Crikey

The former boss of Newspoll Sol Lebovic likens opinion polls to the scoreboard in a football game. The numbers show who is leading but not who will win. “Polls aren’t necessarily predictive, nor should they be”, he wrote in The Australian on Saturday. “As we get closer to the election there is a greater chance they will reflect the final result. However, just as in football, the winner is not necessarily the side that is leading before full-time. The most exciting football games, or elections, are in fact those where we do see late changes.”

And late changes are clearly what a lot of people expect as the betting markets tell a very different story to the opinion polls. Believe the average of the polls showing Labor with around 56 or 57% of the two party preferred vote is what will happen on election day and, using the normal margin of error for polls, Labor would be assessed at well over a 90% probability of winning. The betting markets as measured by the Crikey Election Indicator (based on the Betfair betting exchange) puts the Labor probability at only 67%.

My whiz on all these matters, the man whose horse ratings in the London Sun years ago had the paper describe him as “Superbrain - the man with a computer for a mind” and who earned the title of the greatest horse handicapper in the world from Las Vegas oddsmaker Roxy Roxburugh, sent me a note this morning which might help those of you wishing to put some excitement in to election watching by having a bet:

I have been studying your seat by seat odds paper.

I have tried to calculate a standard deviation that links the poll findings to the betting. I am not 100% sure of the accuracy of what I have done, but I think it’s ok.

My calculation goes like this. The betting says 67% to 33% and that translates to a difference of 0.866 standard deviations under the normal curve.

The polls say 57% to 43%, so in order to link the two pieces of information we have to say one standard deviation in the betting market equals 16.2% of polling information. (57-43)/0.866.

That does seem very high, given the previous history of the polls. But that appears to be the view that the betting market has about the polling information.

I have applied that higher standard deviation to the “at risk” seats and get a more cautious set of odds.

Of course the polls might deserve to be accorded a smaller standard error and the market is being too cautious, but as you know I am a great respecter of betting markets.

One other minor point, I am not sure that the 2004 election margins are the best starting points for Wentworth and Bennelong.

The betting says:

Coalition: 66.7
ALP: 33.3

What does that say in Standard Deviations?

0.433: 0.667493
-0.433: 0.332507

It says that the betting market assessment of the difference is 0.433 + 0.433 = 0.866 standard deviations.

The polls say the two party preferred is:

Coalition: 43
ALP: 57

In order to link the two pieces of information don’t we then need to say (Stdev = pcent):

0.433 = 7%

or

0.866 = 14%

So, in betting market terms, the standard error or one standard deviation is 1/0.866 *14 = 16.166. Applying the standard error of 16.166 to the most marginal seats we get:

Seat	Held By	Present Margin	Forecast Swing	Forecast Margin	ALP % Chance	Mean	Revised ALP % Chance
NSW
Parramatta	Lab	-1.1	9.2	8.1	94.7%	58.1	69.2%
Wentworth	Lib	-2.6	9.2	6.6	90.7%	56.6	65.8%
Lindsay	Lib	-2.9	9.2	6.3	89.6%	56.3	65.2%
Eden-Monaro	Lib	-3.3	9.2	5.9	88.1%	55.9	64.2%
Bennelong	Lib	-4.0	9.2	5.2	85.1%	55.2	62.6%
Dobell	Lib	-4.8	9.2	4.4	81.1%	54.4	60.7%
Page	Nat	-5.5	9.2	3.7	77.0%	53.7	59.1%
Cowper	Nat	-6.6	9.2	2.6	69.8%	52.6	56.4%
Paterson	Lib	-6.8	9.2	2.4	68.4%	52.4	55.9%
Robertson	Lib	-6.9	9.2	2.3	67.7%	52.3	55.7%
Hughes	Lib	-8.8	9.2	0.4	53.2%	50.4	51.0%
NT
Solomon	Lib	-2.9	8.8	5.9	88.1%	55.9	64.2%
QLD
Bonner	Lib	-0.6	9.1	8.5	95.5%	58.5	70.0%
Moreton	Lib	-2.8	9.1	6.3	89.6%	56.3	65.2%
Blair	Lib	-5.7	9.1	3.4	75.2%	53.4	58.3%
Herbert	Lib	-6.1	9.1	3.0	72.6%	53.0	57.4%
Longman	Lib	-6.6	9.1	2.5	69.1%	52.5	56.1%
Flynn	Nat	-7.8	9.1	1.3	60.3%	51.3	53.2%
Petrie	Lib	-7.9	9.1	1.2	59.5%	51.2	53.0%
Hinkler	Nat	-8.8	9.1	0.3	52.4%	50.3	50.7%
Bowman	Lib	-8.9	9.1	0.2	51.6%	50.2	50.5%
SA
Kingston	Lib	-0.1	9.4	9.3	96.9%	59.3	71.7%
Wakefield	Lib	-0.7	9.4	8.7	95.9%	58.7	70.5%
Makin	Lib	-1.0	9.4	8.4	95.4%	58.4	69.8%
Boothby	Lib	-5.4	9.4	4.0	78.8%	54.0	59.8%
Sturt	Lib	-6.8	9.4	2.6	69.8%	52.6	56.4%
TAS
Braddon	Lib	-1.2	8.8	7.6	93.6%	57.6	68.1%
Bass	Lib	-2.7	8.8	6.1	88.9%	56.1	64.7%
VIC
Deakin	Lib	-5.0	11	6.0	88.5%	56.0	64.5%
McMillan	Lib	-5.0	11	6.0	88.5%	56.0	64.5%
Corangamite	Lib	-5.4	11	5.6	86.9%	55.6	63.5%
LaTrobe	Lib	-5.9	11	5.1	84.6%	55.1	62.4%
McEwen	Lib	-6.5	11	4.5	81.6%	54.5	61.0%
Gippsland	Nat	-7.8	11	3.2	73.9%	53.2	57.8%
Higgins	Lib	-8.8	11	2.2	67.0%	52.2	55.4%
Dunkley	Lib	-9.4	11	1.6	62.6%	51.6	53.9%
Kooyong	Lib	-9.6	11	1.4	61.0%	51.4	53.5%
Goldstein	Lib	-10.1	11	0.9	57.1%	50.9	52.2%
Menzies	Lib	-10.7	11	0.3	52.4%	50.3	50.7%
WA
Hasluck	Lib	-1.9	4.4	2.5	69.1%	52.5	56.1%
Stirling	Lib	-2.1	4.4	2.3	67.7%	52.3	55.7%