Friday, 8 July 2016

COMPARATIVE PREDICTIVITY

Every week since round 7, the SWARMS page (check the tab at top of the blog) has been updated with the predictions of where your team and others will wind up at the end of the regular league season.

The current ladder is an interesting beast as there are three distinct groups of teams.

The remaining fixture could amend the above split, but by how much is yet to be determined. Models such as the FMI predictor and others can estimate where the changes will occur, if at all.


As it stands now, the FMI calculations set GWS as the minor premier followed by Adelaide, the Western Bulldogs and Sydney to round out the top 4.
Missing the eight are Port Adelaide and Melbourne in 8th and 9th. All the detail is in the 'WIN' tab of the SWARMS page, as opposed to the simple ladder prediction table added in here.
F M I
1GWS17.9
2Adelaide17.6
3W Bulldogs/FFC17.5
4Sydney/SMFC16.8
5Hawthorn16.5
6Geelong16.0
7West Coast14.4
8North Melbourne14.3
9Port Adelaide11.1
10Melbourne10.2
11St Kilda9.9
12Carlton8.6
13Collingwood7.6
14Richmond7.4
15Gold Coast4.9
16Fremantle4.1
17Brisbane Lions1.8
18Essendon1.5


There have been questions asked about the derivation of these outcomes, The above is based on 'as current' data plus the expectations on performance plus adjustments to teams ratings post each individual simulated game. All of the future factors alter during simulations as the fixtures play out, and repeated simulations are run.

If the above methodology is true and correct has already been debated, and depends on your viewpoint, but the system employed above will remain for the remainder of the season.

What is interesting to do is to not only consider a single simulation outcome as 'the data', but instead look for more data. And this is easy as there are three other (and perhaps better known) systems that produce similar simulations.

Firstly, the Matter Of Stats blog has recently joined the 2016 season prediction stakes and also upped the ante by running 100,000 simulations for greater accuracy [1].

Also to compare are the highly popular (in BigFooty circles) Squiggle predictions [2], and the FootyForecaster end of season table [3]. Its not clear on how many iterations the data is created from in both these, but they do change as the weeks roll on.

     [1]   Additional win expectation data used below not listed on that blog, but tweeted today.
     [2]   Click the 'Prediction' marker at the top of the page to open a new table.
     [3]   Click on 'Season Forecast' on the left of screen.



To make everyone's life easy, the table below aggregates all the predictions together for ease of comparison.

Immediately leaping off the page are some disagreements, such as the number of expected wins and positions of the top 6 teams, as well as some obvious positional agreements, such as West Coast and North Melbourne to fill 7th and 8th, as well as the same four teams in 15th to 18th.

CURRENT AS AT (POST) ROUND 15
F M I MATTER OF STATSFOOTYFORECASTERSQUIGGLE
1GWS17.91Geelong16.71GWS161Geelong16
2Adelaide17.62Adelaide16.42Adelaide162Adelaide16
3W Bulldogs/FFC17.53GWS16.03Hawthorn163GWS16
4Sydney/SMFC16.84Hawthorn15.94Geelong154W Bulldogs/FFC16
5Hawthorn16.55W Bulldogs/FFC15.75Sydney/SMFC155Hawthorn16
6Geelong16.06Sydney/SMFC15.26W Bulldogs/FFC156Sydney/SMFC15
7West Coast14.47West Coast13.77West Coast147North Melbourne15
8North Melbourne14.38North Melbourne13.78North Melbourne148West Coast14
9Port Adelaide11.19Port Adelaide11.29Port Adelaide119Port Adelaide12
10Melbourne10.210St Kilda10.110Melbourne1010St Kilda10
11St Kilda9.911Collingwood9.111St Kilda1011Melbourne9
12Carlton8.612Melbourne9.012Collingwood912Collingwood9
13Collingwood7.613Carlton8.813Richmond913Richmond9
14Richmond7.414Richmond8.614Carlton914Carlton8
15Gold Coast4.915Gold Coast6.215Gold Coast715Gold Coast7
16Fremantle4.116Fremantle5.516Fremantle616Fremantle5
17Brisbane Lions1.817Essendon3.217Brisbane Lions317Brisbane Lions3
18Essendon1.518Brisbane Lions3.118Essendon318Essendon2


There is distinct commonality across all four particularly when you look at the earlier tweet about groups of teams in the current ladder.

All four predictions remain with three groups after 23 rounds - a top 8 separated from a trailing 6 with a distant last 4.

All four also have roughly 14 games as the standard for making the finals, and also a gap of 2 to 3 games between 8th and 9th.

All four also have a second gap of at least 2 games between 14th and 15th.


Which is the most accurate is open for speculation. As an exercise in bench-marking the four modelling systems we can see that all four have enough in common to consider the eventual R23 ladder to be something like what is above.

No comments:

Post a Comment