Tags:
Win Probability

Welcome to my blog! This is my first post so I wanted to start it off with something cool (in my opinion anyway) I've been working on. Basically I wanted a way to quantify the state of our game. I'm of the opinion that all this talk about the game being in a bad shape is an over reaction to one particular way of thinking about what a good game means.

To do this I'm going to use a model I built to predict the probability of a team winning, given the game time remaining, margin and relative scoring abilities of each team from my ELO model.

I've decided to look at a few different ways to quantify a good game of AFL so decided on the following kinds of games that might be deemed 'good':

- Games with lots of momentum swings and scoring.
- Games that could be lower scoring, but are 'in the balance' for a lot of the game, particularly towards the end of the match.
- Games where a team is either a big underdog or overcomes a big deficit to come back and win.

The converse of this is that bad games will be where the favourite runs away with an easy win. To relate these 3 points above back to my model, I'm going to think of these as being equivalent to:

- Games where the win probability varies a lot throughout the match.
- Games where the win probability is very sensitive to a goal being scored either way.
- Games where the winning team had a very low chance of winning at some point of the game.

I'll refer to the three types of games above as exciting games, tense games and surprising games respectively and will show an example of each below:

An example of a high excitement game would be 2016, round 5 Hawthorn vs Adelaide. Adelaide made a good start and lead by 23 points before Hawthorn came back to lead by 8. Adelaide looked to have won the match when they lead by 15 points late in the last quarter before Hawthorn kicked the last 3 goals to get the win. We can represent this in the chart below which shows how the margin and win probability of the home team (Hawthorn) varied throughout the match.

Next we've got tense games. One of these would be the drawn final with extra time in 2017 between Port and West Coast. As this match was in the balance, particularly in the final minutes, a goal either way would have moved the scoring teams chance of winning by a lot. For example, in the last quarter when Port scored 2 behinds in quick succession, this moved their win probability from 70% to 84%! Then even when no scores were made in the final 2 minutes, just due to time passing while they were in the lead, their win probability rose from 84% to 98%. This was before Luke Shuey kicked a goal after the siren to drop that down to 0%.

That brings us to surprising games. A great example is the round 13 game in 2013 where Brisbane overcame a 52 point margin in the 3rd quarter to win after the siren. Brisbane were a 0.04% chance of winning at this point according to my model.

The last thing I wanted to show was how many of each of these games we get per season. I've developed a metric to quantify how extreme a match ranks in excitement, tension and surprise. Below shows the distribution of these metrics each year (up to round 17 2018) since 2008 along with the distribution of all years since 2008. The charts are interactive so you can toggle each year on and off in the legend.

One thing I've noticed is that 2018 seems to have less high excitement games, but more high tension games. This makes sense because scoring is down, but I think defining good games as only high scoring is quite naive and means you miss out on a whole bunch of other kinds of great matches.

We can also see that during 2011, 2012 and 2013, the proportion of matches with low excitement, tension and surprise is higher than normal, probably due to the expansion clubs finding their way. These charts also highlight how great the 2017 season was, with more exciting, tense and surprising matches than almost every season since 2008.

Anyway that's it for now, I'll be exploring these ideas more so let me know if there's a path you want me to go down, always open for suggestions as to what to explore!