An Extra Slice of An Analysis of Strength of Schedule for the Men's 2024 AFL Season
/I was thinking about the Strength of Schedule metric used in this blog from yesterday, and it struck me that, rather than using the raw values of the opponent team’s MoSHBODS rating and (for some metrics) the net Venue Performance Values (VPVs) for a game, we could, instead, convert these numbers into a win probability, which might make the resulting aggregate Strength of Schedule value more readilly interpretable.
CALCULATING A WIN PROBABILITY
Let me explain what I mean by way of an example.
Consider the first game of season 2024, which sees Sydney play Melbourne at the SCG. The relevant MoSHBODS numbers are:
Sydney Combined Rating: +5.49
Melbourne Combined Rating: +9.56
Sydney VPV for the SCG: +1.76
Melbourne VPV for the SCG: -9.75
So, firstly ignoring net VPVs, Sydney’s opponent’s Combined Rating is +9.56. Now MoSHBODS Ratings are measured in points, so a neutral team (ie a team with a Combined Rating of zero) playing Melbourne at a neutral venue would be expected to lose by 9.56 points.
To convert this to a win probability we can use the approximation that game margins are distributed as a Normal with a standard deviation of about 33 points of late. A neutral team’s win probability against Melbourne is therefore given by NORMDIST(-9.56,0,33,TRUE), which is 39%. This is the value we use in the Strength of Schedule metric ignoring VPVs for Sydney for this game.
To include net VPVs, we simply add the net VPV from Sydney’s viewpoint to the Combined Rating, which here gives 1.76+9.75-9.56 = +1.94. Converting this to a probability via NORMDIST(1.94,0,33,TRUE) gives a value of 52%, which is the probability of victory for a neutral team playing Melbourne at a venue that offers them a net +11.51 VPV.
We can proceed in the same way on all the remaining games, both in the fixture and missing from it, to arrive at the numbers in the table below.
We can interpret the numbers in this table as expected wins by an average team (ie rated 0) playing the opponents of a given team.
So, for example, the 11.28 number in the final column of the top table for Hawthorn can be interpreted as telling us that an average team would be expected to win 11.28 games against the opponents that Hawthorn faces in the actual fixture. if all those games were played at a neutral venue (ie where both teams had 0 VPVs).
If we adjust for the actual VPVs of Hawthorn and its opponents on the venues in the actual fixture (but still assuming an opponent of average ability), that number changes to 10.8 expected wins.
Interestingly, if we order the teams based on this latter metric (for which smaller values imply tougher opponents because an average team would be expected to win fewer games against stronger opposition), we arrive at exactly the same ordering as we did with the original Strength of Schedule metric except that North Melbourne and Sydney swap places.
One other insight we can glean from this is that an average team would be expected to win about 1.6 more games facing Adelaide’s schedule (with Adelaide’s VPVs) than facing Hawthorn’s.
Turning lastly to the missing schedule calculations, we see that Brisbane Lions are most disadvantaged by the truncated fixture, and West Coast least, regardless of whether we include or exclude VPVs.
Here, larger numbers imply greater disadvantage and we see that an average team facing Brisbane Lions’ opponents in the missing part of the schedule would be expected to win about 6 games aganist those 11 teams. In contrast, such a team would be expected to win only about 5 games against West Coast’s missing opponents.
The ordering of teams in this table exactly matches that in the corresponding table from yesterday’s blog.
CONCLUSION
If we convert opponent ratings and net VPVs into win probabilities, we can provide Strength of Schedule metrics that are more readily interpretable but that, it turns out, order the teams almost identically in terms of the measured strength of the actual and missing schedules.
