Clean Sheet Conversion Rates
In the last post about the new model analysis I'm doing with the Shots on Target team, I touched on the fact that my current defensive forecasts only forecast goals conceded which is obviously useful for a given week, but could be misleading over a number of weeks as you'd rather have a team with three very low scores and two very high scores than one with five medium scores, even if their average goals conceded were equal. That got me thinking about at what rate teams 'convert' good clean sheet opportunities and while the below data is based on an incredibly small sample size (three gamesweeks) and thus shouldn't be directly relied upon, I think it shows that we're onto something and move prove useful in the future:
Despite having just 60 defensive appearances to go on, the data is already stacking up in a logical order with only a couple of outlier percentages appearing (for example, it is clearly erroneous to suggest that a team with a forecast goals per games of 1.1 has a better chances of clean sheet than a team with a forecast goals per game of 1.0 but with such a small sample size, one or two games can cause slight inaccuracies).
Where I think we'll end up in a couple of weeks when I start to apply all this data, is with a series of plateaus, for example, if you're forecast to concede 0.8 GPG or less you have a ~50% chance of a clean sheet, 0.8-1.0 gives you ~40% etc. I believe that aggregating these chances at clean sheets will give a better measure of future success given the fact that the 4 point gain is a strict win/loss scenario with no bonus points for always holding opponents to a single goal.
The next step will the be to look at how each individual team have performed against expectation to see if we can spot trends of teams who consistently fail to 'convert' good clean sheet chances, though that will need a bit more data before those numbers stabilise. For now, we can at least consider the above data when looking at the 8 week team forecast to give a bit more depth to the likelihood of a team's success.
Despite having just 60 defensive appearances to go on, the data is already stacking up in a logical order with only a couple of outlier percentages appearing (for example, it is clearly erroneous to suggest that a team with a forecast goals per games of 1.1 has a better chances of clean sheet than a team with a forecast goals per game of 1.0 but with such a small sample size, one or two games can cause slight inaccuracies).
Where I think we'll end up in a couple of weeks when I start to apply all this data, is with a series of plateaus, for example, if you're forecast to concede 0.8 GPG or less you have a ~50% chance of a clean sheet, 0.8-1.0 gives you ~40% etc. I believe that aggregating these chances at clean sheets will give a better measure of future success given the fact that the 4 point gain is a strict win/loss scenario with no bonus points for always holding opponents to a single goal.
The next step will the be to look at how each individual team have performed against expectation to see if we can spot trends of teams who consistently fail to 'convert' good clean sheet chances, though that will need a bit more data before those numbers stabilise. For now, we can at least consider the above data when looking at the 8 week team forecast to give a bit more depth to the likelihood of a team's success.
Comments