Monday, 19 December 2016

2016: THE WASH-UP

With the year winding down, its time to review what was. And give time restraints, it follows last years method - short and sharp.
The pressures of employment and family have made the toughest of foes in the fight for time and attention to analysis and blogging.

In terms of the FMI targets, the new system was employed and is used as the key measure (as it should be) but also the old system results will also be listed to show progress or regress.

Further below are charts of each clubs' change in rank week to week, as well as measures on wins and ladder finishes against the FMI preseason prediction.


But first, a look at a few other season goals set earlier this year.



TIPPING RETURN:

"...another correct tipping return over 145 (70%) should be expected, and if 150 (72%) is breached, even better."

Outcome:
Season 2016 fell two short of target, with 143 the final result, and 69.1 in percentage terms.
This remains a decent outcome given the the horror tipping run through rounds 6 to 8. Over those rounds, the tipping average fell from 74% down to 66%.

But the ship was righted with rounds 9 and 10 delivering good returns.

After that recovery all the remaining rounds were tracking above 70%, and the eventual sub-70% result can be sheeted home to the eventual Premiers. The FMI system did not tip them at all in the finals, and those misses weighed heavily.

Old System - 143. (no change)


MONASH UNI:

"Another score over 1850 is the KPI, and nudging past 1900 will be a great return."

Outcome:
Season score of 1796 is well under the KPI, but still strong enough to garner a 10th place finish vs published data.

Also encouraging was the increase again in competition, with other data-based tipsters scoring well.
Well played @TJMillar1, @MatterOfStats, @DownIsTheNewUp, @RankingSW and @FiguringFooty for all finishing in the top 10.


"Target 2016 is to again be top 10 minimum and top 5 would be ideal."

Outcome:
Tenth, so achieved... in the second last week of the season.

Old System:
1772 points (-24 pts vs new) and a 15th place finish (-5).


PERFECT ROUNDS AND MARGIN WINDOWS:

"...target is to achieve 2 perfect rounds."

Outcome:
Missed again. This year there were four rounds with 8 correct. SO close, but not close enough.

Old System:
No perfect rounds (no change).


There was no goal set for exact margins tipped, but 3 games were correctly tipped to the exact margin. The broader goal of being in the 0-12 point range was set this year though, as below.
"...a target of 29% or more of games tipped in the 0-12 point should be achievable."

Outcome:
This is the first time to have set this target, and it may be positioned a little too high. The season result is 24.6%, which is significantly under. To be reviewed.

Old System:
Only one exact was achieved (-2) and the 0-12 margin/game range was 21.2% (-3.4%).


MEAN ABSOLUTE ERROR:

"The MAE is a measure of accuracy of predicted margin to actual margin. As was described to me by a regular reader and twitter correspondent (@SgtButane), a value under 30 would be a very good return."

Outcome:
As before, when there are a run of bad weeks, maintaining a tight MAE is never easy. This season though, a respectable MAE of 30.5 was achieved. Better than last year and an improvement over the old system.

Old System:
31.4, so a significant improvement (+0.9)


SOME FINAL NOTES:

Thank you again readers and followers. This blog thing is literally hanging out there in cyberspace and to have hits, comments, interactions and replies make it worthwhile.
Another increase in annual readership is also something to be happy about, even if there is less content and less quality.

Thanks again to contributors to the FMI series if blogs and twitter accounts. Both those with new ideas and proposals to improve, and those along for a bit of a laugh. All these things make the workaday week just a bit better. The @FMI_CL and @FMI_TL are well received (so you say), and I cant believe how the @FridayNotFooty following has developed.

A final note of thanks to those I have met IRL this year - some before and some new folk as well. Looking forward to doing it all again next year, smarter, and even more accurate.



ADELAIDE 

5th, 16 wins.

Ladder:      Expected: 8th        (Range: 4 - 10)
Wins:         Expected: 12 - 17
Summary:  Played to the top range of their expected level, good season given resources lost. Set for 2017.




PORT ADELAIDE 

10th, 10 wins.

Ladder:      Expected: 7th        (Range: 3 - 9)
Wins:         Expected: 12 - 17
Summary:  Disappointing start set the tone for the season. A wasted effort and little progress.




FREMANTLE

16th, 4 wins.

Ladder:      Expected: 9th        (Range: 5 - 10)
Wins:         Expected: 11 - 16
Summary:  Expected not to make the 8, but went down in a flaming heap, bordering on tanking. Terrible.




WEST COAST 

6th, 16 wins.

Ladder:      Expected: 2nd        (Range: 1 - 3)
Wins:         Expected: 16 - 20
Summary:  6th place decent given the closeness of the top 7, but overall a nett negative year.




BRISBANE LIONS 

17th, 3 wins.

Ladder:      Expected: 14th        (Range: 13 - 16)
Wins:         Expected: 3 - 7
Summary:  Wasted season for the northern-most club. Ship is being righted with new staff and expectatons.




GOLD COAST 

15th, 6 wins.

Ladder:      Expected: 15th        (Range: 13 - 17)
Wins:         Expected: 2 - 6
Summary:  Ebbing and flowing and not really going anywhere. Finished as expected and should be better.




GWS 

4th, 16 wins.

Ladder:      Expected: 12th        (Range: 11 - 13)
Wins:         Expected: 7 - 11
Summary:  Early season set the tone and lined them up for a 1st finals tilt. Top season and look out 2017.




SYDNEY/SMFC 

1st, 17 wins.

Ladder:      Expected: 3rd        (Range: 2 - 7)
Wins:         Expected: 14 - 18
Summary:  Finished around expectations and faltered in the big one. Expecting similar next season.




W BULLDOGS/FFC 

7th, 15 wins.

Ladder:      Expected: 5th        (Range: 3 - 8)
Wins:         Expected: 13 - 17
Summary:  Finished the H&A bang on to expectation, but look at those last 4 weeks - turned it ON.




ST KILDA 

9th, 12 wins.

Ladder:      Expected: 13th        (Range: 12 - 16)
Wins:         Expected: 4 - 8
Summary:  Great effort, the top 8 was lock, so 9th is brilliant. Under the radar all year, but not any more.




RICHMOND 

13th, 8 wins.

Ladder:      Expected: 4th        (Range: 3 - 8)
Wins:         Expected: 13 - 17
Summary:  Huge disappointment, second only to Fremantle, and only just. Time for a re-build.




NORTH MELBOURNE 

8th, 12 wins.

Ladder:      Expected: 6th        (Range: 3 - 8)
Wins:         Expected: 12 - 17
Summary:  Season went to predictions - won early against the minnows, & struggled later when it got hard.




MELBOURNE 

11th, 10 wins.

Ladder:      Expected: 16th        (Range: 13 - 17)
Wins:         Expected: 3 - 6
Summary:  Last of the P.Roos era and their best so far. Should be set for a better 2017 with lots of talent.




HAWTHORN 

3rd, 17 wins.

Ladder:      Expected: 1st        (Range: 1 - 2)
Wins:         Expected: 18 - 21
Summary:  The best has now come back to the field. 3rd worst season losing 200pts, ranked 6th. Era ends?




GEELONG 

2nd, 17 wins.

Ladder:      Expected: 10th        (Range: 8 - 11)
Wins:         Expected: 10 - 14
Summary:  That early season drive (from Dangerfield) pushed them back into contention before plateauing.




ESSENDON 

18th, 3 wins.

Ladder:      Expected: 18th        (Range: 16 - 18)
Wins:         Expected: 1 - 4
Summary:  A year of going backwards in the rankings, but games into kids will send them back up in 2017.




COLLINGWOOD 

12th, 9 wins.

Ladder:      Expected: 11th        (Range: 9 - 12)
Wins:         Expected: 9 - 13
Summary:  Another year of treading water at Olympic Park. One day they will rise, but not without change.




CARLTON 

14th, 7 wins.

Ladder:      Expected: 17th        (Range: 15 - 18)
Wins:         Expected: 2 - 5
Summary:  The new coach and direction has worked wonders. Getting further than bottom 6 will be harder.



Friday, 25 November 2016

DRAFT HISTORY RESOURCE (2016)

Once again, your handy draft day reference of all past draftees, all in a simple Google Sheets doc that you can filter etc

Click the link and there it is.

SO go ahead and see if your draftee is the next
Hamish McIntosh, or
Luke Molan, or
Kayne Pettifer, or
Caydn Beetham, or even
Mark McVeigh or
Chad Cornes.

Tuesday, 15 November 2016

A BETTER TOMORROW II

Regular readers of this blog (yep, just you) will know clearly that there is a better way to balance the fixture than that currently done by the AFL.

It is as proposed over here in this ancient manifest, where the idea went a little too far and for a few thousand words too long as well.

To boil all that down to its main point - the best way to balance the fixture while maintaining 22 rounds is to split teams into three groups that play set teams home and away twice.
But not the AFL's split of top 6 playing each other, mid 6 playing each other etc.

Split them so that each group is composed of two from the top 6, two from the middle 6 and two from the bottom 6.
Those groups of 6 play each other home and away, and each team in that group plays the other teams outside the group once, for a total of 22 games.
Adjustments are made to the location of the 12 non-group games to balance travel requirements for the Victorian clubs.
This makes for a balanced draw. No need to go the full Magilla and have conferences. Just make the splits fair.


Applying the proposed scheme from the manifest, we take the 2016 ladder at left, and convert it such that the groups where teams are paired are at right.
2017 LADDERH&A GROUP (A)H&A GROUP (B)H&A GROUP (C)
1W Bulldogs/FFC1W Bulldogs/FFC2Sydney/SMFC3Geelong
2Sydney/SMFC6Adelaide5Hawthorn4GWS
3Geelong7West Coast8North Melbourne9St Kilda
4GWS12Collingwood11Melbourne10Port Adelaide
5Hawthorn13Richmond14Carlton15Gold Coast
6Adelaide18Essendon17Brisbane Lions16Fremantle
7West Coast
8North Melbourne
9St Kilda
10Port Adelaide
11Melbourne
12Collingwood
13Richmond
14Carlton
15Gold Coast
16Fremantle
17Brisbane Lions
18Essendon


So as mentioned, group C has four journeys for Geelong and St Kilda, so they would do less travel in the cross-group games.

The above, with some work to achieve a fixture where each team has 11 home and 11 away games, will produce a balanced result.
Measuring and proving that balance and fairness, is the next step.



Mutual Aggregation Society

Before reviewing and proving that a better balanced system can be achieved, its worth noting that the doyen of footy stats, MoS, has run up a blog post comparing the five fixture analyses offered by Rohan Connolly, Trow Wheatley, the team at HPN, FMI and of course their own, all residing at this link.

Courtesy: MatterOfStats.com
By combining the five into a single ranking system, MatterOfStats produces an output listing GWS at the top of the tree with Hawthorn, and North Melbourne and Richmond at its base.


The MoS summary pre-empted the same post that was planned for here, but it still left open another comparision.

As is well documented now, the AFL plan is to have the top teams playing against each other, with the middle 6 and bottom 6 also similarly grouped. So rather than just sorting by the average 'ranking position' score across the five analyses, the second approach is to look at the finishing slots differently.

And the thinking is this: Where does any team find themselves grouped by these analyses?

   -  Do they have a 'top 6' toughest schedule assessment?  - If so allocate 1 point per assessment in that group.
   -  Do they finish in the middle six? Two points per assessment in that group.
   -  Finishing in the bottom 6 gives you three points each time you land in that group.
   -  Then tally those numbers, and keep the above MoS assessment to sort out teams with the same finishing group tally scores.

So under those rules, you get a variation on the above, but weighted to the AFL policy of handicapping the race.
FIXTURE EVALUATIONS - GROUPED
RNKTEAMGRP TALLYAVG RANK
1GWS51.8
2Hawthorn52.2
3W Bulldogs/FFC64.8
4St Kilda65.0
5Sydney/SMFC75.4
6West Coast77.6
7Geelong86.2
8Collingwood88.0
9Fremantle97.0
10Melbourne1110.0
11Adelaide1213.0
12Carlton1313.0
13Brisbane Lions1314.0
14North Melbourne1314.6
15Essendon1414.2
16
Gold Coast
14
14.4
Port Adelaide
18Richmond1515.4

Again, there is some commonality between the above and the MoS assessment, but there are a few minor changes to account for some of the key discrepancies between the five methods.



Theoretical Balance - H&A

There once was a time, with 12 teams only where playing each other team once home and away brought a season of 22 weeks in duration.
Everyone played everyone else once. Eleven at home and eleven away. Synergy, and crystals aligning and planets in sync and other such stuff.
Maybe.

Now, with an 18 team season, its all but impossible to get that 'pure' balance into reality. The season would stretch on too long. Players would be pushed too hard. The AFLPA wouldn't allow it (well... not without a shed load of $'s). Other sports would rail against it.

Eighteen teams playing home and away would be a 34 week season (2 x 17), and if you add in two mid-season byes, plus a pre-finals bye, and then 4 weeks of finals - well that gets you 41 weeks of footy.

Forty. One. Weeks.


Which means it cant exist in reality. But an 18 team competition and 34 home and away matches can exist in theory.

Last week, this blog posted an assessment on the current 2017 fixture as per that link. In that post it suggested the AFL achieved its goal of the top teams having a harder schedule and the lower teams an easier fixture.

The next step therefore is to apply the same methodology to the theoretical 34 week season (including the 'sold' home games to regional cities and Shanghai as they are now) and determine the outcome.
The below table shows that output.
FIXTURE BALANCE - 34 ROUND SCENARIO
RNKTEAMLAD POSFMI RANK PTSSUM OPP'N PTSHOME / AWAY BONUSTOUGH DRAW INDEX
1Sydney/SMFC214283685703.74
2West Coast713253706203.31
3GWS4137836957-1003.30
4W Bulldogs/FFC11272371682003.18
5Adelaide6140136910-2003.03
6Geelong313163708002.94
7North Melbourne81175373633002.61
8Hawthorn5130537104-2002.22
9Port Adelaide10118237348-2001.14
10Collingwood1210963752100.43
11Melbourne111023376682000.01
12St Kilda9101737678100-0.41
13Richmond13972377690-1.30
14Carlton14886379400-2.24
15Fremantle16902379090-2.33
16Essendon18693383270-2.94
17Gold Coast1585638000-100-3.12
18Brisbane Lions17630384540-3.14


In the above, we begin to see a breakdown of the AFL model of top-teams having a harder calendar of games, with West Coast bobbing up into second. But lest not forget that West Coast did finish home and away in 6th on percentage and equal on points with 4th placed GWS.

What is more interesting in the above table is the spread of the FMI Tough Draw Index. In the current fixture released by the AFL, at the top of the table was GWS with 4.98 points, and at the foot with the easiest schedule is Brisbane on -3.67 points.
Those extremities have clearly come in closer to the mean,

So recalling that posts' comment on 'spread':
The spread has come down to 8.65 for next year (calculated as the absolute difference between #1 ranked GWS on 4.98 and #18 ranked Brisbane at -3.67).
The previous years were:
   2016:   8.90
   2015:   8.98
   2014:   9.15

So heading to the theoretical 'everyone plays everyone else twice' scheme as applied, that spread is now down to 6.88 from top to bottom.
So yes, it improves the bottom line of even-ing out the clubs schedules for the year.

It's just unfortunate that is 34 home and away season can't wont couldn't happen.



FMI - Manifest Balance

So the other way to look at balancing the fixture is of course via the FMI method listed above.
Split them so that each group is composed of two from the top 6, two from the middle 6 and two from the bottom 6. Those groups of 6 play each other home and away, and each team in that group plays the other teams outside the group once for a total of 22 games. Adjustments are made to the location of the 12 non-group games to balance travel requirements for the Victorian clubs.
Once again, applying the above group splits to the season and including the 'sold' home games to the regions, the same 'tough draw index' calculations reveal the below results.

FIXTURE BALANCE - FMI MANIFEST
RNKTEAMLAD POSFMI RANK PTSSUM OPP'N PTSHOME / AWAY BONUSTOUGH DRAW INDEX
1W Bulldogs/FFC11272240712002.65
2West Coast7132523965-1002.05
3Adelaide6140123813-1001.98
4Collingwood121096244242001.71
5Sydney/SMFC21428234461001.63
6Geelong313162387601.02
7North Melbourne81175239521000.68
8GWS4137823753-2000.64
9Richmond139722467200.34
10Port Adelaide10118224144-2000.26
11St Kilda91017244731000.10
12Hawthorn5130523693-400-0.08
13Essendon1869325230200-0.21
14Melbourne11102324257200-0.21
15Fremantle1690224704100-0.59
16Brisbane Lions1763025043100-1.38
17Carlton14886245290-1.84
18Gold Coast1585624795-300-2.02


Once again, you could be distracted looking at the raw ranking down the left side and start wailing about 'Oh no! Collingwood get the 4th hardest schedule".
But you would be wrong to do that.

Because the right-hand column is the one to see. All the shading on that column has been kept consistent, so you can clearly see there is less strength at either end of the color spectrum, meaning the spread of difficulty has evened out.
Most clubs under this system improve their schedule, with a few getting a slightly harder outcome, but not by any significant standard.

And again looking at the spread across the whole 18 teams, it has been reduced considerably.
The current FMI Tough Draw Index for the 2017 season under the AFL 'handicap' system is 8.65 points.
Working a theoretical full home and away season, this spread reduces to 6.88 points.
Under the FMI manifest, it is brought right down to 4.67 points.


Some might say that this 'single point' of theory does not validate the manifest as being a balancer of the schedule.
To add another data point, it is worth doing the same calculations on the 2016 season past, including re-fixturing the season based on the 2015 ladder split into the three groups as proposed, and allowing for 'sold' home games.
The re-fixtured season would have the below splits of teams that play each other home and away.

H&A GROUP (A)H&A GROUP (B)H&A GROUP (C)
1Hawthorn2West Coast3Fremantle
6Adelaide5Sydney/SMFC4North Melbourne
7Richmond8W Bulldogs/FFC9Port Adelaide
12Collingwood11GWS10Geelong
13Melbourne14St Kilda15Essendon
18Carlton17Brisbane Lions16Gold Coast


And of course travel balance measures are needed to be made such that Melbourne teams are 'out of state' a reasonably consistent number of weeks.

The 2016 schedule was determined to have a spread of 8.90 points, as per last years post.
When the FMI schedule proposal is applied to 2016,  the spread is reduced to 3.09 points.

So that is now two data points to confirm the best way to return to a balanced draw is to run the splits of teams not in the current AFL handicap, but through the full run of the previous years table.
Split them so that each group is composed of two from the top 6, two from the middle 6 and two from the bottom 6.

Simple. Transparent. Fair.
... and as it appears - balanced.


Back to the denizens of League HQ to make a move.