Data is king.
In a world of asymmetric information, a data hound such as I relies on the evidence around me to form views and opinions. While my innovative Monte Carlo simulations are straight data crunchers, there is always the risk of "Garbage in - Garbage out". That is, if my initial primary vote estimates are erroneous, there's no point.
My first step is to look at polling data, particularly the BludgerTrack.
Since doing this, several intelligent people have provided me with ideas for improving my modelling - this is great! Indeed, this is why I have so meticulously outlined my rationale to encourage others to critique and help me improve the data estimates that feed into the Monte Carlo sims.
The following information has recently become apparent:
1. PUP is outspending the other parties
No surprise here, but the level of outspending seems very high by recent standards
2. PUP had a good poll with Morgan
Sure, PUP polled 10.5% in a recent Morgan poll, but this was a sample size of just 247 WA voters, implying an error of ~4%
3. "Internal polling" suggests PUP vote is ~10%
I don't normally rely too much on what's reported as "internal polling", but I have been provided credible evidence from a trusted source that this is actually the case.
Each of these on its own is not a game-breaker. But in the world of Bayesian probability, each new data point builds on our last and we need to modify our conclusions based on all the available data. Is there a Palmer tsunami on? For short, we'll call it the Palmernami.
Incremental analysis
Using the base analysis tools previously referred to, the PUP vote was estimated to be 3.5% owing to a general post election decline in PUP vote as per BludgerTrack. So this becomes a starting point. Progressively, I now modify the PUP% while simultaneously reducing the PUP variation applied by the Monte Carlo Model to +/-10%. For these samples, n=500, and all reduced vote is coming proportionally from other parties. No other changes to methodology.
Senators elected with PUP=3.5%
LIB: 2.91
NAT: 0.07
ALP: 2.06
GRN: 0.93
PUP: 0.006
OTH: 0.02 (DEM, WIKI)
Senators elected with PUP=5%
LIB: 2.86
NAT: 0.04
ALP: 2.03
GRN: 0.95
PUP: 0.10
OTH: 0.02 (DEM)
Senators elected with PUP=7.5%
LIB: 2.39
NAT: 0
ALP: 2.002
GRN: 0.96
PUP: 0.64
OTH: 0.002 (DEM)
Senators elected with PUP=10%
LIB: 2.25
NAT: 0
ALP: 1.996
GRN: 0.91
PUP: 0.84
OTH: 0
Senators elected with PUP=12.5%
LIB: 2.22
NAT: 0.002
ALP: 1.97
GRN: 0.81
PUP: 0.995
OTH: 0
Senators elected with PUP=15%
LIB: 2.33
NAT: 0.05
ALP: 1.95
GRN: 0.67
PUP: 1.00
OTH: 0
(odd - I suspect this LIB upturn likely produced by increasing PUP vote flowing back to LIB via preferences)
Note: Still no HEMP. Not even in a single scenario... I'll do that incremental analysis at some other stage this week I hope.
Just to clarify... The assumption in this analysis is that PUP takes votes evenly from across the board, and doesn't draw more from those who would have voted LIB... correct?
ReplyDeleteYou do have to start somewhere, though I have trouble imagining too many GRN voters thinking "Dammit, he's right! That carbon tax IS bad for WA".
We're running out of time here but I'd be keen to see a new set where most of PUP's vote comes from the LIB, NAT and OTH, with less from ALP and much less from GRN.
Yes - the assumption is correct. What makes it somewhat valid is that the PUP preference rate to LIB/ALP nationally was close to 50%, so it's fair to assume support proportional.
DeleteWhile not many GRN voters would switch directly to PUP, the basic assumption I've made could be seen (if true) where you had voters swapping "half-way" so to speak so a GRN-ALP and a separate ALP-PUP
There's something instructive in the averages and how the game flips as PUP% increases.
ReplyDeleteAt 7.5%, it looks like the vast majority of scenarios elect 2 ALP, 2 LIB and one GRN, with PUP and LIB fighting for the last spot in most cases.
At 12.5 and 15%, then most cases elect 2 ALP, 2 LIB, 1 PUP and then it's LIB vs GRN for the final spot.
10% is a transition space where 2 ALP and 2 LIB are usually elected, and rather than one seat the battle opens up to two seats between GRN, LIB and PUP.
The media assumption always seems to be that PUP would be threatening the GRN seat but this work suggests that this is seldom the case.
If you take into account the suggestion about PUP pinching votes mainly from the LIBs - this might be even more strongly demonstrated.
Yes, indeed 7% seems to be the number PUP need, and polling (above) seems to make it clear it will get this level.
Delete