C/D: TS% & FGA are poor metrics for measuring efficency

[azuresou1]

The intention of TS% is to show how efficient a player is at scoring. To do this, FTAs get weighted at .44 FGA. This means that an And-1 counts as 1.44 FGA for 3 points, while two made shots counts as .88 FGA for 2 points, and actually statistically punishes the player for converting the And-1. An you're just flat out screwed if you make the shot but there's no call, with its 1 FGA for 2 points.

In essence, TS% massively overstates FTs as if they are generated in some void that is arbitrarily added on, rather than as if they were a byproduct of an actual attempt to score.

This stems from the issue that FGA is not, in fact, FGA, since getting fouled and failing to make the basket means you don't get a FGA assessed.

My proposal
FGA SHOULD be measured as any attempt to score on the basket. TS% could then be scraped as a method of calculating efficiency, and PPS would take its place as a pure measure of scoring efficiency.

Thoughts?

[Scott Carefoot]

Because I use TS% a lot in my analysis, I'm curious to hear people's thoughts on this. My contribution is this chart listing the top 15 players in TS% (min. 600 points) so far this season, along with their PPS at the end. Clearly, there's a signficant discrepancy between the two stats. Which is a better measurement of offensive efficiency and why?

[azuresou1]

Both stats aren't great right now, due to how FGA is IMO defined really dumbly.

[DSMok1]

What do you want to measure? TS% is a better measure of who is a better "scorer" efficiency-wise; PPS is a better measure of who is a better "shooter" efficiency-wise.

I like TS%, despite the limitation of the 0.44. Technically, it should be calculated directly from PBP data, with an AND-1 still counting as a single FGA, while 2 regular FT's in the bonus count as 1 FGA, being fouled on the shot and shooting 2 or 3 FT's still counts as 1 FGA, and 1 FT from missing the front end of a 1-and-1 counts as 1 FGA.

If those corrections were made, azure, would you be happy using TS% then?

[azuresou1]

What do you want to measure? TS% is a better measure of who is a better "scorer" efficiency-wise; PPS is a better measure of who is a better "shooter" efficiency-wise.

I like TS%, despite the limitation of the 0.44. Technically, it should be calculated directly from PBP data, with an AND-1 still counting as a single FGA, while 2 regular FT's in the bonus count as 1 FGA, being fouled on the shot and shooting 2 or 3 FT's still counts as 1 FGA, and 1 FT from missing the front end of a 1-and-1 counts as 1 FGA.

If those corrections were made, azure, would you be happy using TS% then?

Not sure I understand the last sentence, but yes, that's what I'm aiming at.

[Scott Carefoot]

What do you want to measure? TS% is a better measure of who is a better "scorer" efficiency-wise; PPS is a better measure of who is a better "shooter" efficiency-wise.

This doesn't compute, since Billups has a better TS% than Howard while Howard has a better PPS than Billups.

[ElGee]

What do you want to measure? TS% is a better measure of who is a better "scorer" efficiency-wise; PPS is a better measure of who is a better "shooter" efficiency-wise.

I like TS%, despite the limitation of the 0.44. Technically, it should be calculated directly from PBP data, with an AND-1 still counting as a single FGA, while 2 regular FT's in the bonus count as 1 FGA, being fouled on the shot and shooting 2 or 3 FT's still counts as 1 FGA, and 1 FT from missing the front end of a 1-and-1 counts as 1 FGA.

If those corrections were made, azure, would you be happy using TS% then?

Not sure I understand the last sentence, but yes, that's what I'm aiming at.

Below the NBA level, they use a "1-and1." Only get a second shot by making the first.

TS% attempts to calculate points per shot. But it fudges with the 0.44 estimation. DS is saying we could literally make it perfect by using play by play data. I agree, although I don't know how much would change things. Players who shoot a lot of free throws might see a small decline.

[Paydro70]

This is sort of a silly point... if you want a perfect measure, you can either alter FGA so that PPS covers free throw accuracy, or you can alter TS% to actually count the number of and-1s and adjust accordingly. It doesn't really make a difference which way you go, they'd both achieve the same goal.

[azuresou1]

This is sort of a silly point... if you want a perfect measure, you can either alter FGA so that PPS covers free throw accuracy, or you can alter TS% to actually count the number of and-1s and adjust accordingly. It doesn't really make a difference which way you go, they'd both achieve the same goal.

Except we currently can't do that, which leads to guys who draw lots of FTAs like Kevin Martin and Maggette have ridiculous TS% despite really not being all that efficient when you consider how many touches they get.

[tommer]

This means that an And-1 counts as 1.44 FGA for 3 points, while two made shots counts as .88 FGA for 2 points, and actually statistically punishes the player for converting the And-1.

I'm not so sure about that -- I think that in virtually any realistic scenario, the and-1 would be more beneficial to overall TS%.

[ElGee]

Wrote a post about the accuracy of TS%: http://elgee35.wordpress.com/2011/03/04 ... -shooting/

Personally, I think it does a pretty darn good job. Obviously we can see outliers can have errors on order of 0.5 to 1% (eg Reggie Evans in 2006) .

[Doctor MJ]

The intention of TS% is to show how efficient a player is at scoring. To do this, FTAs get weighted at .44 FGA. This means that an And-1 counts as 1.44 FGA for 3 points, while two made shots counts as .88 FGA for 2 points, and actually statistically punishes the player for converting the And-1. An you're just flat out screwed if you make the shot but there's no call, with its 1 FGA for 2 points.

In essence, TS% massively overstates FTs as if they are generated in some void that is arbitrarily added on, rather than as if they were a byproduct of an actual attempt to score.

This stems from the issue that FGA is not, in fact, FGA, since getting fouled and failing to make the basket means you don't get a FGA assessed.

My proposal
FGA SHOULD be measured as any attempt to score on the basket. TS% could then be scraped as a method of calculating efficiency, and PPS would take its place as a pure measure of scoring efficiency.

Thoughts?

Well, I'd think everyone pretty much agrees with everything here except the vehemence. As long as we only have the data we historically have, TS% essentially does as well as you can possibly do. I wouldn't be shocked if someone came up with a study that said .44 isn't quite optimal, but that's as far as it goes.

But yes, ideally, scorekeepers would have simply done all this different from the beginning, and then we'd have better data.

Interesting though, I'm not aware of any site trumpeting a true TS%/PPS system for current seasons, and there's no reason we can't have that with the level of detail sites like synergy and hoopdata have. Makes me think I've just missed it. Anyone aware of such a site?

[Doctor MJ]

Because I use TS% a lot in my analysis, I'm curious to hear people's thoughts on this. My contribution is this chart listing the top 15 players in TS% (min. 600 points) so far this season, along with their PPS at the end. Clearly, there's a signficant discrepancy between the two stats. Which is a better measurement of offensive efficiency and why?

TS% is better than the standard PPS. That penalty the OP mentioned for the AND1 in TS% is much more severe with PPS.

An AND1 situation in TS% take 3 points and divides by 1.44 to get a value of 2.08, and the two made foul shots counts for 2 on 0.44 shots for a value of 4.55.

Obviously that's bad, but PPS on the two made foul shots gives 2 points for zero shots, for a per shot value of infinity.

If you're going to use one number for efficiency, use TS%, but I am a proponent of using both 1 number estimates and more detailed estimates when it comes to pretty much everything in the game, depending on the relative breadth vs depth of what you're trying to get your head around.

[Paydro70]

This is sort of a silly point... if you want a perfect measure, you can either alter FGA so that PPS covers free throw accuracy, or you can alter TS% to actually count the number of and-1s and adjust accordingly. It doesn't really make a difference which way you go, they'd both achieve the same goal.

Except we currently can't do that, which leads to guys who draw lots of FTAs like Kevin Martin and Maggette have ridiculous TS% despite really not being all that efficient when you consider how many touches they get.

Kevin Martin is #4 in PPS, and Maggette #9. In TS% they're both lower. PPS inflates the value of free throws more than TS%, because they truly are "free" as currently recorded. The point is that neither one is perfect, and it really doesn't make sense to target TS% as the "problem" when it's actually MORE accurate currently than PPS.

The basis of either one can be altered to be better... except one requires changing the definition of a fundamental of basketball (FGA) and the other would just involve actually counting FTs instead of estimating them.

[azuresou1]

This is sort of a silly point... if you want a perfect measure, you can either alter FGA so that PPS covers free throw accuracy, or you can alter TS% to actually count the number of and-1s and adjust accordingly. It doesn't really make a difference which way you go, they'd both achieve the same goal.

Except we currently can't do that, which leads to guys who draw lots of FTAs like Kevin Martin and Maggette have ridiculous TS% despite really not being all that efficient when you consider how many touches they get.

Kevin Martin is #4 in PPS, and Maggette #9. In TS% they're both lower. PPS inflates the value of free throws more than TS%, because they truly are "free" as currently recorded. The point is that neither one is perfect, and it really doesn't make sense to target TS% as the "problem" when it's actually MORE accurate currently than PPS.

The basis of either one can be altered to be better... except one requires changing the definition of a fundamental of basketball (FGA) and the other would just involve actually counting FTs instead of estimating them.

I think you missed the entire point of my original post. The problem is that TS% and PPS BOTH inflate Kevin Martin and Maggette's numbers, when I'd be willing to bet that they hoist up a similar number of attempts as a guy like Joe Johnson who takes almost no FTAs.

[Paydro70]

I think you missed the entire point of my original post. The problem is that TS% and PPS BOTH inflate Kevin Martin and Maggette's numbers, when I'd be willing to bet that they hoist up a similar number of attempts as a guy like Joe Johnson who takes almost no FTAs.

I didn't miss the point of your original post, those points just didn't make sense so I focused on the one part that did: that we estimate when using TS% when we could be counting. But fine...

The OP makes 3 points:
1) That a player is "punished" for completing an and-1. This is simply false. A player gets an and-1. He makes his free throw: he has a TS% of 1.04 (thanks to the "estimate" on FT attempts). If he misses it, his TS% is .694.
2) That a player is "flat out screwed" if he gets fouled and there's no call. Well duh... no statistic is going to correct for referee error, that is beyond the scope of statistical adjustment. Even doing what you suggest, redefining FGA, will still make a shot worth only 2 points even if there's an uncalled foul.
3) That TS% adds FTs on "arbitrarily" as if they do not count as attempts. This doesn't really make sense, because FTA is a part of the TS% formula. With default PPS, yes, all free throws are a bonus because there is no accounting for FTA. That is not the case with TS%, where a player is only rewarded for more FTAs insofar as he makes them... though since nearly everyone in the NBA shoots better from the line than the field, taking more free throws typically makes you more efficient. I don't understand your difficulty with the idea that a guy who gets to the line a lot *IS* more efficient than a guy who doesn't. Certainly the example of Ellis/Martin and Joe Johnson isn't illuminating, since the former two are more efficient than Johnson from the field in addition to getting to the line more.

Back to the actually useful point from the OP: TS% does only ESTIMATE how many "shots" the player took, rather than COUNT them. Changing the definition of FGA as you suggest would therefore alter the results, but only very slightly. The only difference between "your" PPS and current TS% would be that players who get a disproportionately high or low percentage of their free throws on and-1s would be benefited or penalized (respectively). Players who take a lot of free throws would not see any substantial change in their rating relative to players who do not.

[azuresou1]

You clearly have misinterpreted the points.

1) I clearly state that making and and-1 is detrimental when compared to getting fouled, NOT making the shot, and then making the free throws. Your interpretation is off.
2) The point of the no-call is that making two free throws on a foul is 2 points over .88 FGA while making a shot is 2 points over 1 FGA. How is that even remotely logical?
3) Why do you not get the simple concept that getting to the line still requires a shot attempt? If I drive and get fouled three times and make my 6 FTs, how is that better than me driving three times and making my 3 FGA? It isn't... and yet in the first case your TS% and PPS skyrocket, while in the second you'd still have an excellent TS% and PPS, but not nearly as good.

[mysticbb]

You clearly have misinterpreted the points.

No, he got every point, you just can't handle his answer, because it makes your op obsolete.

The way you are exploring this will never give you a chance to understand the concept of that specific stats (nor any stats at all). The formula for ts% approximates the amount of scoring opportunities a player used. The 0.44 comes from real game data, it works with a big enough sample. Using it for a one game sample might give you inaccurate results. That's just the way it is. TS% would be absolute correct, if it would be derived by play-by-play data rather than using an approximation for the true shooting attempts.

[Doctor MJ]

Wrote a post about the accuracy of TS%: http://elgee35.wordpress.com/2011/03/04 ... -shooting/

Personally, I think it does a pretty darn good job. Obviously we can see outliers can have errors on order of 0.5 to 1% (eg Reggie Evans in 2006) .

OP have you read ElGee's piece here?

Basically the answer is: Stuff averages out pretty darn well. To the point where I don't see a reason for concern unless you're actually concerned with players intentionally missing shots to up their TS%. Hard to imagine that anyone really thinks that's a realistic danger though.

[therealbig3]

Because I use TS% a lot in my analysis, I'm curious to hear people's thoughts on this. My contribution is this chart listing the top 15 players in TS% (min. 600 points) so far this season, along with their PPS at the end. Clearly, there's a signficant discrepancy between the two stats. Which is a better measurement of offensive efficiency and why?

TS% is better than the standard PPS. That penalty the OP mentioned for the AND1 in TS% is much more severe with PPS.

An AND1 situation in TS% take 3 points and divides by 1.44 to get a value of 2.08, and the two made foul shots counts for 2 on 0.44 shots for a value of 4.55.

Obviously that's bad, but PPS on the two made foul shots gives 2 points for zero shots, for a per shot value of infinity.

If you're going to use one number for efficiency, use TS%, but I am a proponent of using both 1 number estimates and more detailed estimates when it comes to pretty much everything in the game, depending on the relative breadth vs depth of what you're trying to get your head around.

I don't think you're actually using the TS% formula though. Using the full formula, for an AND-1, you have a TS% of 104%. For 2 made FTs, you have a TS% of 114%. Much closer.

Here's the math, in case I did it wrong:

AND1: 3 PTS/(2*(1+.44*1)) = 1.04

2 made FTs: 2 PTS/(2*(0+.44*2)) = 1.14

So I don't think it's THAT flawed: obviously, an AND1 is much more desirable than 2 made FTs, so the stat is flawed to an extent, but taking into account that it's not that big of a difference, and in addition to ElGee's post, I think it just confirms that TS% is the best way to measure how efficiently someone scores.