jackman.stanford.edu/blog
• 2013 Australian Federal Election
• Pre-election poll averaging last updated Saturday September 7 10:28 am
• Labor 2PP estimate: 46.2% +/- 0.9
• Trajectory over the last 90 days png pdf
• Progress of two-party preferred count: png
• Analysis of polls, betting markets, election results (in progress): here
• State-by-state pre-election "pendula" pdf png

• Bayesian Analysis for the Social Sciences Wiley; Amazon; errata as of 5/24/13

• 113th U.S. Senate
• ideal point estimates pdf csv 11/22/13
• scatterplot against 2012 Obama vote share pdf
• roll call object: RData
• 113th U.S. House
• ideal point estimates pdf csv 12/6/13
• scatterplot vs Obama vote share pdf svg
• roll call object: RData

## Tuesday August 28, 2007

Filed under: Australian Politics — jackman @ 6:19 pm

Another Queensland seat, another arb opportunity. This time its Bowman. ALP/Coalition prices at 3 agencies:

Sportingbet:portlandbet: 2.15/1.60
portlandbet:sportingbet: 3.00/1.50
centrebet: 2.40/1.50

[whoops, got that the wrong way around too.]

The sportingbet/portlandbet arb there is a guaranteed 4.2% 4.35% profit.

Hinkler is still showing a 7.5% 8.1% profit: sportingbet have 3.25 for Labor, while portlandbet have 1.62 for the National Party candidate.

And South Australian races have popped up on Centrebet, thats 51 seats plus the national race they are covering (plus stuff like the interesting election day market etc, Nov 10 the best backed option there). I’ve added these to my averaging/charting etc over here.

### phone versus on-line polling

Filed under: Australian Politics,computing,statistics — jackman @ 3:51 pm

An interesting match-up is coming in the study of Australian public opinion, and coming fast. I missed this story in the Fairfax papers re the launch of You.gov‘s joint venture in Australia with Photon (on the day this story ran we were in the good care of United Airlines, SYD-SFO). I’ve been close to Internet ventures in the past, having had a ringside seat during the creation of Knowledge Networks and Polimetrix here in Stanford/Palo Alto, and have seen some of the competition between traditional polling methods and on-line, and even the competition between different on-line approaches.

The Australian case will be very interesting. Compulsory voting is kind of a two-edged sword. On the one hand, the electorate is the age-eligible citizen population, and the Census tells you pretty much what your sample should look like, nationally, or even state-by-state or electorate-by-electorate. On the other hand, what’s left of the “digital divide” (sounds so quaint in 2007) can be a real bugger when you’re trying to hit the entire population (think older, more rural-regional than urban, less educated people, who skew towards offline).

## Monday August 27, 2007

Filed under: Australian Politics,statistics — jackman @ 1:23 pm

ABS released Basic Community Profiles today. The Parliamentary Secretary to the Treasurer “congratulated the ABS for putting vital Census information back into the hands of all Australians in a very accessible form.”

But ABS is still yet to release data relevant to an interesting question: the number of citizens over the age of 18 (the age-eligible electorate). Total number of citizens, yes. Detailed age breakdown, yes. But the number of adult citizens have disappeared from the Census summaries published thus far, after being there in 1996 and 2001.

Again, I ask, why is such a basic fact about the country not (freely) available from ABS or AEC…?

### Labor probability of winning, averaging across 3 betting agencies

Filed under: Australian Politics — jackman @ 1:26 am

My daily analysis of the betting agencies prices has been expanded to perform some averaging of the 3 betting agencies I am following (centrebet, sportingbet and portlandbet). Averaging across these three agencies, Labor is currently (Sunday pm) tipped to win in 75 seats. The distribution of the agency-averaged, seat-by-seat Labor probabilities is skewed, and so the expected seat count for Labor is 72 seats (i.e., summing the agency-averaged, seat-by-seat probabilities).

In the headline, national market, the average of the 3 implied probabilities of a Labor win is .61. The average of the 3 implied probabilities of a Labor win in Bennelong is .38.

## Sunday August 26, 2007

Filed under: general — jackman @ 10:19 pm

Long-time Australian expatriate, Stanford professor, and a giant of science. 3 degrees from the University of Sydney, PhD Cambridge, had been at Stanford since 1955. Obit from the Stanford Report.

## Saturday August 25, 2007

Filed under: Australian Politics — jackman @ 3:39 pm

As of Saturday afternoon (August 24), Portlandbet.com has Labor at 1.80 in Swan (in from 1.85), sportingbet.com has the Coalition at 2.40. There is still a small arbitrage possibility there (ignoring the possibility of a minor party win), down to 2.9% profit from the 4.5% profit it was sitting at for most of the past week or so. I suspect there is going to be less easy money out there from here on in.

This movement in Swan on portlandbet.com brings it (just barely) into the “tipped to be a Labor retain” column, and means portlandbet.com no longer has Labor tipped to lose any of its seats. Details are at my analysis page.

Sunday morning update: Hinkler is offering 7.5% 8.11% in profits at the moment, arbitraging the 1.62 price for the Coalition at Portlandbet.com against the 3.25 for Labor at sportingbet.com. Again, the possibilities for arbitrage profits are based on the assumption that ALP and the National party candidates as the only horses in the race.

The details of the (closing) arbitrage opportunity in Swan, if you care and/or this arbitrage stuff seems mysterious: consider a total stake of size $$S$$ split in the proportions
$$r = .5714 = 2.4/(2.4+1.8)$$ and $$1 – r = .4286$$. Invest
$$rS$$ at Portlandbet.com on Labor to win Swan and $$(1 – r) S$$ on the Coalition to win Swan at sportingbet.com. If Labor wins the seat: your ROI is
$$[0.8 r - (1-r)]S \approx .029 S$$. If the Coalition wins the seat: ROI is $$[1.40(1-r) - r]S \approx .029 S$$.

Of course, the minor party candidate could (!) win the seat, which is kind of like the “00″ on the roulette wheel…

## Friday August 24, 2007

Filed under: Australian Politics — jackman @ 11:20 pm

Centrebet currently has Labor tipped to win in each of the 5 Tasmanian House of Reps seats, but their “clean sweep of Tasmania” market has prices of 1.90/1.72/301.00 Labor/Neither/Liberal (and so an implied probability of 0.47 of a Labor clean sweep). This is an interesting: the only Tasmanian seats in any real doubt are Braddon (1.50/2.40 ALP/LIB, implied probability of a Labor win is 0.61) and Bass (1.47/2.50, implied probability of a Labor win is 0.63). In the other 3 Tasmanian seats we have Labor win probabilities of .954 (Denison), .78 (Franklin), and .81 (Lyons).

Under assumptions re efficient market pricing, then given these prices & implied probabilities, we can consider outcomes in each seat as independent (i.e., assuming that if the events weren’t independent then traders/punters could/would exploit this information in their positions). Via this reasoning we get the probability of a Labor clean sweep at .23 (multiplying the seat-by-seat probabilities). Contrast the .47 probability of a Labor sweep in the “Tassie sweep” market, suggesting that there are some arbitrage possibilities lurking here. But again, the bookie’s take is probably squashing the prospects of getting these prices/probabilities to converge.

## Wednesday August 22, 2007

Filed under: Australian Politics — jackman @ 12:28 pm

Mumble forwarded me a comment he received from one of his readers:

Just a point on the betting markets (and I am not a great believer in
their predictive powers). It is possible for a majority of seats to have the coalition as
favourites but the market still to be consistent with Labor being
overall favourites. To give a basic example, in a 150 seat parliament, Labor could have a
90% chance in each of 70 seats, and a 45% chance in each of the rest. In
this case you would expect Labor to win .9 x 70 plus .45 x 80, giving 63
+ 36 or 99 seats, despite being favourites in only 70. No inconsistency
there.

In theory, no, there is no inconsistency there (and a nice example, all round numbers). Except thats not what the seat-by-seat probabilities of a Labor win look like, at least at portlandbet.com (which has some interesting prices in some Labor marginals that differ markedly from what you can get elsewhere, but I digress). See my Monte Carlo simulation of Labor’s expected seat tally given the current portlandbet.com seat-by-seat probabilities here.

## Monday August 20, 2007

Filed under: Australian Politics — jackman @ 11:02 pm

How about these puns in the media on the Rudd NYC strip club matter:

“Kevin point-oh-seven” (Annabel Crabb in the SMH, attributed to a producer on ABC’s Insiders, a pun on the title of Rudd’s website)

“its now in the lap of the Gods” (Tony Jones, Lateline, Monday night)

“a redefinition of the term `swinging voters’ ” (Tony Jones, Lateline, Monday night)

“In politics the polls go up, the polls go down and in the end as the saying goes, there’s only one poll that counts. Unless of course it happens to have a scantily clad dancer attached to it.” (Michael Brissenden, 7.30 Report, Monday night)

And there was also this remarkable exchange on the 7.30 Report, which I found almost uniquely/delightfully Australian in that the Leader of the Opposition’s credentials re being able to handle his piss are called into question, and, in response, the Leader of Opposition points to his previous career as a diplomat…

KERRY O’BRIEN: So when something like this happens, people will make their own judgment about how they feel about you being in a strip club, or having drunk too much, or a combination of both. But once something like this breaks, it can then also…

KEVIN RUDD: One led to the other, I’ve got to say.

KERRY O’BRIEN: It can also then become a matter of honesty in the way you deal with it. So was part of your problem on that night that you were a bit of a “two pot screamer”, to use the Aussie vernacular?

KERRY O’BRIEN: No, I mean, you speak to people around Canberra, Kerry, and you’ll find I’m not a big drinker, never have been.

KERRY O’BRIEN: (interrupting) But can you handle it?

KEVIN RUDD: As you know from the operation of… Well, of course, and I’ve also worked as a diplomat, I’ve been in the foreign service. I mean, these things, you get used to it over many years. I’ve served many overseas countries where these things are quite important. But if you look back honestly in your own life, any bloke does, you can point to one or two occasions where you’ve had too much to drink. I remember very clearly as far as…

KERRY O’BRIEN: But there are some of us, Mr Rudd, who’d find more than one or two. But in my experience, if I can put it like that, if a person usually has to consume a great deal of alcohol to suffer memory loss from a night on the tiles.

### betting market trivia

Filed under: Australian Politics — jackman @ 6:08 pm

I’ve added a few more graphs to the collection available here:

1. Scatterplot of Labor and Coalition prices, all 150 seats from portlandbet.com PDF; study this carefully and you’ll see the house’s profit margin quite vividly (see Mumble’s comment re the vig).
2. Summary of ALP seat count, given seat-by-seat probabilities from the prices offered by portlandbet.com in all 150 seats. PDF This is kind of fun; you read off the probabilities of a Labor win in each seat, and then flip 150 coins, each with probability equal to the corresponding seat-specific probability of a Labor win; count how many coins up for Labor, repeat many times, inducing a distribution over the number of Labor seats given the seat-by-seat probabilities. This Monte Carlo exercise treats the outcomes in seat as conditionally independent, which is equivalent to a strong assumption of market efficiency; i.e., given the prices on offer in each seat, we treat the outcome in each seat as conditionally independent of the outcome in any other seat (i.e., no cross-seat arbitrage possibilities, since to the extent the outcome in seat A is correlated with the outcome in seat B, we’re assuming that information is already factored in the prices offered in the seats). Anyway, given the seat-by-seat portlandbet.com prices, the probability that Labor wins 76 or more seats is only .20, given the prices as at 5pm Monday.

Sportingbet added Sturt (SA) to their seat-by-seat markets on August 16. They are now covering 41 seats. As of 5pm yesterday (Sunday), they had Labor at $3 in that seat, the Coalition at$1.40, and $26 for any other candidate, equivalent to a .31 chance that Labor wins the seat (returned Chris Pyne, Lib, with 56.8% 2PP in 2004). Portlandbet.com had pretty much the same:$3 to \$1.32.

Generally, there has been very little day-to-day change at either betting agency over the week that I’ve been looking closely, and virtually none at sportingbet.com, save for the addition of extra SA seat. Its a Newspoll Tuesday tomorrow; that might generate some movement?

You’d have to think that some of those marginal seat markets are very thin. With prices like 1.85 to 1.85 as we’re seeing from portlandbet.com in Swan (WA), it could well be tough to attract punters; imho, there are many more attractive propositions out there for the politically-inclined punter (again, see the comments re the vig) and the betting agencies probably don’t feel a lot of pressure to make the prices more attractive in those quieter markets, at least not just yet. At the same time, sportingbet is offering 1.50/2.40 ALP/LP in Swan; one wonders if/when these differences will get washed out.

Next Page »