Latest Monte Carlo simulations using updated BludgerTrack estimations

What a week it's been. Saturday night and I fnially have time for another update.

The timing of this is propitious as yesterday saw a significant move in the polling estimates produced by Poll Bludger's "BludgerTrack"

In this weekly update, the LNP 2PP moved from 49.9% to 48.2%, on the back of weak polling performance by the Government. Additional information is that the Greens vote is slightly up and the LNP performance in WA is down even more than the trend.

I have also made the following modifications to my model:
1. Wirrpanda effect
I have removed the +1% variation to LIB and -1% to NAT. While I still believe that David Wirrpanda positively increased the NAT vote, he did this in Group U. This time, NAT has drawn the much more favourable Group B. Given the fact that NAT and LIB are imperfect substitutes*, I would expect a small proportion of LIB voters to instead vote NAT instead. Rather than modelling a different "LIB-NAT Substitution Effect", I've just deleted from my model the Wirrpanda effect.
*I use the term imperfect substitutes. LDP Senator-Elect David Leyonhjelm considers NAT as a front for LIB - Certainly not true in WA!

2. Could Not Find effect
Again, I still believe that the OTH vote will be lower this time due to the focus on a Senate only election. But the model in telling me last week that just 7.8% would vote OTH just seemed unrealistically low. So rather than arbitrarily increasing the vote, I decreased the reduction (double negative, sorry) from 2/3 to 0. (ie, deleting the effect). Perhaps another piece of supporting evidence to the OTH not decreasing by much may be the higher number of groups in this election - an even smaller font than last time.

3. Double Monte Carlo
Previously, I was applying a variation to each party's vote. While this was successful in modelling each party's own vote and variations therein, the model was actually weak in modelling the total OTH vote, as applying 28 individual variations will mainly cancel itself out. Any fair scenario analysis will consider the impact of the total OTH vote varying significantly from its mean.

In my update last week, I applied the following variations to parties' votes:
LIB: +/-15%
ALP: +/-20% (I suspect I may have typoed by stating I'd assumed a +/-15% for ALP. Apols)
GRN: +/-30%
All others individually: +/-50%

This week, my revised methodology applies two variations to minor parties' votes:
a) Universal variation: +/-40% (not to LIB, NAT, ALP, GRN, PUP)
b) Individual variation: +/-30%
and where the two variations are applied multiplicatively.

This ensures that the minor party vote fluctuates, and that each individual minor party's vote also varies within the varying OTH vote.

4. Revised minor party votes
I have adjusted minor party votes to increase the number of different minor party levels. Minor party vote is based on the vote the party got last time, but arbitrarily adjusted for Group.


Major party votes:

	Model	Last wk	2013
LIB:	37.1%	-4.5%	-2.1%
NAT:	4.5%	+0.6%	-0.6%
ALP:	31.5%	+0.6%	+5.0%
GRN:	12.3%	+0.1%	+2.8%
PUP:	3.5%	-	-1.5%
OTH:	11.0%	+3.2%	-3.6%

Minor party votes

Group	Party	Est Vote	'13 Grp	2013
A	WIKI	1.30%	L	0.75%
AE	SEX	1.13%	I	1.49%
J	LDP	0.87%	B	3.43%
C	IND	0.87%	-	-
S	SFP	0.78%	G	1.04%
N	FF	0.61%	N	0.67%
G	AMEP	0.61%	W	0.59%
AD	AJP	0.52%	T	0.74%
D	DEM	0.43%	R	0.29%
T	HEMP	0.43%	D	1.07%
E	PIR	0.43%	-	-
W	AFLP	0.35%	V	0.44%
X	GOD	0.35%	C	1.64%
AB	DLP	0.26%	-	-
AC	KAP	0.26%	M	0.30%
Z	RUA	0.26%	Y	0.29%
V	SMOK	0.26%	A	0.67%
I	VEP	0.26%	-	-
AG	ODR	0.17%	Q	0.17%
Q	SPORT	0.17%	X	0.23%
L	BAP	0.09%	-	-
H	FREE	0.09%	-	-
M	MUT	0.09%	-	-
U	REP	0.09%	-	-
AF	SA	0.09%	E	0.09%
Y	SEC	0.09%	J	0.11%
O	SUS	0.09%	P	0.10%
K	VOICE	0.09%	H	0.09%

 

Outcomes:
Despite significant change in methodology, there is surprisingly little change in outcomes. LIB is now not quite certain to win 3 Senators, while ALP is modelled winning a surprising 3 Senators on 10% of occasions. This is unlikely - it would require ALP vote exceeding 35%. While this is within the Monte Carlo simulated range, it is well beyond what is likely.

Very likely (>90% likelihood):
LIB: 3 (3rd Senator with 91% likelihood)
ALP: 2 (100%)
GRN: 1 (93%)

Unlikely:
NAT: 7%
ALP3: 6%
WIKI: 1.2%
PUP: 0.6%
DEM: 0.5%

In terms of a left-right breakdown, note that my modelling does not show NAT being elected when 3 LIB are elected - the NAT is an unlikely substitute for the 3rd LIB Senator.

No HEMP?
I am now modelling HEMP at 0.43%. This is much lower than the 1.07% it got last time. It's worth considering HEMP's vote in other states in the 2013 election:
WA: 1.07%, Group D
QLD: 0.90%, Group E
VIC: 0.59%, Group F
NSW: 0.69%, Group G
SA: 0.58%, Group S (33 groups)
TAS: 0.51%, Group K (23 groups)
WA: ??%, Group T (33 groups)
So while 0.43% is slightly less than what it polled in TAS and SA, it's not massively so and only indicative of the fact that total OTH vote is forecast to decrease at this election.


Conclusions:
The more things change, the more they stay the same. I've revamped my modelling, applied a different share of minor party vote, used latest polling data, and everything is pointing to a 3-2-1 outcome.

I'm still not getting HEMP elected in even 0.1% of scenarios, but it does tend to hoover the late-elimination minors if they can last that long. But given they're moving from Group D to Group T surely its vote will not exceed 0.7%?