Beating the Brownlow – 2012

Last year, I wrote about beating the Brownlow and promptly haemorrhaged cash by submitting my worst gambling performance on the AFL’s night of nights since first receiving word that you could actually bet on the thing. As noted in that preview, the 2010 result barely squeezed onto the profitable side of the ledger, but last year gave back roughly the amount of money won in 2009 and 2008 combined. As over-thinkers like me are prone to doing, a thorough investigation was undertaken – the good news is that the problem boils down to simple variability in probabilities, and can easily be better allowed for.

Before we get to that, a few primers on the methodology for coming up with ratings (tinkered with slightly for this year) and some headline points to note:

Firstly and most importantly, this is not one of those ‘leaked Brownlow votes’ pieces of analysis. Rather, it uses publicly available votes from the media fraternity, along with some adjustments (explained in more detail later on) to estimate a series of ‘Brownlow ratings’, which are then converted into recommended process for various betting markets.

Secondly, despite some of the short prices on offer and cases where a person is seemingly a ‘sure thing’ to win a particular market, no such certainties exist. In 2008, Nick Riewoldt (at odds of $1.20 to top the St Kilda vote) and Dean Cox ($1.30 to top the West Coast vote) were both rolled. Even more amazingly in 2009, Dane Swan (at odds of $1.03 to top the Collingwood vote) was edged out by Scott Pendlebury, while Sam Mitchell (also at odds of $1.03 to top the Hawthorn vote) tied on 13 votes with Brad Sewell. These should act as cautionary tales for anyone thinking of loading up on one of the many short-priced favourites on offer in 2012 or putting them all into a multi.

And finally, whilst the ratings system does its best to take into account discrepancies between the views of journalists awarding player votes and the views of umpires doing same, it is far from definitive (as last year’s results rammed home, to my eternal chagrin).

 

Brownlow ratings methodology – a four-step process

Step 1:

Take the match-by-match publicly available votes from journalists and convert them into a 3-2-1s for each game, then add them up to establish a quasi-Brownlow count as voted on by journalists.

By and large you could simply look at the ‘Player Of The Year’ leader boards to get a sense of the short list for the overall winner, but to analyse club-by-club markets and the like you need to delve into the detail.

 

Step 2:

Using the outcomes from Step 1, add the number of times each player ‘polled’ to the ‘votes’ they received.

This one delves a little more into punting philosophy and can best be explained by example. Suppose you’re evaluating a head-to-head market where according to Step 1, one player has polled 12 votes in four games (four best-on-ground performances) and another played has polled 11 votes across seven games. Who should be the favourite in that market?

I would argue that the 11 votes across seven games player should be slight favourite, and here’s why – in polling seven times, there is some upside risk that some of the 1-vote and 2-vote games could be elevated into 2-vote or 3-vote games, and thus their ‘upper limit’ may be 15 votes.

Compare this to the four best-on-grounds player. Granted, there may be some games where they didn’t poll with journalists but could do so with umpires, but their upside risk is limited while their downside risk (i.e. the chance that some of the 3-vote games turn into 1-vote or 2-vote games) is more pronounced.

Adding the votes plus times polled together gives 12 + 4 = 16 versus 11 + 7 = 18, which I think from a ratings perspective offers a more robust basis for evaluating risk/ reward in each market, which as punters is fundamentally what we’re trying to do!

 

Step 3:

Use ‘Step 1’ estimates and observed outcomes from 2008 to 2011 to estimate any bias between the votes of journalists and the votes of umpires.

This step is the product of being stung in successive years by either (a) betting on a player whom journalists clearly rate more highly than umpires and/ or (b) betting against a player whom umpires clearly rate more highly than journalists.

What we need here is a ratio of umpire votes to ‘adjusted journalist votes’, which will be referred to as a ‘Brownlow Ratio’ because I lack imagination.

In order to avoid spurious cases where umpires awarded two votes to one from the journalist (causing a Brownlow Ratio of 200 per cent), analysis has been restricted to cases where across the three years there were at least ten votes allocated by either the umpires or the journalists (and in many cases both).

For example, Adelaide’s Scott Thompson received 53 Brownlow votes over the past three years (15 in 2008, 11 in 2009, 15 in 2010 and 12 in 2011) compared to 39 ‘journalist votes’ over the same period (11, three, 14 and 11 respectively). His Brownlow Ratio is therefore 53/39 = 136 per cent. On that analysis alone, if the journalists allocated him 11 votes, you could expect the umpires to award him around 15 votes.

Now compare this to Hawthorn’s Luke Hodge. Hodge received 42 Brownlow votes over the past three years (two, seven, 16 and 17 respectively in the last four years) compared to 59 ‘journalist votes’ over the same period (11, seven, 26 and 15). His Brownlow Ratio is therefore 42/59 = 71 per cent. On that analysis alone, if the journalists allocated him 17 votes, you could expect the umpires to award him only around 12 votes.

Some of these ratios can get pretty skewed and are often distorted by a single year out of the four. Moreover, last year provided a number of high profile cases that bucked the trend of the previous three years. Last year I set a range where Brownlow Ratio could not drop below 80 per cent, nor exceed 125 per cent. In light of last year’s changes in the dynamic, I’ve narrowed the range to 85-115 per cent for 2012.

Most if not all of you aren’t reading this to evaluate statistical methods, so let’s put some real life examples to the test as per below:

Key players for whom 2008-2011 ‘Brownlow Ratio’ set at maximum 115 per cent

Scott Thompson – Adelaide (53 Brownlow votes, 39 ‘journalist votes’)

Patrick Dangerfield – Adelaide (18, 14)

Scott Pendlebury – Collingwood (65, 39)

Jobe Watson – Essendon – (46, 40)

Joel Selwood – Geelong (73, 51)

Gary Ablett – Gold Coast (101, 87)

Lenny Hayes – St Kilda (54, 46)

Matthew Priddis – West Coast (44, 25)

Key players for whom 2008-2011 ‘Brownlow Ratio’ set at minimum of 85 per cent

Brent Harvey – North Melbourne (47 Brownlow votes, 62 ‘journalist votes’)

Dean Cox – West Coast (26, 45)

Key players for whom 2008-2011 ‘Brownlow Ratio’ lies between 85 and 115 per cent

Simon Black – Brisbane – 113 per cent (54 Brownlow votes, 482 ‘journalist votes’)

Chris Judd – Carlton – 114 per cent (91, 80)

Marc Murphy – Carlton – 93 per cent (50, 54)

Dane Swan – Collingwood – 101 per cent (82, 81)

Matthew Pavlich – Fremantle – 98 per cent (45, 46)

Steve Johnson – Geelong – 86 per cent (36, 42)

Sam Mitchell – Hawthorn – 103 per cent (73, 71)

Lance Franklin – Hawthorn – 100 per cent (54, 54)

Nick Dal Santo – St Kilda – 108 per cent (55, 51)

* Note: Trent Cotchin has only had one year of significantly polling (2011), where he received 15 Brownlow votes and 16 ‘journalist’ votes. The small sample size worries me, so I’m assigning him a 105 per cent ratio, inflated purely because everyone else already seems to think he’ll get more votes than I think.

 

Step 4:

Take (Step 1 + Step 2) x Step 3 ratio and scale it back to ensure that only 1,188 rating points (i.e. 6 points per game x 198 games) are allocated in total.

Strictly speaking this isn’t necessary, but since a total of 1,188 votes are allocated via applying the 3-2-1 process for each of the 198 home and away season matches, scaling the ratings back to this level brings them closer to approximating how votes may be allocated to each player.

 

Converting ratings into prices, and prices into recommended bets

So far this looks much like last year’s preview, but the promised change in methodology arrives at this point.

Last year I said “As a rule of thumb, I would not recommend a bet unless the available price was 20 per cent or more above the rated price for those rated around $2.00 or shorter, and 50 per cent above the rated price once you start to look at longer-priced options.”

In trawling through old records, the trend every year became clear – I made money when finding the big overlay that also had a high predicted probability, but lost money on ‘overlays’ with lower chances of coming off.

To try and get past that, this year I’m adding one new criterion – bets can only be recommended where the available price exceeds the rated price by one standard deviation or more.

To explain what that means, consider a case where I’ve rated someone $1.25, or an 80% probability of winning. For a single event (like winning a Brownlow market), the standard deviation of something with probability p is p x (1-p). Or in this case, standard deviation = 80% x 20% = 16%. That means that one standard deviation either side of my rating is either 96% (around $1.04) or 64% (around $1.56).

Compare that to a case where I’ve rated someone $2.50, or a 40% probability of winning. The standard deviation is 40% x 60% = 24%, so one standard deviation in either would take us to 64% (around $1.56) or 16% (around $6.25).

Notice how much the required overlay goes up as the probability of winning goes down – $1.56 is only around a quarter again on top of $1.25, but $6.25 is a 150% increase on $2.50. When you’re dealing with variability associated with:

(1) The thoughts of the journalist votes I used to frame the ratings;

(2) The different outcomes you might get by simply changing where you source the votes from; and

(3) The different results you get again when switching from those watching from the stands to the umpire themselves,

Demanding a greater overlay as your rated price increases is probably a good idea.

Using this rule, here is a summary of the relationship between some rated prices and what needs to be on offer for me to recommend a bet in 2012:

$1.20 rated price; $1.44 required price for a recommended bet

$1.40 rated; $1.96 required

$1.60 rated; $2.56 required

$1.80 rated; $3.24 required

$2.00 rated; $4.00 required

$2.50 rated; $6.25 required

$3.00 rated; $9.00 required

$3.50 rated; $12.25 required

$4.00 rated; $16.00 required

 

Brownlow Ratings and recommended bets

Okay, we’ve gone through the details of how these ratings have been constructed, now let’s get to the good stuff. Who should you back and at what prices should you do so? Let’s start with the overall winner and Round 12 overall leader, before moving into the team-by-team and finally the custom group and head-to-head markets.

 

Overall winner (* indicates a player ineligible to win the Brownlow Medal due to suspension)

Ratings:  Watson 36.0, Dangerfield 30.0, Ablett 29.0, Cotchin 25.0, Thompson 24.0, Pendlebury 24.0, Joel Selwood 24.0, Hayes 24.0, Swan 19.5, Mitchell 19.0

Rated odds:  Watson $5.50, Dangerfield $7, Ablett $8, Cotchin $13, Thompson $17, Joel Selwood $17, Hayes $17, Pendlebury $21 (he missed a few weeks, so has less upside than the other on a 24.0 rating), Swan $26, Mitchell $31

Summary: Nothing much to report here. Ablett and Cotchin both seem huge unders to me, most likely as a result of ongoing public discussion of their chances.  

Recommended bet:  Nil

 

Top five

Ratings:  As above

Rated odds:  Watson $1.20, Dangerfield $1.60, Ablett $1.70, Cotchin $2.80, Thompson $3.20, Joel Selwood $3.20, Hayes $3.20, Pendlebury $5.00, Swan $6.00, Mitchell $8.00

Recommended bet:  Lenny Hayes is available at $11 (TAB Sportsbet) and I think he’s well over the odds at that quote. He polled 54 votes in three years between 2008 and 2010 and he meets two of my favourite ‘underrated’ criteria: (1)  likely to poll well early in the season (potentially four of the first five games), and (2) likely to poll well in some low profile matches (Gold Coast, Western Bulldogs twice, and Melbourne). Given that the Saints won 12 games, he could top 20 votes.

 

Round 10 leader

Ratings:  Watson 20, Pendlebury 18, Dangerfield 13, Thompson 12, Stanton 12, Ablett 11, Cotchin 11, Hayes 11

Rated odds:  Watson $3.50, Pendlebury $4.00, Dangerfield $10, Thompson $12, Stanton $12, Ablett $15, Cotchin $15, Hayes $15

Summary: Similar issues here to the overall winner market, although in this instance I think that the two clear favourites have correctly been identified.

Recommended bet: Nil

 

Most votes from each club

As indicated very early on in this column, occasionally a red-hot favourite in a club market can be defeated. As a consequence there are certain players whom while deserving of such marked favouritism, do not warrant any value at quotes of $1.10 and shorter and thus should not be wagered upon, nor should they be taken on. These favourites are listed below:

Essendon:  Jobe Watson (rating of 36.0, next best Essendon rating of 12.5)

Gold Coast:  Gary Ablett (rating of 29.0, next best Gold Coast rating of 6.5)

Richmond: Trent Cotchin (rating of 25.0, next best Richmond rating of 11.0)

In addition, there are a number of club markets with key players taken out being offered by bookmakers this year, so ratings will also be provided below for the majority of these scenarios.

However, there are also a number of team markets where there will be very few votes to go around, and thus betting would be a riskier proposition than is desirable. As a consequence, no ratings have been provided for the Gold Coast (without Ablett), Greater Western Sydney, Melbourne and Port Adelaide markets.

Adelaide

Ratings: Dangerfield 30.0 (rated odds of $1.50), Thompson 24.0 ($2.50)

Recommended bet: Dangerfield is not far away from being backable – if he drifts to $2.25 or better during the course of the next week, he could be backed for one unit.

Brisbane

Ratings: Black 18.0 ($2.00), Redden 12.0 ($3.80), Rockliff 8.0 ($7.00)

Carlton

Ratings:  Murphy 11.0 ($2.40), Judd 10.5 ($3.00), Carazzo 7.5 ($13), Gibbs 7.0 ($15), McLena 7.0 ($15)

Collingwood

Ratings:  Pendlebury 24.0 ($2.20), Dane Swan 19.5 ($2.80), Beams 11.5 ($6.00)

Recommended bet:  The $6.00 available for Pendlebury at Centrebet is a great price. He missed a few weeks midseason with injury, but may have polled in eight of the first nine games (hence he’s right in the ‘Round 10 leader’ markets) and will take some catching. Umpires love him too, polling a total of 45 votes in the past two years.

Essendon

Ratings without Watson: Stanton 12.5 ($1.15), Hurley 5.0 ($15)

Fremantle

Ratings:  Pavlich 13.0 ($3.30), Hill 11.5 ($3.80), Fyfe 11.5 ($3.80), Mundy 8.0 ($7.50)

Ratings without Pavlich: Hill 11.5 ($2.80), Fyfe 11.5 ($2.80), Mundy 8.0 ($5.50)

Geelong

Ratings:  Selwood 24.0 ($1.40), Hawkins 14.5 ($4.00)

Ratings without Selwood: Hawkins 14.5 ($1.80), Johnson 9.0 ($4.50), Kelly 8.0 ($6.00)

Hawthorn

Ratings:  Mitchell 19.0 ($2.20), Lewis 14.5 ($5.00), Sewell 13.0 ($6.50), Franklin 12.5 ($7.00)

Ratings without Mitchell: Lewis 14.5 ($3.20), Sewell 13.0 ($3.60), Franklin 12.5 ($3.60), Burgoyne 6.5 ($15), Rioli 6.0 ($17)

Recommended Bet: $21 at Centrebet about Jordan Lewis in the market without Mitchell is a fantastic price. While he had a quiet middle third of the season, he may well have polled on four or five occasions in the opening third of the season and was a contender for best on ground against Collingwood, Gold Coast and Sydney late in the year.

North Melbourne

Ratings:  Harvey 12.5 ($3.50), Swallow 11.5 ($4.00), Wells 11.0 ($5.00), Petrie 10.5 ($6.00), Todd Goldstein 7.0 ($12)

Richmond

Ratings without Cotchin:  Deledio 11.0 ($2.40), Tuck 9.5 ($3.50), Ivan Maric 8.0 ($5.00)

St Kilda

Ratings:  Hayes 24.0 ($1.15), Dal Santo 11.0 ($15)

Sydney

Ratings: Jack 13.5 ($2.80), Kennedy 13.0 ($3.00), Richards 9.0 ($7.00)

West Coast

Ratings:  Priddis 11.5 ($3.60), Kerr 11.0 ($4.40), Cox 9.5 ($5.50), Naitanui 9.0 ($7.00)

Western Bulldogs

Ratings: Griffen 9.5 ($2.20), Boyd 9.0 ($2.40)

Recommended Bet: The Bulldogs had a shocking year, with 13 of their 17 losses coming by 34 points or more. As a consequence, rounds three to eight (where four of their five wins for the year arrived) will all but decide who of Griffen and Boyd wins this poll. At $6.50 with Centrebet, I’m keen to back Griffen.

 

Head-to-head markets

There are a significant number of these markets around the place, and the majority of them seem pretty accurately priced to my eye. There were, however, two exceptions which I think are worth an investment:

Jordan Lewis (14.5 rating, odds of $1.30) to beat Harley Bennell (6.5 rating, odds of $4.40) at $1.80 with Sportsbet

Lenny Hayes (24.0 rating, odds of $1.33) to beat Brett Deledio (11.0 rating, odds of $4.00) at $1.95 with Sportsbet

 

 

Summary of recommended bets

0.5 units on Lenny Hayes to finish top-five overall at $11 (TAB Sportsbet)

1 unit on Scott Pendlebury to receive the most votes for Collingwood at $6.00 (Centrebet)

0.5 units on Jordan Lewis to receive the most votes for Hawthorn (excluding Sam Mitchell) at $21.00 (Centrebet)

1 unit on Ryan Griffen to receive the most votes for Western Bulldogs at $6.50 (Centrebet)

2 units on Jordan Lewis to poll more votes than Harley Bennell at $1.80 (Sportsbet)

2 units on Lenny Hayes to poll more votes than Brett Deledio at $1.95 (Sportsbet)

 

Thanks to Quinn Rooney/Getty Images AsiaPac for use of the photo