Newspoll released 1500+ sample poll of South Australian voting intentions today in The Oz that you can take a squiz at over here. What this allows us to do is use that data to run one of our monte carlo election simulations on those figures to get a bit of a handle on the most likely outcome of today’s result (assuming that the election result comes in within the MoE of the Newspoll figures) and attach some implied probabilities to possible outcomes.
The headline results of the simulation come in like this:
The most likely result is Labor winning 21 seats (where 24 is needed to govern outright), which would mean a total result of 21 ALP, 21 Liberal and 5 independents. You can’t get much more of a split parliament than that!
If we move across to the cumulative implied probabilities that measure the probability of the ALP winning at least X number of seats, this is what we get:
To read the chart, simply choose a number of seats at the bottom, go up to where the red line intersects and trace across to the left axis. That gives us the implied probability of the ALP winning at least that many seats today based on the Newspoll this morning. Tabling up the key numbers we get:
So the most likely outcome, based on the Newspoll, is a perfectly hung parliament with a 21/21/5 split between ALP/Liberal/Independents. However, nearly all the uncertainty in the numbers is actually on the upside for the ALP – meaning the second most likely result by far is 22/20/5 with the third and fourth being split 20/22/ 5 and 23/19/5
If the election result matches the polls, it more likely than not that the ALP will form a minority government – but those probabilities aren’t so great as to be a sure thing, just a likely thing.
25 Comments
Isn’t the third most likely result from the top graph, the ALP winning 20 seats?
I predict that the Liberals will win the seat of Morialta.
It is Grog -I mangled that in my rush out the door.
Didn’t stop the ABC news this morning, on the basis of the same Newspoll, predict that ALP would get a thrashing.
The ABC’s wishful thinking. There must be so really bitter Labor haters in the ABC now days, unfortunately they seem to have control over content.
Hey Possum, with your simulation, you have 5 independents. Does the simulation take into account the possibility of Kris Hanna dropping below the Libs, leading to a possible Labor gain in Mitchell?
Additionally, is the 5th independent Karlene Maywald?
It does Al (and Maywald is counted as an Indie here). I kept getting an average of 4.6 independents per iteration, rounding out as 5.
The media this morning is going on about how all this is bad for Rudd. Surely this is about old state governments, and with how this country tends to like their state and federal governments of different persuasions, it might be good news for Rudd if there’s more Lib state governments? Is the media just in a let’s bash Rudd mood?
Thanks Poss. At this point a minority government on either side would be a good outcome to me. If the Libs win I don’t want them to blow another billion on that stupid stadium and RAH plan.
what difference would 51/49 lib or 50/50 or 51/49 alp make to yr simulations
still reckon 24 seats alp is possible
Mick,
An ALP TPP of 49 makes it 22 seats most likely with 23 seats next most likely, followed by 21 seats.
An ALP TPP of 50 makes it 23 seats most likely with 21 seats next most likely, followed by 24 seats.
An ALP TPP of 51 makes it 24 seats most likely with 23 seats next most likely, followed by 25 seats.
kitkat – some in the media will gaze deeply into their navels about the federal implications of state results. Column miles will be written – and none if it will amount to a hill of electoral beans.
Poss
Did you use newspolls “Adelaide” and “Rest of SA” numbers?
Oh getting so bored, 6.00 hurry up!
The Liberals and Family First are crying foul. There were fliers purportedly from Family First which were ALP documents given out in marginal seats. How would this affect the votes if all Family First voters followed them?
“You can’t get much more of a split parliament than that!”
Does that make it a well hung parliament?
I was right about Morialta – the first seat to change hands.
With 70% counted, the ABC predicts 25 seats to Labor, a result on the simulation with just a 2% probability. Any comments about the predictive capacity of pre-election polls, Scott?
hmmm. According to the ABC, the actual primary results (at 72 percent of the count) came in within the newspoll MoE of 2.5%. Yet Labor has 25 seats: the cdf above gave that a 2% chance. What gives?
Toypoodle, I think the probability indicator assumes a uniform swing. The swing tonight was anything but uniform.
Via Poll Bludger and Antony Green comes this description of the non-uniformity of the swing.
“The lack of bang for the Liberals’ buck is illustrated by some numbers crunched by Antony Green: the swing was 1.7 per cent in marginal Labor seats, 7.7 per cent in safe Labor seats and 11.3 per cent in very safe Labor seats. There was also an average 8.3 per cent swing in the 14 seats already held by the Liberals.”
Possum
As noted above, the actual outcome looks like it will be an 8% to 2%, or possibly even lower, probability event, with significant non-uniform seat swings, no doubt influenced by a targetted electoral campaign (amongst various other local factors).
Have you seen any analysis which compares actual outcomes of various elections over time, with the probabilities generated by “accurate” polls (I imagine, late polls, where 2PP vote was within margin of error), to see whether actual seat outcomes are, over time, becoming “less probable” – potentially to point to the efficacy of seat targetted electoral campaigns?
Any way of factoring the prob of smaller swings in marginals into the model?
I’ve got a new post up that looks at some the stuff canvasses here in comments.
http://blogs.crikey.com.au/pollytics/2010/03/22/sa-election-%e2%80%93-skew-not-skewered/
Toypoodle,
Yep – feeding in marginal seat polling that we get in Fed election campaigns would do the job! For State elections – probably not.
Lao – When I was developing the sim, I ran it back over 5 federal elections and about half a dozen state ones to make sure it was accurate. It was.
The non-uniformity where there’s a very strong relationship between two party preferred and swing is what knocked it around. The new post gets into that.