Strength of Schedule in the 2016 AFL Fixture

It’s AFL fixture time again, and that means plenty of attempts to evaluate the unbalanced AFL fixture and work out who’s got the soft ride and who’s being screwed by a tough draw. This post is our analysis based on a number of Strength of Schedule calculations.

If you want the short version before we dive in:

  • North and Adelaide have it the toughest
  • Carlton Essendon and Geelong have it the easiest
  • The fixture successfully gives the bottom 6 teams the strongest weighting on the easy side of a fair schedule, but GWS have an anomalously tough run for a team that finished 12th
  • Comparing 2015 to 2016, Port Adelaide, Geelong and Essendon should expect a bump of at least 1 expected win next year from just the fixture
  • Conversely, GWS should expect -0.9 wins and Hawthorn and Adelaide -0.8 wins compared to 2015 based on the fixture.

And the caveats:

  • These are all based on strength of opponent alone
  • Strength of Schedule makes a difference of about half an expected win between the hardest and the average fixture, and less than 1 win between the hardest and easiest draws
  • These assessments will turn out to be interestingly wrong – last year for example Hawthorn projected to have the 3rd hardest draw and it turned out to be the 3rd easiest

What is Strength of Schedule?

The Strength of Schedule measure is often used in US sport, specifically the NFL and Collegiate Sports, to equate records where the schedules are not perfectly even. Take for example NCAA Division I College Football, where each team plays around 12 games, but have 128 teams in the division. In order to best work out who was actually the best in order to play Bowl games (or the new NCAA College Football Playoff), the selection committee leans on the strength of schedule as a tiebreaker for teams on the same, or similar, records. In the world of College Football, this can be worth millions of dollars.

The most basic way of determining SOS is to average out the winning rates of all teams that each team faces. This is the method that the NFL and the NCAA use in their formulations. Some of the more advanced statistical methods account for other factors, such as Home Ground Advantage and extra weighting towards stronger teams.

We’re sticking to the most basic formula, but will also use a couple of different inputs as well as straight win-loss records. All data is based on results from the 2015 AFL Season. We’ll first assess the fixture as a tool of competitive equalisation, and then go into discussing some different measures of fixture strength.

Does the fixture achieve competitive equalisation?

Strength of Schedule on Win-Loss is a straightforward measure of the cumulative winning percent of each team’s opposition in 2016. A team with a 50% winning record and who played every opponent the same number of times would have a strength of schedule of 0.500. Strength of Schedule tells us how far each club’s draw deviates from that ideal.

Of course, since teams don’t play themselves that’s an inherent influence on draw difficulty. If every team played each other once in a 17-week season, Fremantle’s opposition would have had a strength of 8.2 wins in 2015, while Carlton’s would have 8.8 wins. That translates to a fair draw for Fremantle being 0.484 and for Carlton being 0.519.

Here’s a chart looking at strength of schedule for each side based on the win-loss record of their opponents.

Strength of Schedule assessment of fixture weighting

The AFL’s fixturing goal is to give the sides who finished top 6 last year (post-finals) the hardest set of matchups, and the bottom 6 the easiest. We can compare the impact of their double-up matches by looking at their actual Strength of Schedule compared with what a “fair” one would be.

Does the weighted AFL fixture “equalise” things competitively? No, it overcompensates compared to a balanced draw, which is the actual goal. The fixture does not merely seek to handicap teams to the same level of difficulty, it seeks to promote “competitiveness” by reducing the exposure of the worst sides to difficult opponents.

We can see that the weighted fixture does effectively achieve its goal at the bottom end. The worst sides, which would have the hardest fair fixtures, get the biggest downward deviation to draw difficulty via their double-ups. This leaves them with the weakest schedule strengths, not just average ones.

At the top the outcomes are more mixed – GWS are getting a set of opponents that drags them further from their “fair” draw difficulty than Fremantle or Sydney from the top group are getting. Richmond, likewise, have a more strongly weighted schedule than Sydney.

Note that removing the impact of not playing one’s own team can be seen here clearly. Sydney have an easier overall Strength of Schedule than Collingwood or the Bulldogs, but those sides’ harder draws are due simply to playing Sydney rather than playing Collingwood or the Bulldogs.

Other measures of strength

The above analysis is based on actual win-loss outcomes as the most basic way to measure Strength of Schedule. Let’s now move into a couple of other measures which try to account for the impact of luck, in the form of close games and goal-kicking accuracy.

While calculating wins via actual results is common sense, it may not be the fairest way to account for the abilities of each team. For example, Hawthorn in 2015 lost a lot of close games to finish third on the ladder, but few would argue that they weren’t the dominant team.

To compensate for the vagaries of close results, we’ve stolen an idea from the father of sabermetrics, Bill James, and tried to apply it to the AFL (To read more about pythagorean expectations and the AFL check out the excellent Matter of Stats on the subject).

James looked at this formula to try to work out how teams “should” have performed, as opposed to how they actually did. By doing so, you can estimate how “lucky” team was during the course of a season.

James did this by looking at runs scored against runs conceded for each team. In an AFL context, looking at “points scored” against “points conceded” for each side is the most appropriate approach. The logic is that the more you outscore opponents by, the more often you should win.

A basic rundown of the Pythagorean Expectation is available on Baseball Reference, or via MAFL. The formula we have used is as follows:

Win % = (Points For ^3.87) / ( (Points For ^ 3.87) + (Points Against ^ 3.87)).

We’ve borrowed our exponent for this from MAFL, who got this number (3.87) in his earlier piece on this subject. We’ve used the Classic James formula, rather than the later models that MAFL has relied on.

Thirdly, we looked at goal-kicking. Teams rarely have the same accuracy for and against over a series of years. If a team is unusually accurate, or inaccurate, it tends to balance out a little the next season. This formula tries to account for that.

To account for accuracy luck more comprehensively we’d need access to something like Champion Data’s expected accuracy which draws on comparing goalkicking from various positions to the expected accuracy from that position. We don’t have that, so we just use the basic scoring shot data.

The formula is:

Win % = (Scoring Shots For ^3.87) / ( ( Scoring Shots For ^ 3.87) + ( Scoring Shots Against ^ 3.87)).

So who has the hard and easy draws?

Pulling the above together, we get the following Strength of Schedule across actual win-loss outcomes, and for two sets of expected win-loss outcomes based on scoring differentials and accuracy:

Strength of Schedule of 2016 fixture

Based on opponents, North Melbourne and Adelaide have the most difficult draws – Adelaide have it harder on actual win-loss record, but North have it harder on Pythagorean expected outcomes. This difference basically reflects that Hawthorn on its scoring dominance and luck with accuracy rates more as a 19-win team than a 16-win team, and North play them twice.

GWS also look like they have an anomalously difficult run for a team which finished 12th, as do Melbourne, especially based on the Pythagorean measures which rate their double-up opponent Hawthorn highly.

At the other end, St Kilda and Essendon get the easiest draws while Geelong and Port Adelaide, who finished 9th and 10th, shape as bolters in 2016.

Sydney and West Coast both sit in the bottom half of Strength of Schedule with almost exactly 0.500 draw, reflecting (as we saw above) that they’re good teams who don’t play themselves, and that their moderate burden from the weighted fixture does not outweigh that.

What does this mean in terms of actual games?

Let’s just take a step back here. Strength of Schedule matters. It changes the expected outcomes for each team. But a common mistake we make is to look at each game as either a “win” or a “loss” rather than a probability of each team winning, and thus overestimate the impact of the fixture. If a team plays North Melbourne they might have a 30% chance of winning. If they play Brisbane they might have a 70% chance. The difference between those two opponents is not 1 win just because one opponent is easier, it’s 0.4 expected wins. Strength of Schedule doesn’t decide premierships, it rarely even impacts ladder position all that much.

To illustrate that, we can add up the above Strength of Schedule measures into actual advantage over the course of the 22 game season and see, the impacts don’t look so bad:

Strength of Schedule expressed in season win advantage

Effectively, compared to a perfectly fair draw, North Melbourne and Adelaide are between 0.4 and 0.6 wins behind St Kilda and Essendon are 0.3 or 0.4 wins ahead. The gap between the hardest and easiest draw based on opponents is about 1 win.

The hardest possible draw (something like Carlton playing last year’s top 5 twice) would be about a 1.5 win disadvantage. The easiest possible draw would be around a 1.7 win advantage. Collectively that’s a range of 3.2 expected wins for mathematically possible draws.

While those differences are significant, they’re also a situation that wouldn’t occur under the current AFL Fixture rules. Within the constraints of an 18-team, 22 week season, the weighted fixture is pretty successful at minimising Strength of Schedule differences.

How did 2015 turn out?

The other difficulty with fixture weighting to compensate for unevenness in team numbers is that expectations based on the previous year’s ladder don’t always pan out.

We ran this analysis last year, and projected Geelong to have the hardest draw, followed by Port Adelaide. That sort of happened, but Port’s draw went from tough to hilariously brutal thanks to the Bulldogs and Adelaide’s improvement. Port Adelaide’s double-up opponents ended up finishing 1st, 3rd, 4th, 6th and 7th on the ladder. Given the Crows are a fixed opponent and the limitations of the AFL’s weighting policy, only replacing the Bulldogs with West Coast or Richmond could have made their draw harder.

Melbourne’s draw also became unexpectedly tough (6th hardest) due to improvement in the Bulldogs, GWS and to a lesser extent St Kilda.

On the other hand, Hawthorn’s draw, projected as the third toughest, turned out to be one of the easiest draws as Essendon, Carlton and Port Adelaide fell off various cliffs while Geelong also regressed.

To finish then, bearing in mind that all this analysis will turn out to be interestingly wrong, here’s a comparison of the actual 2015 Strength of Schedule and the projected 2016 Strength of Schedule to see who gets the most benefit heading into next year:

Strength of Schedule impacts, 2015 and 2016

From this we can expect Geelong and Port Adelaide to start confident of an improved 2016, and since they both just missed finals (finishing 9th and 10th), they will expect to push for a finals berth. Essendon and Gold Coast, too, look set to rebound a little.

Hawthorn theoretically present as likely to slide, as do Adelaide, North Melbourne and GWS. Remember that this is just the impact on expected wins from the strength of opponents, and plenty else will change between now and the end of next year. We, for one, won’t back Hawthorn to slide just because their fixture is harder.


It’s OVER – Saying goodbye to the 2015 AFL Trade Period

After a hectic last day, the 2015 AFL Trade Period is officially over! Here’s a quick “by-the-numbers” look:

  • 37 trades were lodged;
  • 4 Free Agents moved clubs;
  • 3 Compensation picks were generated;
  • 15 trades involved future draft picks;
  • 7 trades involved draft picks alone;
  • 3 trades were made involving more than two teams;
  • 13 players whose first names start with the letter “J” were traded.

HPN ended up writing 32 articles in the less than two weeks, but it felt like a lot more. We’d like to extend our thanks to all those who submitted comments and read our articles, and all of those who engage with us over Twitter, on Reddit, via Facebook and on the site itself.

We will be back with a recap of the entire period over the weekend, with a list of winners and losers. After that, HPN’s focus changes to the AFL draft, with the first edition of the Consensus Phantom Draft hopefully being released sometime next week.

Thanks for reading, and watch this space.


In Yarran Richmond Trusts #afltrades

Richmond fill one of their most glaring needs at the deadline, and Carlton accumulate another high pick.

Carlton value in = 980 points – Pick 19.

Richmond value in = 1312 points – Chris Yarran.

VerdictUnfair trade (Richmond gains 332 points of pick and player value, or 1.34 points back for every point given up).

For most of Thursday, it felt like this trade was going to be a touch-and-go proposition. As indicated above, on demonstrated value alone Yarran is worth more than pick 19 in isolation. For a different look on why, this piece by The Age’s Liam Mannix demonstrates why in a nutshell.

But Yarran wanted out of Carlton, and right now Carlton doesn’t have a need for players like Yarran. As difficult as it to frame, Carlton’s goal in 2016 isn’t necessarily to win games of football, it is instead to shape their list into a position where they can win games of football in 2019. Like the Saints recent teardown of their list, the Blues are rebuilding from the ground up around a select few older heads who want in for the long term. With his desire to leave the club, Yarran didn’t fit in those plans. If Yarran stuck around for an extra year before heading off during the next offseason, he might actually perversely hurt the club by helping them win a couple of marginal, if not meaningless, games next season, which would hurt their subsequent draft position.

Make no bones about it; Yarran is a good football player. Perhaps even a very good or great one if placed in the right situation. Last season, even though Carlton had an inability to score goals, Yarran was flung away from the forward line for much of the year. While Yarran has played some good footy running through the midfield or down back, he’s most dangerous with the goals in sight. This is likely why Richmond recruited them, as they were in desperate need for a somewhat defensively minded small forward who could punish the opposition if given space.

Earlier in the week when analysing the trade that sent Richmond’s pick 31 and their 2016 second rounder to Gold Cost as a part of the Dixon trade, we wrote:

“Richmond obviously have plans for the pick, reportedly offering it for Yarran, but if those plans don’t come to fruition they are going to look quite silly, and probably the first team to have conclusively stuffed up future pick trading. Early reports being that Carlton weren’t interested in pick 19 for Yarran – and though they may relent, they’re right that it’s modest unders for him (his true value is around pick 9). And even if the swap happens, if Carlton would prefer pick 19 alone to pick 30 and pick 31, that seems like a strange call on their part.”

This quote remains true, but Richmond’s ability to make the Yarran trade work somewhat redeems their involvement in the Dixon move. The Yarran trade was ultimately a game of chicken, and Carlton blinked first.

Western Bulldogs all but pay Sydney to take Michael Talia #afltrades

If Sydney do slightly better in 2016 than they did this year, then the Bulldogs have given Talia away for less than nothing. (Feel free to steal that example of a proper if-then statement, Damo)

Sydney value in: 1200 points (Michael Talia 815 points, pick 69 – 385 points

Western Bulldogs value in: 400 points (2016 pick ~68 points)

Verdict: Very unfair trade, Sydney get back 3 points of value for every point given up.

After the “talking to his brother” scandal, the Bulldogs have turned hard on Michael Talia and are turfing him for potentially less than nothing, making him a free hit at filling a dire need in the Swans’ future plans. Specifically, the nominal value of the Swans 2016 pick is 68, but that could easy fall below the 69 the Dogs traded to Sydney via either a better Swans performance (making a prelim final), or a couple of free agent compensation picks pushing the pick down the order. Unless Sydney slide down the ladder, the Bulldogs would probably have been better off simply delisting him.

Now, there’s a strong chance that Michael Talia is not a good player and he may never take a regular spot at the Swans, but they are desperate for a key position defender with any sort of experience between the ages of 20 and 30, and Talia fits that bill at no cost.

Talia goes into the mix with Aliir Allir and Xavier Richards for the key defender positions soon to be vacated by Ted Richards (33 in January) and Heath Grundy (30 in June). He fills a gaping hole in the Swans list profile, with the only other candidates being Sam Reid who is probably needed up forward with Goodes’ retirement, or mid-sized defenders like Rampe and Laidler. Spare a thought also for Alex Johnson, a 23-year-old who hasn’t played since the 2012 grand final due to an ACL injury and subsequent infection complications. He won’t play again before 2017, if ever.

Talia has played 9, 3 and 14 games over the last three years (an average of 8.7), and turned 22 in February. He got a Brownlow vote once. Based on his record so far, we’d therefore project him to play about another 81 games in his career, making his fair value around pick 37. Talia reaching that projection would be a very good result for Sydney.

Swapmageddon – Four swaps involving eight clubs and zero players #AFLtrades

Picks fly every which way as clubs align their selections with their draft strategy. Gold Coast has a bet on a Fremantle slide.

Sydney value in: 1630 points (pick 36 – 820 points, pick 37 – 810 points)

West Coast value in: 940 points (pick 23)

Verdict: Theoretically unfair if West Coast intend to make three live picks. Sydney get back 1.73 points of draft pick value back, but 1.20 points of academy bid matching. This pick swap actually would have made the Jetta-Sinclair swap very fair at 2121 points vs 2166 points.

For West Coast, they are placing a significant premium on moving their first pick to 23 (they previously had 31). This suggests to us a strategy of meeting their minimum 3 selection obligation with 2 live picks and then a senior list upgrade for at least one of their four rookie players, because otherwise 31, 36 and 37 looks a better potential draft yield than 23, 31 and 62, even in a weaker draft year.

Carlton value in: 1870 points (pick 11 – 1170 points, 2016 pick ~47 – 700 points)

Bulldogs value in: 1930 points (pick 20 – 970 points, pick 21 – 960 points)

Verdict: fair, Carlton get back 0.97 points for each point surrendered.

As if to demonstrate our earlier argument that the swap of picks isn’t inherently a symptom of Academy bidding, here’s a trade of picks which is all about old fashioned draft positioning. Carlton are loading up hard on high picks in the 2015 draft, while the Bulldogs – quiet during this trade period – are content to slide down to get two picks at 20, 21. Along with 30 that’s now their first three picks. This is a fair swap focused on different drafting priorities.

Collingwood value in: 1670 points (pick 27 – 900 points, pick 66 – 430 points, pick 77 – 230 points, pick 84 – 110 points)

GWS value in: 1930 points (pick 34 – 840 points, pick 53 – 610 points, pick 63 – 480 points)

Verdict: fair trade. Collingwood get back 0.86 points for every point given up. GWS get back 1.13 points for each point traded in Academy bid matching terms.

This will help GWS land Hopper and Kennedy so they’ll be happy. For Collingwood, they improve their first pick from pick 34 to pick 27 while sacrificing pick 53 and 63 for 66 and 77 as a second and third pick. They are paying a small premium in later picks for that upgrade to 27 but exiting this draft almost completely was probably always the price for executing a beautifully balanced trade for Treloar, and one expects that they’ll look to fulfill minimum drafting obligations with at least one of their three rookies or by delisting and redrafting someone. The movement of 84 to Collingwood seems pretty meaningless but may serve some administrative purpose or seek to “even out” the trade in the eyes of AFL overseers.

The other consideration is that the difference between 53 and 66 probably isn’t huge. In a widely considered weak draft pool, few clubs will still be using live picks in that range so if the Pies have a speculative punt in mind they’ll probably still be able to get them.

Gold Coast value in: 1410 points (pick 56 – 570 points, 2016 pick ~34 – 840 points)

Fremantle value in: 1340 points (pick 35 – 830 points, pick 61 – 510 points)

Verdict: fair trade. Fremantle get back 0.95 points for every point traded (depending on their draft position in 2016).

Gold Coast have no academy kids worth high bids this year but are positioning themselves well for a stronger crop next year including potential first round pick Brad Scheer. With this move they take in 6, 16, 29 and 56 as their first draft picks this year and now have Melbourne’s first rounder and second rounders from Richmond, Port Adelaide and Fremantle in 2016. If the ladder in 2016 stays close to this year’s, that gives Gold Coast something like picks 3, ~6, 21, ~28, ~30 and ~34 in the 2016 draft.

The Suns will therefore be barracking hard against Melbourne, Richmond, Port Adelaide and Fremantle next season. We also note that the Suns waited until the last moment, after it was clear Fremantle would not get McCarthy, before taking their bet on Fremantle’s 2016 fortunes. Fremantle once again failing to secure a key forward presumably increases their chances of a slide down the ladder and a better pick for the suns next year.

For the Dockers, their trade period leaves them with 22, 35 and 61 this year, having earlier used pick 16 in the Bennell trade.

Bastinac, Aish and a variety of draft picks change hands near the death #AFLtrades

Collingwood value in: 2756 points – James Aish (1306 points), pick 34 (840 points), pick 53 (610 points). Value out: 3410 points – pick 26 (910 points), pick 28 (890 points), pick 47 (700 points), 2016 2nd rounder ~ pick 26 (910 points). Total = -654 points.

North Melbourne value in: 3500 points – pick 17 (1000 points), pick 26 (910 points), pick 28 (890 points), pick 47 (700 points). Value out: 5561 points – Ryan Bastinac (1946 points), pick 34 (840 points), pick 38 (800 points), pick 40 (780 points), pick 53 (610 points), 2016 3rd rounder ~ pick 55 (585 points). Total = -2061 points.

Brisbane value in: 5021 points – Ryan Bastinac (1946 points), pick 38 (800 points), pick 40 (780 points), 2016 3rd rounder ~ pick 55 (585 points), 2016 2nd rounder ~ pick 26 (910 points). Value out: 2306 points – James Aish (1306 points), pick 17 (1000 points). Total = +2715 points.

Verdict: Unfair trade. Brisbane the huuuuge winners, North the big losers, Collingwood lose slightly but get the most promising player out of it.

Given the quick movements of the final day, we can’t spend too much time on this one, but it’s worth noting that the HPN formula rates Bastinac quite highly due to his “elite” loading for his high Rising Star finish, and his ability to play a lot of games at a young age. Aish’s form seemed to dive off a cliff this year, with his only area of statistical improvement being an increase in RB50s per game. Aish also spent a lot of time in their NEAFL team this year, and will have to work hard to crack the Collingwood midfield rotation.

North, on paper at least, gave up a lot for not a whole lot in return, but this loss would be mitigated if they don’t have any intention of using more than three draft picks this year. That would turn this result from a big on paper loss to a moderate one at best.

Pick swap – Hawthorn 53 and 58 for GWS 48 #AFLtrades

A simple win-win swap of picks, not a “loophole”, not evidence of a broken system.

Hawthorn value in: 690 points (pick 48)

GWS value in: 1130 points (pick 55 – 585 points, pick 58 – 545 points)

Verdict: theoretically unfair but practically win-win. GWS get back 1.63 points of value for what they give away. However, the Hawks probably wouldn’t use half the value they gave away anyway, and GWS’ actual benefit is much lower (1.29 points back) because they won’t use the picks live.

Hawthorn are giving away two mediocre picks for one slightly better one, with the likelihood that they were only going to use one pick here anyway (they also have pick 15 and 18).

There’s been some grumbling on social media that this represents a “loophole” whereby GWS can generate more pick value by treating draft picks as a tradeable commodity. This is because picks 55 and 58 are worth more bid-matching points than pick 48. For GWS it represents a modest upgrade of 75 points, or enough to meed a bid at pick 50. This might help GWS secure Eastlake product Harry Himmelberg as well Hopper and Kennedy. It’s a pretty slight upgrade, but it’s a win for them regardless, while also being a win for Hawthorn.

This is what trade period is all about – clubs generating value through the judicious exchange of commodities (picks and players) to meet needs.

This isn’t a “loophole”, it’s just about clubs having different needs in a market. GWS need more picks, Hawthorn need less, so they swap 2 for 1.

Now, the thing with the AFL’s draft pick value chart for academy and father-son selections is that it gives every pick an explicit values. And once you assign every pick a number, if you add two picks together they will equal some single better pick. That’s just the additive property of integers – numbers sum together to produce bigger ones. Under the AFL’s draft chart system, picks 55 and 58 add up to pick 43 as a completely fair trade. The difference between pick 43 and 48 isn’t a cause for intervention or panic, it’s just the fuzziness caused by who has what picks available and who has what priorities.

This is the sort of trade that could have happened before the market bidding system – if GWS intended to take more late picks and Hawthorn only wanted their minimum 3. GWS need more picks because of the Academy system in this case, but that’s not an inherent requirement for this sort of trade to occur. For example, in 2013 Collingwood traded 3 picks for 2 of West Coast’s picks in order to change their draft position from 11 to 6. West Coast wanted local boy Dom Sheed and figured he’d still be there at 11. Collingwood presumably wanted Scharenberg who would not drop to 11. The clubs had different needs, so exchanged picks accordingly. Collingwood gave up three picks for two, but got a better first pick as a result. It was a positive-sum game, both clubs got what they wanted.

In actuality, the value of 55 and 58 is substantially more than 48, not just the slight upgrade suggested by the Academy system. The benefit the Giants are getting is a lot smaller in academy points than it would be with the value of the live picks.

55 and 58 as live picks means two chancey players, both of whom have a possibility of producing games for their club. They have expected outputs of 58 and 54 games respectively. That’s compared to one pick at 48 with a slightly higher expected output (69 games). By pick 48 there’s little premium available, so all three picks are, in technical terms, a “crapshoot”. If Hawthorn wanted more live picks this would be a bad move for them because they lose one roll of the dice.

But what this trade illustrates is that value has a context. Hawthorn don’t value pick 58, so they’re happy to use it to improve the odds on their third pick in the National Draft. It’s not a win-win because of the Academy system itself, it’s a win-win because one club needs more picks than the other one does.