Another interesting take on the merits of sports analytics thinking in the realm of public policy. Clearly incentivising for maximum expected value works but I doubt we can take the emotion out of policy completely. On a related topic, the expected value of life is a key issue in autonomous vehicle legislation. Coders have to tell the car to make a decision in an A or B crash fatality scenario and the legal framework doesn't yet shield them from the liability for telling the car to kill Granny, I believe.
As a committed FF player back in the day, I'm keen to understand more about how you went about valuing Mo. I was successful in the later 00s mainly down to a few useful heuristics on which player types over perform start of season price, which probably would appear in the data: attacking wing backs who took freekicks or corners, creative midfielders who also took set pieces and finally have as many penalty takers in the team as possible (I dreamt of a British Chilavert). My MVP was Cesc Fabregas who cost much less than Gerrard or Lampard but had 28 goal involvements in 09-10!
On to Mo, it would be great to understand how you valued the following:
- Availability. As the Americans like to say, it is the best ability. Given his playing style and age he has shown remarkable durability playing an average of 33.45 90s in the Premier League over the last eight completed seasons. His work ethic is famous but there has to be some luck in there that a hamstring never went. Did you weight for this and did Mo over perform?
- Set pieces. How do you factor set piece taking into your analysis? The number of penalties (and to a lesser extent free kicks and corners) awarded to a team is volatile, but favours better teams, and who takes them depends on availability, form and transfer activity. Clearly Mo taking and converting 9 pens last season (24% of his goals) was a massive factor in his record points total.
- Did you have an over-performance level at which you were happy to sell your MVP? Given fixtures and the variance in chance conversion I imagine there were seasons when Mo was 3 or 4 goals over expected part way through a season and his value was sky high. When you add in the seasons where Egypt have qualified for AFCON (which is always during his time at Liverpool) then there is clearly an arbitrage opportunity. I'm not sure what the transfer policy was in your league, but regardless would you have considered trading him in for an under performer in the same position e.g. a Bryan Mbeumo (any chance to mention the Bees will be taken) and take the cash to improve the rest of the squad? I'm sure you would never be loyal to the level of a sub-optimal performance in the TWFC cup.
Such a high-effort comment, the Gods of the blogosphere commend you, whoever They may be.
I'm planning to write (much?) more about Fantasyland modeling and the lessons that translate to "real life" (including how it teaches you the importance of incorporating new info about the likes of Bryan and the Bees and updating accordingly rather than hoping your original prediction comes true), but to take your specific Qs now as quick hits:
- Availability. I projected this as the weighted average of games played in the previous 3 seasons, with the weights derived from running regressions on historical data, and then regressed this to a league-wide mean (on the basis that injuries are somewhat randomly distributed). This is one of the places I'd make leg-break adjustments. At least as important was projecting minutes played per game: you get double appearance-points if you play 60 or more mins, and being on the pitch longer means you have more opportunity to score goal-and-assist-points.
- Set pieces. I didn't adjust for this specifically, other than penalties. Free kicks and corners show up in the historical data via expected goals and expected goals assisted, and therefore in the goals and assists projections. I had a penalty-taker variable, which was 0-1: 0 for most players, 1 for Mo, and e.g. 0.5 each for Saka and Odegaard before we knew who'd be on pens. Penalty points per game were penalty-taker x minutes per game % x points per penalty (factoring in likelihood of scoring and the points lost for missing) x penalties per game. I didn't adjust likelihood of scoring (by player) or penalties per game (by team) because they're both mostly noise.
- Trades. I never traded Mo, and don't remember every trying to. Maybe that was blind loyalty, but I think it's moreso because a) he was under-valued by other managers, b) the structure of the game makes it hard to do one-for-multiple trades of the kind you'd need to bridge the value gap between Mo and other players, and c) my rivals were generally more focused on making sure they didn't lose trades than trying to win them, which meant trades were rare (despite there often being positive-sum opportunities).
Another interesting take on the merits of sports analytics thinking in the realm of public policy. Clearly incentivising for maximum expected value works but I doubt we can take the emotion out of policy completely. On a related topic, the expected value of life is a key issue in autonomous vehicle legislation. Coders have to tell the car to make a decision in an A or B crash fatality scenario and the legal framework doesn't yet shield them from the liability for telling the car to kill Granny, I believe.
As a committed FF player back in the day, I'm keen to understand more about how you went about valuing Mo. I was successful in the later 00s mainly down to a few useful heuristics on which player types over perform start of season price, which probably would appear in the data: attacking wing backs who took freekicks or corners, creative midfielders who also took set pieces and finally have as many penalty takers in the team as possible (I dreamt of a British Chilavert). My MVP was Cesc Fabregas who cost much less than Gerrard or Lampard but had 28 goal involvements in 09-10!
On to Mo, it would be great to understand how you valued the following:
- Availability. As the Americans like to say, it is the best ability. Given his playing style and age he has shown remarkable durability playing an average of 33.45 90s in the Premier League over the last eight completed seasons. His work ethic is famous but there has to be some luck in there that a hamstring never went. Did you weight for this and did Mo over perform?
- Set pieces. How do you factor set piece taking into your analysis? The number of penalties (and to a lesser extent free kicks and corners) awarded to a team is volatile, but favours better teams, and who takes them depends on availability, form and transfer activity. Clearly Mo taking and converting 9 pens last season (24% of his goals) was a massive factor in his record points total.
- Did you have an over-performance level at which you were happy to sell your MVP? Given fixtures and the variance in chance conversion I imagine there were seasons when Mo was 3 or 4 goals over expected part way through a season and his value was sky high. When you add in the seasons where Egypt have qualified for AFCON (which is always during his time at Liverpool) then there is clearly an arbitrage opportunity. I'm not sure what the transfer policy was in your league, but regardless would you have considered trading him in for an under performer in the same position e.g. a Bryan Mbeumo (any chance to mention the Bees will be taken) and take the cash to improve the rest of the squad? I'm sure you would never be loyal to the level of a sub-optimal performance in the TWFC cup.
Keep up the saber-med-rics!
Such a high-effort comment, the Gods of the blogosphere commend you, whoever They may be.
I'm planning to write (much?) more about Fantasyland modeling and the lessons that translate to "real life" (including how it teaches you the importance of incorporating new info about the likes of Bryan and the Bees and updating accordingly rather than hoping your original prediction comes true), but to take your specific Qs now as quick hits:
- Availability. I projected this as the weighted average of games played in the previous 3 seasons, with the weights derived from running regressions on historical data, and then regressed this to a league-wide mean (on the basis that injuries are somewhat randomly distributed). This is one of the places I'd make leg-break adjustments. At least as important was projecting minutes played per game: you get double appearance-points if you play 60 or more mins, and being on the pitch longer means you have more opportunity to score goal-and-assist-points.
- Set pieces. I didn't adjust for this specifically, other than penalties. Free kicks and corners show up in the historical data via expected goals and expected goals assisted, and therefore in the goals and assists projections. I had a penalty-taker variable, which was 0-1: 0 for most players, 1 for Mo, and e.g. 0.5 each for Saka and Odegaard before we knew who'd be on pens. Penalty points per game were penalty-taker x minutes per game % x points per penalty (factoring in likelihood of scoring and the points lost for missing) x penalties per game. I didn't adjust likelihood of scoring (by player) or penalties per game (by team) because they're both mostly noise.
- Trades. I never traded Mo, and don't remember every trying to. Maybe that was blind loyalty, but I think it's moreso because a) he was under-valued by other managers, b) the structure of the game makes it hard to do one-for-multiple trades of the kind you'd need to bridge the value gap between Mo and other players, and c) my rivals were generally more focused on making sure they didn't lose trades than trying to win them, which meant trades were rare (despite there often being positive-sum opportunities).
I look forward to reading more about Fantasyland