What is the point value of a turnover?

[logical_art]

Is there an accepted conversion ratio for the value of a turnover in terms of points forfeited?

[Doctor MJ]

Not that I'm aware of. Different metrics use different weights for every stat.

If you wanted a starting point I'd go with the average number of points per possession, which is typically around 1.05 in recent years.

[ElGee]

Not that I'm aware of. Different metrics use different weights for every stat.

If you wanted a starting point I'd go with the average number of points per possession, which is typically around 1.05 in recent years.

I use points per possession as well. It's a tricky thing if you actually think about it though. The bottom line is turnovers are bad bad bad. They are a possession without a shot AND often spike efficiency going the other way.

[Doctor MJ]

Not that I'm aware of. Different metrics use different weights for every stat.

If you wanted a starting point I'd go with the average number of points per possession, which is typically around 1.05 in recent years.

I use points per possession as well. It's a tricky thing if you actually think about it though. The bottom line is turnovers are bad bad bad. They are a possession without a shot AND often spike efficiency going the other way.

Part of what makes it so tricky is how you turn the ball over. If you lose the ball in a classic steal in the back court, that probably means 2 points the other way meaning the turnover could be thought to be worth -3.05 points.

[mysticbb]

If you wanted a starting point I'd go with the average number of points per possession, which is typically around 1.05 in recent years.

Points per possessions have turnover included. A turnover costs a team a scoring opportunity. Thus the scoring efficieny would be the way to go. Currently the average ts% for the league is 0.54. That makes it 1.08 as the value.

Well, the other point is the increased amount of fastbreak opportunities for the opponent. That is the tricky part as you and Elgee already mentioned. Basically an own turnover gives the other team an increased offensive effiency too. Thus the value should be higher than 1.08. Typically I use 1.1 or 1.2 in a simple linear model.

[Doctor MJ]

Points per possessions have turnover included. A turnover costs a team a scoring opportunity. Thus the scoring efficieny would be the way to go. Currently the average ts% for the league is 0.54. That makes it 1.08 as the value.

Well, the other point is the increased amount of fastbreak opportunities for the opponent. That is the tricky part as you and Elgee already mentioned. Basically an own turnover gives the other team an increased offensive effiency too. Thus the value should be higher than 1.08. Typically I use 1.1 or 1.2 in a simple linear model.

Well getting it perfect starts getting really tricky.

Average TS% is 0.54, but if a shot is missed and the team gets an offensive rebound then there's more opportunity to score which increases the value of a possession. On the other hand, not forcing a turnover early in a possession doesn't mean that there won't be a turnover later in the possession, so it doesn't make sense to ignore that possibility, and thus that decreases the value of a possession. And then as we're all agreeing, there's the matter of what kind of deviation from typical likelihood of success for the opponent comes with a turnover.

I think it's wise for any statistical modeler to just start somewhere based on the types of observations we've made here, adjust a bit for common sense, and move forward until there is reason to see a need for change. Your 1.1 or 1.2 seems perfectly reasonable to me in that regard.

[ElGee]

Points per possessions have turnover included. A turnover costs a team a scoring opportunity. Thus the scoring efficieny would be the way to go. Currently the average ts% for the league is 0.54. That makes it 1.08 as the value.

Well, the other point is the increased amount of fastbreak opportunities for the opponent. That is the tricky part as you and Elgee already mentioned. Basically an own turnover gives the other team an increased offensive effiency too. Thus the value should be higher than 1.08. Typically I use 1.1 or 1.2 in a simple linear model.

Well getting it perfect starts getting really tricky.

Average TS% is 0.54, but if a shot is missed and the team gets an offensive rebound then there's more opportunity to score which increases the value of a possession. On the other hand, not forcing a turnover early in a possession doesn't mean that there won't be a turnover later in the possession, so it doesn't make sense to ignore that possibility, and thus that decreases the value of a possession. And then as we're all agreeing, there's the matter of what kind of deviation from typical likelihood of success for the opponent comes with a turnover.

I think it's wise for any statistical modeler to just start somewhere based on the types of observations we've made here, adjust a bit for common sense, and move forward until there is reason to see a need for change. Your 1.1 or 1.2 seems perfectly reasonable to me in that regard.

Turnovers are actually the hardest stat for me to assign value to. I find they end up being one of the most highly weighted statistic (if not the highest) and that makes me uncomfortable. And I'm certainly not someone who ignores scoring like WP does. It's hard to get around that giving up a possession without even getting a shot is really really bad in a sport that averages over a point per possession.

Adding to what Dr. MJ said, we should see a relationship between when a TOV took place in a possession and the value of that turnover. If the turnover "eliminates" a possession with an expected value, and the EV of a possession decreases as the shot clock decreases, then we should see that in the value of a TOV. In other words, players don't give up 1.05 pts if they commit a TOV at 2 on the shot clock.

An ideal model would incorporate both the time-value of a TOV AND what happens going the other way relative to that opponent's expected value on offense. I have many blog posts on related topics already queued up so we can get into this more in the future, although personally I think the concept is more important than getting it perfect in a model.

[DSMok1]

I'm planning eventually to parse PbP to get "state" values for basketball (for instance, a possession started by a steal generates how many points?), but a basic look would be:

You have the ball, dribbling up the court. A neutral case would be a jump ball; in this situation (having the ball) your team has about a 1.05 point advantage, if there are infinite possessions left. You lose the ball, and now the other team has it, at about a +1.25 point advantage (possessions started by turnovers are higher expectancy--should differentiate between turnovers that end in a dead ball and "steals"). Thus the value of the turnover would be something like 2.3 points.

Does that make sense?

[Jimmy76]

It would be nice if they started tracking turnover types, throwing the ball out of bounds or stepping out of bounds probably leads to less opponent points than missing or even making a basket since you get so much time to set up your defense

[ElGee]

I'm planning eventually to parse PbP to get "state" values for basketball (for instance, a possession started by a steal generates how many points?), but a basic look would be:

You have the ball, dribbling up the court. A neutral case would be a jump ball; in this situation (having the ball) your team has about a 1.05 point advantage, if there are infinite possessions left. You lose the ball, and now the other team has it, at about a +1.25 point advantage (possessions started by turnovers are higher expectancy--should differentiate between turnovers that end in a dead ball and "steals"). Thus the value of the turnover would be something like 2.3 points.

Does that make sense?

I think your math is off, no? In your example, a turnover would be -1.25 points. You lose the possession from turning it over (1.05 points) and then increase the efficiency of the other team's possession - that they are to get anyway - by 0.20 points.

[DSMok1]

I'm planning eventually to parse PbP to get "state" values for basketball (for instance, a possession started by a steal generates how many points?), but a basic look would be:

You have the ball, dribbling up the court. A neutral case would be a jump ball; in this situation (having the ball) your team has about a 1.05 point advantage, if there are infinite possessions left. You lose the ball, and now the other team has it, at about a +1.25 point advantage (possessions started by turnovers are higher expectancy--should differentiate between turnovers that end in a dead ball and "steals"). Thus the value of the turnover would be something like 2.3 points.

Does that make sense?

I think your math is off, no? In your example, a turnover would be -1.25 points. You lose the possession from turning it over (1.05 points) and then increase the efficiency of the other team's possession - that they are to get anyway - by 0.20 points.

Yes, I think you're right... I think. These things always confuse me. Yeah, I think you're right.