Certified Legendary Thread Race for the flag, in squiggly lines

Final Siren

Mr Squiggle
Aug 18, 2009
3,858
16,278
AFL Club
Richmond
Just on the "Home" factor, i previously did some analysis and the home advantage in finals was massively correlated to victory, is there any evidence in your model to suggest the home factor is more prevalent in finals?
Nope, I haven't looked much at finals specifically. This is just the same model used in home & away games. I'd be interested in seeing what you've got, though.

I know home sides do win a lot in finals, but of course you expect them to, since the higher-ranked teams get home finals. Does it happen even more than you should expect? That's probably hard to know for sure, since the sample size is fairly small (or else fairly old). But you could certainly crunch some numbers.
 

(Log in to remove this ad.)

Final Siren

Mr Squiggle
Aug 18, 2009
3,858
16,278
AFL Club
Richmond
Nope, I haven't looked much at finals specifically. This is just the same model used in home & away games. I'd be interested in seeing what you've got, though.

I know home sides do win a lot in finals, but of course you expect them to, since the higher-ranked teams get home finals. Does it happen even more than you should expect? That's probably hard to know for sure, since the sample size is fairly small (or else fairly old). But you could certainly crunch some numbers.
Ok I got interested enough to investigate. It's a little tricky to get information on which was really the nominal home side, since some data sources don't seem to think that's important and record the teams in any order. (Or, worse, winners first.) But I think I have it right.

And yes, home sides do win more than you'd expect! If you tip the home side in every final for the last 25 years, you get 160 right and 59 wrong, at an impressive accuracy of 73.1%. This is better than any of my algorithms could do by 1 tip.

Over the last 10 years, it's the same story: 64 right and 22 wrong at 74.4%.

Over the last 5 years (including 2014 so far): 31 right, 10 wrong at 75.6%.

What's also interesting is that you can ignore where they're playing. Quite a few teams have played "home" finals in the enemy's state over the years - West Coast in particular seem to have played a bunch of "home" finals in Victoria, and usually lost them. But on balance, it evens out: it doesn't matter whether you change your tip when a non-Vic team plays a "home" final in Melbourne.

Grand Finals, however, may not work so well. Tipping the home team over the last 10 years gets you 4 right and 6 wrong: worse than a coin toss and worse than all of my algorithms, except (sometimes) the one that does a coin toss. Over the last 25 years, it's 12 right and 13 wrong, also quite a lot worse than more sophisticated approaches. And again, it doesn't matter if you base this on nominal home team (which favours the team that finished higher on the ladder) or geographic home team (which favours Melbourne teams). There's not a big data set here, but for whatever reason, the home advantage that is so significant in regular finals seems to evaporate on the biggest stage.
 

swansfan51

Moderator
Jun 3, 2007
12,111
19,583
AFL Club
Sydney
Other Teams
The Exers
Great analysis Final Siren, a potential spanner in the works is that the home side in finals is usually the better performed side, so they'd be expected to win most games anyway?
 

El_Scorcho

Brownlow Medallist
Aug 21, 2007
25,979
62,497
AFL Club
Port Adelaide
Other Teams
Aston Villa, San Antonio Spurs
FinalSiren , how different would Port be looking now on the squiggle if we'd really put our foot down in the 2nd half and gotten to a 100 point victory in that final? Would it be enough to make us more competitive with Freo this week?
 

Final Siren

Mr Squiggle
Aug 18, 2009
3,858
16,278
AFL Club
Richmond
Great analysis Final Siren, a potential spanner in the works is that the home side in finals is usually the better performed side, so they'd be expected to win most games anyway?
Exactly right, but home sides seem to have won finals even more than that. For example, the squiggle's default algorithm of ISTATE-91:12, which considers form as well as home state advantage, is 151-68 (69.0%) in finals since 1988, whereas HOMER is 160-59 (73.1%).

I have one algorithm that justs pips HOMER in finals tipping over the last five years (WEARY, which makes adjustments for days' break between games: 32-9 at 78.1%), but it's slightly worse over a longer time period. Every other algorithm is unable to improve on the simple system of tipping the home team.
 

Final Siren

Mr Squiggle
Aug 18, 2009
3,858
16,278
AFL Club
Richmond
FinalSiren , how different would Port be looking now on the squiggle if we'd really put our foot down in the 2nd half and gotten to a 100 point victory in that final? Would it be enough to make us more competitive with Freo this week?
Ah, that what-ifs... bear in mind that plenty of teams in the past have also no doubt slipped into cruise control once the game was in the bag. So there is probably a real difference between a team that wins a final by 100 pts and a team that could have but didn't.

Anyway: more competitive, sure, but the squiggle would still be tipping Fremantle by 10. For it to tip Port you needed another 160 points, i.e. a 290-75 victory. Or alternately kicking the same score (132) but keeping Richmond below 18 pts.
 

Billy ray

Brownlow Medallist
Nov 10, 2013
12,651
18,820
The Valley near the Alley
AFL Club
Gold Coast
Grand Finals, however, may not work so well. Tipping the home team over the last 10 years gets you 4 right and 6 wrong: worse than a coin toss and worse than all of my algorithms, except (sometimes) the one that does a coin toss. Over the last 25 years, it's 12 right and 13 wrong, also quite a lot worse than more sophisticated approaches. And again, it doesn't matter if you base this on nominal home team (which favours the team that finished higher on the ladder) or geographic home team (which favours Melbourne teams). There's not a big data set here, but for whatever reason, the home advantage that is so significant in regular finals seems to evaporate on the biggest stage.
I wonder how the bookies have gone with grand finals over the years, has the favourite (where the $'s are) won more often than not?
 

Billy ray

Brownlow Medallist
Nov 10, 2013
12,651
18,820
The Valley near the Alley
AFL Club
Gold Coast
Need the wider big footy community to see this thread. Easily the best thing on here
agree
needs to be a sticky, on front page of main board.
and new season with new thread "Race for the flag, in squiggly lines - 2015" with a link to this thread which can be renamed "Race for the flag, in squiggly lines - 2014" and split if need be for the year earlier.
also, deserves thread of the year, every year, forever and ever.
 

Belnakor

Brownlow Medallist
Apr 10, 2005
19,512
10,648
Perth
AFL Club
Fremantle
Nope, I haven't looked much at finals specifically. This is just the same model used in home & away games. I'd be interested in seeing what you've got, though.

I know home sides do win a lot in finals, but of course you expect them to, since the higher-ranked teams get home finals. Does it happen even more than you should expect? That's probably hard to know for sure, since the sample size is fairly small (or else fairly old). But you could certainly crunch some numbers.
it was just a simple regression model i'd have to dig up, basically the biggest factors were ladder position + home ground, and i think confidence went up significantly if it was also a final.
 

(Log in to remove this ad.)

Final Siren

Mr Squiggle
Aug 18, 2009
3,858
16,278
AFL Club
Richmond
Hey, do most tipping comps consider a draw to be a right tip or wrong? I have the squiggle marking it as a wrong tip but not sure if that's standard.
 

Belnakor

Brownlow Medallist
Apr 10, 2005
19,512
10,648
Perth
AFL Club
Fremantle
Exactly right, but home sides seem to have won finals even more than that. For example, the squiggle's default algorithm of ISTATE-91:12, which considers form as well as home state advantage, is 151-68 (69.0%) in finals since 1988, whereas HOMER is 160-59 (73.1%).

I have one algorithm that justs pips HOMER in finals tipping over the last five years (WEARY, which makes adjustments for days' break between games: 32-9 at 78.1%), but it's slightly worse over a longer time period. Every other algorithm is unable to improve on the simple system of tipping the home team.
it would probably be more interesting if we still had the previous finals system, where the team that lost the first week played away regardless of their previous ranking iirc.
 

Final Siren

Mr Squiggle
Aug 18, 2009
3,858
16,278
AFL Club
Richmond
Tipping the GF

The squiggle algorithm ISTATE-91:12 is not very good at tipping Grand Finals. It's 15-10 (60%) over the last 25 years (ignoring the drawn Grand Final).

This is because it awards a 12-pt home ground advantage to Melbourne-based teams at the MCG, yet on Grand Final Day this doesn't seem to eventuate. If anything, it's the non-Victorian team that usually overperforms. Why this might be, I have no idea. But it's what's been happening.

Picking the home team, as discussed above, doesn't work either, despite being an excellent system for regular finals.

The best algorithm I have for Grand Finals is OFFDEF-75. This works like ISTATE-91:12 but with two little differences:
  1. It doesn't award any home ground advantage
  2. It is a lot more sensitive to recent form
While ISTATE-91:12 weights the most recent game at 9%, with 91% coming from previous games, OFFDEF-75 weights the most recent game at 25%, with older games being a relatively modest 75%.

It's a small tweak and a simple system, but OFFDEF-75 has tipped Grand Finals at 20-5 (80%) over the last 25 years, and 9-1 (90%) over the last 10 years. (The one it missed is Hawthorn 2008, which all of my stats swear shouldn't have happened.)

OFFDEF-75 is only good for Grand Finals. It doesn't do that well during the home & away season, because it doesn't account for interstate home advantage, and it gets too carried away with individual results. It's not great in regular finals, either, since it doesn't care who is the home team, which seems to matter a lot. But in Grand Finals it's a superstar!

I can't tell you what OFFDEF-75's tip is for the Grand Final yet because it needs to see teams' form in the semis and prelims. But I thought it was interesting to see a statistical basis for the theory that more than anything it's finals form that wins premierships.
 

Sydney Bloods

Brownlow Medallist
Apr 7, 2012
18,188
13,935
Sydney
AFL Club
Sydney
Other Teams
Coney Island, GWS, The Exers!
Finals Week 1, 2014


Port Adelaide skipped ahead to a very nice position this week on the back of their comprehensive demolition of Richmond. In fact, not only have the Power fully arrested their decline, but they're in a better position now than they were at any point this season, even better than when they were a game clear at the top of the ladder. They're so close to a bunch of premiership cups, they could lick them, if they wanted to, and were feeling a bit weird.

The squiggle doesn't think Port have enough Power to beat Freo in Perth, though, as I'll post later. But it's been a pretty amazing journey for the Power this season.

The Hawks beat Geelong by a bit more than expected, but otherwise there wasn't much squiggle movement from the other finals games.

The interactive squiggle stopped updating over the weekend; sorry about that. It's working again now!
Good write up as always but you missed a great (ok terrible) pun. see edit.
 

Final Siren

Mr Squiggle
Aug 18, 2009
3,858
16,278
AFL Club
Richmond
so
phase 1 - ISTATE-91:12 during the year
phase 2 - HOMER in first three weeks of finals
phase 3 - OFFDEF-75 for Grand Final

or as the underpants gnomes would say:
phase 1 - get underpants
phase 2 - ?
phase 3 - profit
If you'd done this over the last couple of decades, you would have been a solid tipster during the H&A season, a brilliant tipper during finals, and a genius at Grand Finals.

What we want to know, though, is whether it's sustainable! I think the answer is yes. The part that looks suspicious is HOMER, because there doesn't seem to be much logic in why the home team should win regardless of all other factors - even form! And when building models, it's easy to fall into the trap of mistaking a random pattern in the data for meaning, and creating a model based on a coincidence.

For example, given perfect data, I'm sure I could perfectly model the 2014 season based on the underwear of fans who attended the game - because out of tens of thousands of seats, there would be one particular seat where every time a fan sat there wearing boxer shorts, the home team won, but every time the person in that seat was wearing tidy whities, the home team lost. But this wouldn't be any good at predictions. It would just be a neat coincidence.

So is HOMER a coincidence or meaningful? The more I investigate, the more I think it's meaningful.

The first piece of evidence is that HOMER works much better in finals. During the home & away season, HOMER tips at around 58% accuracy; during finals, it's 75%. There are two major differences between the H&A season and finals: the team that finished higher on the ladder is always the home team, and the competing teams are evenly matched. You never get 1st vs 18th in a final. And since 2000, you never get 1st vs 8th, either. It's all matches between teams that are very close on the ladder: 1st vs 4th, 2nd vs 3rd, 5th vs 8th, 6th vs 7th.

And sure enough, if you look at 1994-1999, which was the old McIntyre Final Eight, where week one was 1st vs 8th, 2nd vs 7th, etc, HOMER isn't quite as impressive. It's still good - in fact you can still get 75% accuracy out of it, if you make a few tweaks like awarding home advantage to the geographic home team rather than the nominal home team, for those games where West Coast played "home" finals in Melbourne. But many other algorithms can also get that kind of accuracy, or better. It's only once the finals system switched in 2000 that HOMER started outperforming everything else.

The second thing to note is that HOMER is unswayed by Week 1 Finals form. We know this has happened a lot since 2000: two top-4 teams lose their Qualifying Finals and slink into a playoff against Elimination Finals winners, and the EF winners look like superstars and everyone talks them up... then the QF losers come out and beat them. The history is pretty compelling: even if Port and North upset the odds this weekend, there's a massive weighting in favour of QF losers - especially since these are games between teams that finished as close as 4th and 5th! All things being equal, you'd expect 5th to beat 4th an awful lot more than has actually occurred. That suggests there's a hidden factor (or two) at work here; whether it's confidence or motivation or pacing or something else. HOMER doesn't know what it is, but it thinks the team that loses the QF has it.

And one more thing. There seems to be some kind of spooky "19th man" voodoo around being the home team. I've only figured this out recently, because I used to think home ground advantage was all about geography. And earlier this year, I made what turned out to be a fairly grievous mistake by "fixing" the squiggle to award home ground advantage based on geography, rather than nominal home team status, which seemed sensible and easier to explain. But it does worse at tipping most years, and this turned out to be one of them, with the "improved" version falling behind by 3 tips. So the original, un-improved ISTATE-91:12 currently has 146 tips and the "improved" version, using geographic home advantage, only 143 (counting the draw as a right tip). Since 1994, the difference between the two is a whopping 23 tips in the original's favour. I should not have made that fix. Being the home team is a big deal.

So, on balance, blindly tipping the home team in a final isn't nearly as dumb as it sounds. In a game between two closely matched opponents, you're preferring the one that finished higher on the ladder, you're backing QF losers to bounce back next week, and you're believing in home ground advantage.
 
Last edited:

Top Bottom