improvement of 0.029 of a quota in McKim's position relative to the Liberals, which is far from meaningless in the context of a projection that had McKim's final margin at .110 of a quota.  The main reason for this is that the way the postcount votes are folded out causes below-the-lines to come in late, resulting in an increased vote for parties with high BTL rates.  Labor are down from 4.381 quotas to 4.361.  One Nation have also dropped back slightly from .337 to .334.  Because of changes in Colbeck's personal vote it is now still even closer as to whether or not he is excluded before One Nation (if he stays at 4.00% my projection still has him a whisker ahead, but not if his final total is a bit lower.)

If Colbeck is cut we're looking at about 0.58 of a quota of Liberal votes to be thrown with Singh on about 0.927 of a quota, Bilyk on about 0.88, McKim on about 0.78 and McCulloch on about 0.634 (according to my sampling).  Anything from Colbeck that doesn't flow to Bushby (including the substantial number of 1 Colbeck 2 Singh BTLs) would leak at full value.  Although on my observation Labor was getting more Liberal ATLs than either the Greens or One Nation, there should be enough exhaust that any Bilyk/Singh surplus won't now be very large.  It seems highly unlikely One Nation would catch the Greens at 0.24 votes/vote on Liberal preferences with votes also exhausting and going to Labor, especially on what I have seen of Liberal votes going all over the place.  But if One Nation are actually doing much better on preferences than my sampling shows, or the Greens much worse, then they would have some sort of chance.

Tuesday 12:30 pm: There are now no unapportioned votes showing, with Singh currently on 6.12% and Colbeck on 3.97%.  It is notable that both these candidates have performed less well on out-of-electorate prepolls, absents and provisional votes as compared to booth votes and postals, so it may be there are still some votes to add for them in these categories, which would cause their tallies to still increase slightly.  The AEC are now expecting the button press towards the end of the week according to the ABC.

Tuesday 5:50 pm: The Mercury now reports the count could be finalised tomorrow.

I should comment briefly on the argument made by Jody Fassina in that piece and elsewhere, which is that Colbeck starts ahead of McKim and has a greater pool of right-wing votes available to him.  What that argument misses is that the fourth Liberal David Bushby will start a long way short of quota, and until Bushby crosses quota, Colbeck does not get a single above-the-line preference from anywhere, while McKim does get above-the-line preferences (as well as below-the-lines) and will gain votes over Colbeck at just about every micro-party exclusion.  This is why the Liberals actually need a much greater share of micro-party preferences to go to them than go to the Greens (or for the Greens to get hardly any micro-party preferences at all), if Colbeck is to win.  That could happen, (and indeed my projection does show Colbeck closing about half the gap on ticket votes) but it is not as easy as the argument suggests.

Tuesday 10:00 pm: Unconfirmed but I've heard on the grapevine that the button press is tomorrow afternoon. On the assumption that the figures are final (with Singh and Colbeck as above), I've entered revised estimates for the main contenders into my model.  It now (by a tiny margin which I place approximately no trust in) has Colbeck excluded before One Nation. If Colbeck outlasts One Nation, it expects McKim to beat him by about .14 of a quota (though the usual caution about possible errors in my preference sampling applies.).  If he doesn't, we enter the unsampled world of whether Liberal voters prefer One Nation to the Greens enough to put McCulloch in (I rather doubt it).

Button press imminent - results and discussion will be posted on a new thread.

87 comments:

  1. Great to get some actual insight into Senate preferences, thanks Kevin.

    I think the tables need a couple of fixes. The minor party preference table seems to have flipped the numbers in the ATL and BTL columns. The tracking of the final few stages has many quotas decreasing from column 2 to column 3, which shouldn't happen.

    ReplyDelete
    Replies
    1. Ta. Both fixed. Second was just a copy and paste error (forgot to paste special: values) so the cells recalculated in the original printed version only, it wasn't an error in the original simulation fortunately.

      Delete
  2. Good effort for 3am. Some of your column 3 figures are less than column 2?

    ReplyDelete
  3. Appreciate your article, thank you

    ReplyDelete
  4. Very good insight, it would be interesting if the remaining seat is Bushby vs Colebeck and the margin is close enough to demand a recount anyway. One would have to say that the senate reforms have been a success and strongly highlighted that part preference deals have little in common with how the majority of people will preference if pressed to make a choice.

    I think some form of ballot rotation would be a good idea for the senate, although i never got why people donkey vote by actually putting in a legitimate vote for a party.

    ReplyDelete
  5. Senate recounts are actually extremely rare (fortunately). The 2013 WA recount happened because the old system created tipping points between micro-parties way down in the preference cutup. In this one whether the Renewable Energy Party outlasts the Science Party by ten votes or the other way around has no bearing on the outcome, so a recount would only be triggered by a very close margin at the end or one of the final stages. The margin in my simulation is close, but it's not that close; it's still over 1000 votes.

    ReplyDelete
  6. Given the publicity for Singh (a left wing Hobart based senator) do you think there's a big risk that greens votes go '1' greens BTL then '2' singh (before going back to the greens ticket/McKim) or even vote female and go greens candidate number 3 then Singh before going back to the greens tally? Thinking about it and I wonder if within ticket 'leakage' could be a lot higher for the greens (although obviously not in actual raw number of votes given they only polled 1.4 quotas) given the risk to Singh, a female left wing Hobart based candidate and the higher btl rate for greens voters in general.

    ReplyDelete
    Replies
    1. I have so far seen very little leakage from the Greens ticket BTLs and they are helped by McKim tracking for about 1.5% in his own right, meaning that only about 0.2 of a quota is available to potentially leak. So a 10% leak would cost McKim 0.02 quotas, but I doubt it is that high, and I've been rather stingy in what I've given him for leaks from the other parties anwyay. The point about Reynolds is a good one and difficult to say since votes for her are so rarely seen; that said it looks like she'll only have about 0.05 quotas anyway.

      Delete
  7. Glad about the "Virtually no-one votes for just one party above the line."

    i.e. most people followed the stated directions (even if those few votes were formal via the savings provisions)..

    Did you notice any partial-Donkey votes.
    eg. 1 in Labor, then 2,3,4,5,6 following (maybe not so relevant in Tasmania, but would not be surprised if this occurs in states with the larger papers)

    ReplyDelete
    Replies
    1. Family First aside, not really. But I did notice a small level of suspiciously clustered micro-party voting where the voter had voted 1 for one micro and then the remaining five for the other micros near it on the ballot. This reminds me a little of the "proximity preferencing" effect that is observed with a single long list of independent candidates, as in a council election - voters are more likely to send their next preference to a box within a few above and below the candidate they have put a given number for.

      Delete
  8. NXT lost its number 2 candidate due to illegibility. What effect do you think this may have had?

    ReplyDelete
  9. The ineligible NXT candidate was still on the ballot paper and able to get votes but would have been disqualified if he won a seat. He was never going to get many votes anyway though.

    Whether the bad publicity much affected the vote for the NXT ticket (1.5%), or its preference flow, is hard to say. NXT's vote in Tasmania is virtually the same as their vote in Victoria, and in Tasmania they have the problem of competition from Jacqui Lambie. Perhaps they would not have done that much better than that here anyway, though I know at least one voter who said they preferenced NXT lower because of it.

    ReplyDelete
  10. This is very interesting Kevin.

    Can you give a thumbnail sketch of the paper trail from Polling Place to the appearance in the VTR? How much stuff is actually passing through the scanners? The counting in Tasmania is unique among the States - no other State is entering BTL data into the system at this rate - usually not at all, except for the Ungroupeds.

    You might get a similar question from other baffled analysts soon.

    ReplyDelete
  11. Something seems to have altered as there has been no change to the BTL totals for candidates by division for Tasmania for the last two and a half days.

    As I understand it raw votes from the booths are packaged up at one counting centre (which I haven't been to yet - I think this includes raw counting of the 7% or so of remaining post-count votes from time to time). Then they are taken to the main scanning centre, put through the scanners and the scanned output is then verified or corrected (as needs be) on screens by a bank of about 25 data entry operators working at a time. There's a process for dealing with difficult votes (of which there are quite a few - there are some appallingly illegible votes) by a series of raises up a chain, and I think that if the entry operator makes any change to what the image recognition software has said then the vote is automatically sent to another operator for rechecking. (It's quite common to see what is obviously the same vote first on one side of the room then again on the other). On Tuesday when I was over there something like 28,000 papers had been scanned in the morning, but they weren't going to clear that many through data entry. Some more were then scanned later. Nonetheless based on the number of votes I was able to sample in a count where most votes are not worth sampling I expect they are clearing well into five figures per day. The entry staff are seriously fast, except when they slow down for a difficult vote.

    The more mysterious part is the points at which the totals broken down into ATL and BTL appear in the VTR. The staff themselves don't know when votes are going to come up, because apparently votes only appear segregated by booth when all the votes for that booth are recognised as complete by the system - which may take a while because votes from one booth are not necessarily all processed together.

    A further complication is that some booths have been appearing in the VTR counts with a breakdown into Ticket Votes and Unapportioned, but no breakdown of the Unapportioned (which are probably all BTLs at this stage) by candidate. However when a booth appears in the count in this fashion, it generally soon afterwards disappears and presumably goes back to being all Unapportioned. There seemed to also be an alphabetical component in the way booths of this kind appear with a Ticket Votes only breakdown and then disappear.

    ReplyDelete
    Replies
    1. I have some sympathy for any crazy strategy of how to do the count better. I wonder about the voter scanning their own ballot at prepolling centres. If there was a code on the ballot they could be checked against each other once counting started. Or electronic voting could be give you a print out receipt like with shopping with a your vote just down the page. You then put your ticket in the box and it is reconciled with the saved electronic vote in the count.

      I'm sure there are pros cons and barriers to any scheme, but id say even at major city per polling booth and on the day for example town hall in Sydney it might makes 20% sonething votes easier.

      Could have something similar for postal, as we have no idea who is actually filling in those votes anyway. You vote electronically, but get a printout to mail in that gets reconciled but is instant to count.

      Delete
  12. Kevin

    Based on what you've seen what do think the chances are the for the second Greens candidate in WA (the incumbent Rachel Siewert) Looks like the final two spots will come down to grn, one nation and nat. Do you think one nation will attract a lot of preference flow? Could the libs even catch the Nats on preferences?

    ReplyDelete
  13. On current WA Senate figures Lib 6 will get eliminated quickly. One Nation should attract enough preferences to stay ahead of at least one of their rivals if not both. Nationals vs Greens for the final position is too close to call without detailed scrutineering data. It does seem that a right-wing preference mix should favour the Nationals but maybe not enough if the final Green to Nats gap on primaries grows in the post-counting.

    ReplyDelete
  14. Thanks Kevin, I'd forgotten that the libs would likely be eliminated very early and so unable to receive the preferences they had.

    ReplyDelete
  15. 5-4-2-1 looks like a bad result indeed for the Coalition indeed in Tasmania, considering that Lambie is effectively Labor on most economic issues.

    If the LDP get two seats (in NSW and Qld) in the Senate, then the Coalition Government effectively needs 9 out of 9 (NXP+ON+LD+Hinch) crossbenchers for their economic policies if Labor and Greens are opposed, and effectively 8 out of 8 (NXP+ON+Hinch+Lambie) on civil and social liberties.

    We could be looking at nearly three years of stalemate, stagnation and deadlock. Potentially longer if the Coalition narrowly win a third term.

    ReplyDelete
    Replies
    1. Even if I was a Liberal voter, I'd say serves them right. I'm sure many will go through and do analysis of what an ordinary half Senate election would look like. I'd also add being able to choose any date rather than forcing it now might have helped too.

      Delete
  16. Yes if it does end up 5-4-2-1 that will be an absolute disaster for the Liberal Party in Tasmania and the Coalition more generally, creating an 8-4 split against the government in many major policy areas. If this occurs I will have some lengthy comments about who is to blame.

    I guess the plan would be that most of the crossbench get 3-year terms and are defeated in 2019. However the non-Coalition parties could all conspire to prevent this from happening if they wished.

    ReplyDelete
    Replies
    1. I believe that putting Abetz at the top of the ticket, and the fact that all three pre-existing Liberal MHRs were considered very conservative even within their party, has hindered the performance of the Liberal Party in Tasmania. It is very sad to watch them denying themselves of any responsibility in the post-election post mortem.

      Perhaps they might hatch a plan to assign Senators by state population and reduce the number of Tasmanian Senators from twelve to four.

      Delete
    2. "Perhaps they might hatch a plan to assign Senators by state population and reduce the number of Tasmanian Senators from twelve to four."

      Not possible under the constitution.
      Tasmania, as one of the "founder" states of the Commonwealth is guaranteed same number of Senate seats as the other "founder" states.

      Delete
    3. There would have to be a referendum passed to change the Constitution and Tasmania as the affected state would have to vote in favour.

      Delete
    4. If 1st elected is used, the 6-year crossbench is likely 3 Green, 2 NXT, 1 One Nation and 1 JLT.

      If recounting is used, there is a chance of an extra 2 Green 6-year Senators and not much other change to the crossbench.

      Delete
  17. Hi Kevin,

    I'm trying to follow your latest numbers to see how you are getting Singh with ~6% of the vote or ~0.8 quotas even with the latest update, based on updates from AEC.

    I can easily see how Singh's BTLs account for just over 6% of the total ALP vote, certainly not the entirety of the Tasmanian voting public. Are you privy to a set of numbers (sitting inside what is currently listed as Unapportioned) that we don't see?

    At the time of writing: Singh BTL 6954, ALP Unapportioned 16049, ALP total 107953, Tas formal total 319393.

    ReplyDelete
    Replies
    1. See my comments in the Thursday update. By this stage virtually every unapportioned vote is a below the line vote in a booth where the BTL votes are not yet separated by candidate but the Ticket Votes have been included in the count. Furthermore, for the post-count categories (absent, provisional, out-of-electorate prepoll and postal) the BTL votes in some categories in some electorates are not even in the count at all, or in others are only partially included.

      Delete
    2. Clarification appreciated, thanks.

      It looks to me like we can already conclude that regardless of how many Unapportioned ALP go to Singh from this point forward, Singh beating Short is almost certainly inevitable.

      Taking a simplistic view with some round numbers:

      If Singh has 0.8Q out of 4.4Q total, then 3.6Q elect the first 3 senators leaving 0.6 for Bilyk and potentially seeing Singh elected as the 4th ALP senator (or not - depending on ATL flows to ALP).

      Even if all of the Unapportioned on the AEC website turned out to be ticket votes and Singh only started with 0.3Q, then 4.1Q out of 4.4Q elect the first 4 ALP senators leaving Short on only 0.1Q, competing with the smallest parties for early elimination.

      Delete
    3. (That rough example ignores the other ALP BTLs, assuming they roughly flow down the ticket in an insufficiently different fashion to ATLs)

      Delete
    4. Yes there is no realistic doubt that Short will be excluded early and that if Labor wins five then Singh wins. It is perhaps unlucky for Short, who was disadvantaged by the use of an inappropriate quota in the original ALP preselection, which denied him any chance he may have had to be preselected ahead of Helen Polley.

      Delete
    5. So with the last 2 updates, it looks like it is playing out exactly as you have described, in that a very large proportion of the Unapportioned are now Singh BTL votes and I expect another large slug of these at the next update.

      When you were looking at the votes to get an estimate of micro-party preference support, did you get any chance at all to look at how Singh's BTLs were breaking - whether they stay in the ALP ticket or leak out to somewhere else? It's possibly the inverse of the question posed by "Curious" above. I would imagine that the strength of the "Save Lisa Singh" message may have held back the Green primary vote a bit, and that some of these might return in BTL #2s and #3s...?

      Delete
    6. Singh's leak a lot, or would if they were distributed (which looks unlikely to be the case at any value worth noticing). There is a big leak to McKim and a significant one to Colbeck. I didn't sample for very long and those I was sampling were from the south but 20% of what I saw was leaving the ticket at number 2.

      Delete
    7. Ah, of course. They won't actually "return" since Singh will either just scrape to a quota vote by vote, or be elected on a large partial quota.

      Thanks.

      Delete
  18. Also of interest is the number of 1Ps that Abetz has picked up (0.115 of a quota) - about 2/3 of Colbeck and twice Urquhardt in #1 for ALP.
    Quite possible that a number of these could have Colbeck high up in secondary preferences

    ReplyDelete
    Replies
    1. They do. About 20% reach Colbeck ahead of Bushby in my sampling. 1 Abetz 2 Colbeck is quite common. Also note that Abetz's 1P is actually going to be about 0.34 of a quota because of the issues mentioned above.

      Delete
  19. I wonder how many votes go 1 Colbeck, 2 Abetz!!?

    ReplyDelete
    Replies
    1. That alas I didn't sample. It would probably be inflated because anyone who went 1 Colbeck then down the Liberal ticket would have gone 2 Abetz. My perception was that 1 Colbeck 2 non-Liberal was common, but not as common as 1 Singh 2 non-ALP.

      Delete
  20. When you say "repulsive Inclusive Gregory", Kevin, I presume the part that's repulsive is that it's "unweighted" rather than "weighted"? Nothing wrong with weighted inclusive Gregory, and hopefully we can persuade JSCEM to recommend it during the next Parliament (assuming the next Parl last more than 3 months...)

    ReplyDelete
  21. Kevin
    In the comments above someone talked about the possibility of reducing the number of Tasmanian senators. You quite rightly pointed out the constitutional impossibility of that. What I am wondering though, is if there might ever be a move to increase the number of ACT and NT senators? If the number of senators from each territory was doubled from two to four, they would still have more population per senator than Tassie. If this could be done simply by changing them over to half senate elections like the states, then DD’s aside, only two senators would still be elected each election. Given that the two-senator system invariably results in one senator for each major party, I imagine both labour and the liberals might potentially be keen on this idea as a way of diluting the influence of the minor parties/cross-benchers. Would that require a constitutional change, or could the major parties legislate this if they cooperated?

    ReplyDelete
    Replies
    1. Solution to fix entirely unfair population to Senate seat ratios in Australia.

      Add to another entirely unfair population to Senate seat ratio!

      I suppose both territories are likely to grow and at least in the NT case inevitably at some point in the future get state status.

      On a macro level making NT and ACT more unfair will partially neutralize the unfairness of SA and TAS.

      Delete
    2. The thing is the problem is not the number of senators for the smaller states, it's the nexus between the numbers in the Reps and the Senate. We have had 150 reps for quite a few years and the populations in some divisions are becoming very large. We need more lower house seats to bring down the number of electors on average to say around 60k which would give MPs some chance of properly representing their divisions and getting around them in the case of the really big ones. But that is stymied by the nexus. If the number of Reps were to go up to say 170 we would need to go to 14 senators per state. In the smaller states, this would mean nearly the entire population would have someone they know to stay with when visiting Canberra.

      Delete
    3. They tried to get rid of the Nexus Clause in 1967 but it was soundly defeated (the majority of voters voted differently in the 2 referenda that day). Only the DLP was campaigning against it then, if they had not, it might have been defeated by only being passed in 3 states but the DLP opposition probably swung Queensland and may have swung Victoria (the DLP`s strongest state causing what may have been a pass into a 69 to 31 crushing) and the Commonwealth from a majority.

      If it could not pass then it cannot pass now because their are now non-big 2.5 parties across the spectrum to drum up massive opposition in all states.

      There is the added issue that the Nexus also effectively sets the minimum size of the HoR at 72+. Scrapping the nexus clause without some new minimum size for the HoR, like they seem to have tried in 1967 (there is very little information on the internet about this referendum), would reduce the state minimum size of the HoR to only 30 (5 per original state) and the causes issues with proportionality and that may mean that the referendum would need to pass in every state to pass. Good luck convincing the Tasmanians to support scrapping the nexus.

      There is no constitutional obstacle to 13 Senators per state (although that would be a little messy dividing them up into long and short terms). This could be used to increase the Parliament over 2 or 3 elections.

      NT statehood would allow the HoR to be increased without adding Senators to any state but the NT (and thus for fairnees the more populous ACT) would likely need more Senators for that. A referendum or successful overturning of the precedent that the territories` Senators and MPs are not included in the nexus would have a siliar effect.

      Delete
  22. They could legislate it under Section 122 of the Constitution. However an especially unreasonable expansion might be struck down by the High Court, which was not unanimous in dismissing challenges against the initial addition of Territory Senators. Also a hazard of an expansion of the ACT to four Senators for the Coalition would be loss of a seat to the Greens in any double-dissolution, so I would not expect them to support it.

    ReplyDelete
    Replies
    1. Section 122 should be ammended to require territory represenation in both houses of Parliament, mandate inclusion territory representation in the nexus and specifically cap territory representation so that it is not allowed to be higher than if a territory or group of territories represented together are not allowed more representation than if they were an original state.

      Delete
  23. So McKim will not be elected?

    ReplyDelete
    Replies
    1. He probably won't get quota, but that is not the same thing as not winning. I think his chances are pretty good, but there are a few ways he might lose.

      Delete
  24. What are the chances of Singh winning a 6 year term?

    ReplyDelete
    Replies
    1. Very low unless the Senate decides to do something odd. Under the two most-discussed allocation methods she will get a three-year term.

      She could get a six-year term under the order-of-election method if she either polled a quota on primary votes, or polled so close to a quota that BTL votes from Anne Urquhart put Singh over the line (and this would be dependent on Urquhart being elected ahead of Abetz, which may well not happen). However it is now extremely unlikely that Singh's primary vote will be that high.

      A reminder for those not following that issue closely that the Senate decides who gets 6 year terms and who gets 3 year terms, and can do so however it likes.

      Delete
  25. Perhaps this is as good a place as any to ask, and I have a feeling you have explained it already and I missed or misunderstood it.

    When exactly are individual candidates on a ticket eliminated? Is it only when the entire ticket is eliminated?


    Is it more or less low polling ungrouped candidates first, then low polling parties, then entire parties even with a long list of candidates that *might* (just like everyone else I suppose) receive below the line preferences?



    If that's the case I'm not sure who the system advantages. It seems quite complicated (particularly if you run two or more tickets) to work out if you want to hit quota, get just over, get just under, rely on some #2 votes either from someone on your ticket or another ticket and so on.

    ReplyDelete
    Replies
    1. I noticed I didn't mention that I am assuming a mix of BTL and ABL votes such that your lower down candidates are receiving (and sending) some votes from all over the place

      Delete
    2. Although there are above the line votes for parties, all votes (above or below the line) are actually counted as votes for candidates. So an above the line vote for a party starts as a number 1 vote for a particular candidate.

      The first stage is that candidates with more than a quota are elected, and their surpluses distributed, then any candidates who go over quota as a result are elected and their surplus redistributed, and so on.

      Once that cycle finishes, exclusions start from the bottom up. In Tasmania, the first candidates to get the chop will probably be the support candidates from the CEC and ALA, and then after that it will be a mix of ungrouped candidates and the support candidates for the other micro-parties for a while. A party is eliminated when all its candidates have been excluded.

      The support candidates for some of the better-polling parties will actually outlast some of the worse-polling micro-parties entirely. For instance when the CEC are completely eliminated (they will likely be the first grouped party excluded), all the Labor, Liberal and Green candidates will still be in the race, as may be the number 2 JLN and Sex/HEMP ticket candidates.

      Delete
  26. I'm struggling to see how McKim can get elected, given the high number of votes with the conservative microparties. Surely in the distribution of preferences these would likely go to the Liberals, then Labor, rather than the Greens?

    ReplyDelete
    Replies
    1. So far in my sampling, preferences of the right-wing microparties are indeed going more to the Liberals than the Greens, but the catch is that a lot of their preferences are going to Labor as well, and also many of their preferences are exhausting among the right-wing micros without reaching any of the big three. So the Liberals don't seem to be getting a massive leg-up over the Greens on the right micro preferences, and the stronger flows to the Greens on the left-micro preferences (almost none of which go to the Liberals) are more or less cancelling it out.

      We can easily think of parties as "conservative" but that doesn't mean their voters love the Liberals. For instance people have an image of One Nation as an ultra-right party but in fact its voters have a very entrenched national history of only leaning to the Coalition over Labor modestly on a 2PP basis nationally (55:45). Given the 2PP drubbing the Liberals have received in the Reps in Tasmania (probably in excess of 56:44), it's not surprising that even the right micros aren't preferencing them all that strongly.

      Delete
  27. When is the AEC going to start distributing preferences?

    ReplyDelete
    Replies
    1. It's not very long from start to finish once they do - once all the numbers are entered in it takes just minutes. Some versions of the distribution programs in these things even have deliberate delays programmed in to reduce the anticlimax of an instant result!

      I have not heard officially but believe Tasmania is headed towards next week, probably later rather than sooner.

      Delete
    2. If they could upload the ballot papers' preferences somewhere then we could do it ourselves, before the final result, which would be fun or useful for some.

      I doubt that they would want to host it, but perhaps providing a file just once around the mid point to approved parties and news agencies (for a fee if necessary) is possible. Then the ABC website could have a actual preference flows we could have a look at and 'guess' how the other votes will flow.

      Delete
    3. It would be terrific if they did publish preference data updates during the count. I have a program that would allow me to do a full count. However I'll likely be limited to just replicating the official count. Unless they happen to publish preference data before announcing results of the section 282 recount... in which case I'll be attempting to beat them to it. Unfortunately very unlikely.

      Delete
  28. Kevin, from your references to seeing things on the screen I presume you've inveigled your way into the Senate counting centre, as someone's scrutineer or as recognised objective expert observer? As well as reporting on the patterns in the voting, do you have anything to say about the general atmosphere and behaviour inside the Fujibunker? Do the operators generally agree fairly readily and quickly on what numbers they're seeing? Has the OCR usually got it right in the first place? Are the major party scrutineers being pains in the arse or do they generally recognise that a 1 however messy is a 1? And anything else worth mentioning?

    ReplyDelete
    Replies
    1. Yes I am technically an appointed scrutineer, mainly having been appointed so I can gather data and watch the process rather than necessarily making a nuisance of myself. At times I've been there there have usually been between one and three other scrutineers about at a time and they've mainly been data-gathering and objecting to little if anything. The image recognition does fine in most cases but struggles with cases of bad writing (of which there are frightfully many including some appallingly illegible BTLs that go well beyond 12) and also with voters who insist for some reason on only putting part of the number in the square. There are some votes for which the OCR has basically no idea and the operators have to enter everything from scratch. It is difficult to tell whether the operators are getting the same read of the same ballot paper as once it is checked and queried by one it won't necessarily pop up on the other side straight away. I have at times watched the operator who has to deal with the too-hard basket cases that have been raised to a high level and they are often frightfully illegible votes that are beyond help. Having watched a similar process for the Hobart Council elections 2014 (but without the scanning) I am finding the workers in this case to be much faster. There are some points I may look at more closely and post further comments on - my main emphasis has been on preference-sampling and it is very difficult work because of the high proportion of votes that are not relevant to the outcome.

      Delete
    2. Thanks Kevin. Further comments will also be appreciated but that satifies my curiosity for now. But as to those who "insist" on going outside the square, perhaps they have poor sight or arthritic fingers and just can't DO small numbers. I'm trying to imagine a large print and large boxes version of the ballot paper - eughhh, too big, too big!

      Delete
    3. The Roman Numeral all box BTLs must take longest to put in.

      The Senate boxes, especially the BTL boxes, should be twice (or in the case of BTL boxes thrice) as wide because of the larger number of numbers.

      Delete
  29. Are you allowed to take screen shots or videos, KB? (Actually for max democratic transparency it seems to me that the whole thing should be videoed and put up on Youtoob for wonks to study at leisure.)

    And that bit about the progress graph on the wall makes a mockery of the statement I got as to "why we do not provide any other data" (Explanation for those not privy to my emails to Kevin - I asked the AEC whether we could have a daily progress report of how many %-complete the count was, and they said "we tell you the provisional quota, as to why we do not provide any other data see this page" which was simply a page explaining the scanning and scrutiny system, upon which Kevin is now ably reporting. So why they can't tell us how far complete the count (ie the full scan of preferences) is remains a mystery - except that they're all somewhere "on the spectrum" and find communicating with other (ugh) people a bit challenging. Got any alternative explanation/diagnosis, Kevin?

    And if it's taking from a day or 3 after the election until next week to do Tassie with its 350,000 voters, how long will it take in NSW with 4.5 mill? Do you know if they have proportionately more scanners in Sydeney, KB? Are they telling us things like that or is that something else we have to find out from scrutineers?

    ReplyDelete
    Replies
    1. I don't know anything about other states beyond what is bouncing around in said emails. Scrutineers are explicitly not allowed to take photos of individual votes. The AEC staff here have been as informative and cooperative as they can whenever I have asked them for information of any kind.

      Delete
    2. Tasmania has fewer and smaller ballot papers to enter but it has a much higher BTL rate and they have more work per ballot paper although not as much more as under the old system.

      Delete
  30. Good to hear. But their high-ups and the "Public Enquiries Team" who replied to my query certainly don't have the same attitude

    ReplyDelete
  31. As an inveterate contrarian I'm loving seeing Lisa Singh's vote edge closer to 3/4 of a quota in her own right... Could she surpass Catryna Bilyk and end up 4th?

    ReplyDelete
    Replies
    1. This is possible - especially if Singh does better than I expect on the postcount votes that are still to be counted - though my model suggests Bilyk will probably overtake Singh in the cutup and be elected before her. Singh will move very slowly on preferences because she can only receive below the line preferences until Bilyk is elected.

      Had Labor polled more poorly in ticket votes, or received a much worse preference share than I am seeing, then Singh could well have been elected at Bilyk's expense.

      Delete
  32. I wish the AEC broadcast the button push. It'd be like a cross between Lotto and Eurovision. They could artificially slow down the elimination of each candidate like it was a Eurovision vote tally.

    ReplyDelete
  33. Speaking of "how long will it take NSw with its 4.5 million", does anyone know what happens if the count in any state actually isn't complete by the deadline for return of the writs?

    I presume they can't just declare the election void and start again - does the Governor General or state Governor have some way to extend the writ?

    ReplyDelete
  34. It appears that a consequence of delaying the return of the writ for any state(s) is that any Senator in the delayed state(s) who is a Minister would cease to be a Minister as of August 9, although they could of course be reappointed once elected. See http://blogs.abc.net.au/antonygreen/2016/03/senate-electoral-reform-double-dissolutions-and-section-64-of-the-constitution.html

    ReplyDelete
  35. Actually Kevin they could probably be reappointed immediately. The general view among consty lawyers is that under the letter of s 64 a non-MP minister could be reappointed for another 3 months every 3 months. If that were done simply to avoid the minister having to face an election, it would clearly be against the spirit of s 64 (and a G-G with a sense of propriety might say no) but if it was just a temporary expedient in the case of delayed return of writs I can't see any real problem.

    And Andrew, I think a doctrine of necessity would have to apply to the return of the writs - the law does not demand the impossible. The return date on the writ would have to be interpreted as "directory" rather than "mandatory" - demanding substantial compliance rather than strict compliance.

    ReplyDelete
  36. Hi Kevin,

    Canberra Times are talking up Colbeck's chances. Your thoughts on Colbeck getting up over Mckim?

    http://www.canberratimes.com.au/federal-politics/political-news/liberal-richard-colbeck-back-on-track-to-defy-the-factional-fix-and-get-reelected-in-tasmania-20160722-gqbopa.html

    ReplyDelete
    Replies
    1. It's a new system and that article assumes things about preference flows when I don't think anyone involved has actually sampled the preferences. Colbeck may beat McKim in the way suggested, but what I was seeing (as discussed in the article) was a lot of supposedly "right-wing" preferences going to Labor and a lot of One Nation votes just exhausting, and a heck of a lot of voters who just don't think about "right" and "left" the way analysts do. Colbeck "leapfrogging" McKim in terms of starting position (if that was what the term meant) means nothing because Colbeck cannot receive any above the line preferences until Bushby has reached quota, so McKim will race ahead of him until that happens.

      Delete
  37. Hi Kevin, any ideas about whether Victorian Greens senator Janet Rice might lose her seat?

    ReplyDelete
  38. I haven't seen any scrutineering from Victoria but I expect Rice will be re-elected easily. Reasons:

    1. She has a large lead over the micro-parties.
    2. The right-wing micro-party preferences will be going to the Liberals and Hinch as well as each other.
    3. In Tasmania the right-wing micro-party preferences are fragmented. Preferences flow from religious right micro-parties to each other and "rural-right" micro-parties to each other, but not much between the two groups. I expect it would be similar in Victoria.
    4. A lot of micro-party preferences will exhaust.
    5. An important difference with Tasmania: Labor in Victoria will be reduced to its four elected candidates and hence out of the count quickly meaning the Greens should be able to hog preferences from the left-wing micros of which there are many in Victoria.

    I would say Rice's position appears a lot stronger than Nick McKim's.

    ReplyDelete
  39. Thanks Kevin! While I am disappointed in the reality of your conclusion, I salute the breadth and depth of your analysis... Cheers, George

    ReplyDelete
  40. Wouldnt the drop of .029 of liberal quota come off bushby, no Colbeck, as presumably they are above the line votes that are over sampled, and Colbeck would have performed consistently with those later counted BTL votes. This shouldnt change Bushby getting elected earlier, he would just hav less votes to transfer to Colbeck, but at the point McCulloch (or Colbeck) gets excluded that hasn't happened yet.

    ReplyDelete
  41. Thanks, I have now hopefully fixed up my explanation of that bit as there were a few bits that were not well explained and also one error (which was of the sort you suspected). The .029 Q is McKim's improvement relative to the Liberals, most of which comes from McKim having more votes; the Liberals are only down .006. In terms of Colbeck's starting position, yes, all else being equal any drop in Liberal quota comes off Bushby rather than Colbeck, but it also happens that Colbeck himself has dropped .004 Q on primary votes since I did the last projection. This actually has him still very marginally ahead of One Nation at the critical point in the projection (not marginally behind as I first had it), but given the errors involved in sampling it is basically a tossup at that point either way.

    ReplyDelete
  42. O-K, I just looked at the Tas Senate count on the VTR and the number of unapportioned is down to zero at last. How much else do they have to do before they can click on the "Calculate Results" button, Kevin? (Or enter RUN on the command line, or whatever else they do?)

    ReplyDelete
    Replies
    1. There may be more non-booth votes to add but I am not sure there are (there can't be many anyway) - non-booth BTLs do not appear in the system as Unapportioned prior to being added. Once all primaries are finalised there is a 24 hour wait to notify candidates (etc) until the button can be pressed.

      Delete
  43. Ok thanks. 24 hours seems excessive but I can see some notice period is necessary so everyone can gather round and watch. (Or they could live-stream it - it is 2016!)

    ReplyDelete
  44. The delay is also to allow for objections against the final primary totals, though I am not sure on what basis those might occur.

    ReplyDelete
  45. Kevin, I understand there's not a lot of theatre in the button press. Just run the program and all the results spit out soon after?

    ReplyDelete
  46. Often there are ties between unsuccessful candidates that have to be resolved by manual random draws. This typically happens at the very early stages only. Apart from that yes, it takes a few minutes.

    ReplyDelete