Why I'm not a TS% fan

[Dr Positivity]

I think I'm going to stop using TS%, for the most part. Here's why: I believe combining stats does us a disservice. My level of belief in a stat comes down to variables. The more variables that go into a stat, the less I trust it. Of course this is because we don't know which of the variables led to the conclusion. TS% has a handful of variables. It takes 2 point efficiency, 3 point efficiency, FT line efficiency, and the level of defensive pressure placed on the player and how 'hard' the shot he's taking. There's also the question of late shotclock and end of the half heaves that could alter the numbers. If all we had was the TS% numbers, we would have no idea how the player got there. Now we can look at the particular 2, 3, FT% stats aside from the TS% and more or less put together exactly how a player came to that overall efficiency. But this doesn't have anything to do with TS%. We could look at the particulars without it. These particulars have less variables, generally 3 FG% percentage for example comes down to the player and the team guarding him, and over the course of a season that latter one shouldn't have a big impact. In the rare case of playing in a system like the Phoenix Suns, that could also boost numbers.

The analogy I use is this. As staticians, our job is to take the game at its surface and zoom in with a microscope. The first zoom level is ppg, rpg, apg, bpg basic stats. You zoom more and you get some of our combined stats like TS%, PER, WS. You zoom in on the TS% and you can see what makes IT up: 2, 3, FT efficiency, and the defensive pressure and heaves stats we can't measure (yet). So to me taking those particular stats and combining it into TS% is actually going BACKWARDS in the microscope zoom. It simplifys the stats, when I believe IMO looking at the stats more broken apart and deeper in, is more valuable. A players' raw TS% does not tell us as much as looking at his 2 point, 3 point, and FT numbers. Those particulars tell us much more about what the player is effective at. We should be able to piece together on our own that a player who gets a lot of points from the FT line at a good percentage, while shooting a high number from 2 and a high number from 3, should be one of the league's best scoring options. TS% tries to simplify all shooting efficiency into one number, but why should we want that? There are so many different types of shots and scoring attempts that measuring it all by the same stat is doing us a disservice because it tells us less instead of more.

I believe we should have more stats, not less. I'm going to start emailing basketball-reference to include raw 2P% on their page, frankly it's embarrasing they don't have it, it'd be the easiest thing in the world to calculate and one of their most important basic stats. But in the future I think the most important new stats will be stuff like "20 feet jumper %", "Within 5 feet of the basket %", differentiating between open and contested 3s, and of course, defensive stats. These are the next step inward on the microscope zoom. Combining stats just doesn't interest me much because the only basis for it is coming up with quickfire arguments and making our knowledge of the game less complex

I feel the same way about baseball and OPS btw. OPS and I have always got on the wrong foot because there's no theoeretical basis for it to adding together OBP and SLG to me. It's pretty obvious they're not exactly equal. It'd be like adding together ppg, apg, and rpg and using it as the ultimate basketball evaluator. Maybe if they had a multiplication equation for it, that would make sense. Every time you get on base, * it by SLG. But the most obvious solution by far is just to forgo the laziness and talk about them separately. OBP makes sense, SLG makes sense, instead of combining them which doesn't make sense, just use 2 separate tools to evaluate a players batting. But w/e.

[Toizumi]

Well, on 82games they do have a chart that shows the shot selection (in percentages) for each player, divided in these categories:
- Jump
- Close
- Dunk
- Tips
- Inside (Close, Dunk, Tips combined)

They also show the FG% players shot in each category ans dome more stats. They only thing missing is a stat that shows what a players percentage on 2pters (jumpers) is like. This can easily be calculated though (with the jumper% stat, fga and 3pta stats).

I'd like stats like these to show up in regular box scores. Not a big fan of TS either, or any advanced stat (that uses a lot of calculations) for that matter. Pure stats I like. The list I named above are stats that aren't messed with (percentage of were a player takes his shots from).

[jicama]

You don't really lose any information by using a combined shooting stat. No one has thrown out the old FG%, FT%, and 3FG%.
I agree totally that 2FG% should be a separate entry. It is a separate event, after all.

But unless you want to look at 3 different accuracies, and then also look at 3 different frequency-of-attempt rates, even the shooting% don't tell you how effective a player was at getting points.

I can't look at six columns of % and tell how players rank in their overall shooting%. That's what TS% does.

[penbeast0]

On the other hand, I like TS% because it gives one of what I think of as the most basic stats, points per shot (it does adjust a little because the NBA doesn't count a shot with a foul as a shot unless it went in).

I do like combined stats where they give valuable information. TS% for me does combine numbers that make sense to combine; I don't like combined stats where the information is trivial or the aggregated numbers are poorly related (assists/turnovers is useful, assists + steals is not very)

[Ripp]

TS% is pretty legit...as penbeast0 said, it is basically points per shot. Obviously there are a lot of factors that go into how many points per shot any player produces, but it is sort of the first thing you'd like to look at when measuring offensive efficiency.

it takes 2 point efficiency, 3 point efficiency, FT line efficiency, and the level of defensive pressure placed on the player and how 'hard' the shot he's taking. There's also the question of late shotclock and end of the half heaves that could alter the numbers.

Defensive pressure is probably measured by something like USG%. If you use up a lot of possessions, presumably the defense pays a lot of attention to you...
Late shot clock heaves are not going to be a significant component of a player's TS%..makes no sense to discard a stat because of a minor error/flaw like this.

I believe we should have more stats, not less. I'm going to start emailing basketball-reference to include raw 2P% on their page, frankly it's embarrasing they don't have it, it'd be the easiest thing in the world to calculate and one of their most important basic stats. But in the future I think the most important new stats will be stuff like "20 feet jumper %", "Within 5 feet of the basket %", differentiating between open and contested 3s, and of course, defensive stats. These are the next step inward on the microscope zoom. Combining stats just doesn't interest me much because the only basis for it is coming up with quickfire arguments and making our knowledge of the game less complex

Agreed, more stats are useful for analyzing things. But the more data/stats you have, the more complexity and difficult you have interpreting the stats. There is always a tradeoff in life between descriptive power and ease of interpretation...the more complex you make things, the more difficult it is to intepret sometimes.
Like, stats like TS% make the game of basketball easier to understand rather than more complicated, imo.

[Doctor MJ]

On the other hand, I like TS% because it gives one of what I think of as the most basic stats, points per shot (it does adjust a little because the NBA doesn't count a shot with a foul as a shot unless it went in).

I do like combined stats where they give valuable information. TS% for me does combine numbers that make sense to combine; I don't like combined stats where the information is trivial or the aggregated numbers are poorly related (assists/turnovers is useful, assists + steals is not very)

Yeah, TS% isn't really a true combined stat a la PER - it's something that would be clear as day from the box score if the original basketball statisticians had a clue (and we literally would be calling in "Points Per Shot", which we can't do right now because it's come to mean something else).

As far as the combined stat issue in general, I get the purist perspective disliking it - but it's just so practical. If I'm trying to get a gauge of many players very quickly, it's a tremendous help. Agree they get overused by people who don't seek the nuance behind them, but just because a tool can be misused doesn't mean it shouldn't be used at all.

[NO-KG-AI]

I actually think TS% is one of the best, and most effective combined stats. Really, as long as you are differentiating guys who create shots, and guys who finish from creators, you get a really nice picture of how efficient their scoring is.

[azuresou1]

I actually think TS% is one of the best, and most effective combined stats. Really, as long as you are differentiating guys who create shots, and guys who finish from creators, you get a really nice picture of how efficient their scoring is.

This is one of the biggest problems IMO with advanced stats.

The other is that advanced stats don't tell HOW a player got those stats - a player who scores 25 pts in the first three quarters and then sits the 4th because his team blows out the other team has the same stats as a player who scores only, oh, 15 points in those first 3 quarters, and then scores 10 points in garbage time against bench scrubs.

[mysticbb]

I actually think TS% is one of the best, and most effective combined stats. Really, as long as you are differentiating guys who create shots, and guys who finish from creators, you get a really nice picture of how efficient their scoring is.

This is one of the biggest problems IMO with advanced stats.

The other is that advanced stats don't tell HOW a player got those stats - a player who scores 25 pts in the first three quarters and then sits the 4th because his team blows out the other team has the same stats as a player who scores only, oh, 15 points in those first 3 quarters, and then scores 10 points in garbage time against bench scrubs.

You just pointed out what NO stats can do. Those things are numbers, they need a proper interpretation, that isn't a problem of advanced stats, that is a problem of all stats.

Anyway, having ts%, fg%, 3p% and ft% you can basically say how a player plays on offense in terms of scoring. I used an example for this. Let me compare the 2007/08 JR Smith and Jason Kapono here.

Smith had 46/40/72 in that season, Kapono had 49/48/86. Now the non-advanced stats numbers will give you the impression that Kapono was more efficient as a scorer, in fact he had the higher fg%, 3p% and ft%. But a look at ts% reveals something here. JR Smith had 60.3 ts%, Kapono had 56.1 ts%. How is that possible? Well, rather easy, Smith used his ability to drive to the hoop (or cutting, slashing) and his range to either get a close shot, 3pt shot or a free throw. All of those are rather efficient ways to score. Kapono on the other end took a lot of long 2pt shots, had a lot of uncontested shots and didn't create a lot of of FTA. Thus overall his higher shooting percentages didn't translate in a more efficient scoring game.

With just having those 4 numbers fg%, 3p%, ft% and ts% you can approximately say how a player scores on offense. Not only how efficient, but you can also get a good impression about the way he is scoring. And that is the reason I'm a big fan of ts%.

[Doctor MJ]

This is one of the biggest problems IMO with advanced stats.

The other is that advanced stats don't tell HOW a player got those stats - a player who scores 25 pts in the first three quarters and then sits the 4th because his team blows out the other team has the same stats as a player who scores only, oh, 15 points in those first 3 quarters, and then scores 10 points in garbage time against bench scrubs.

mystic just gave a good reply by my own words: Not only do basic stats not inherently provide this information, but the people who actually do work to provide the the splits you want are always advanced stat people.

[mopper8]

No stat can tell you everything you need to know about a player. For example, if Keyon Dooling pounds the ball into the court for 8 seconds at the top of the key, makes a move, pulls his dribble back out, makes another move, still gets no where, and then with 5 seconds left on the shot clock passes into Brooke Lopez in the post, and Lopez misses a hurried layup attempt, that play does not register at all in Doolings ass%, TO%, TS% or any basic stat, etc etc, though it makes Brooke Lopez look worse. If Lebron James goes barreling down the lane to throw up a layup but gets called for a charge, that TO on a scoring attempt is ignored by all FG% metrics.

That being said, TS% is probably the best stat at approximating a player's scoring efficiency because it controls for Simpson's Paradox (mysticbb gave a good example of that in the post above). In one tidy neat number, it gives us a pretty good idea of the likelihood that a player will score a point when he attempts to score a point, in essence. At the end of the day, that's not the measure of a player's offensive worth, and as such TS% isn't the be-all/end-all, and it does miss some nuance for sure, but its probably the best tool we have IMO for measuring scoring efficiency

[Ripp]

I actually think TS% is one of the best, and most effective combined stats. Really, as long as you are differentiating guys who create shots, and guys who finish from creators, you get a really nice picture of how efficient their scoring is.

This is one of the biggest problems IMO with advanced stats.

The other is that advanced stats don't tell HOW a player got those stats - a player who scores 25 pts in the first three quarters and then sits the 4th because his team blows out the other team has the same stats as a player who scores only, oh, 15 points in those first 3 quarters, and then scores 10 points in garbage time against bench scrubs.

Regarding the first point, why not look at usage rate and assist rates? If your usage rate is high and the fraction of shots which you are assisted on is low, then you are likely creating a lot of your own offense.
Regarding your latter point, how frequently will that type of event happen? If it doesn't happen very often, then it won't corrupt the stat by much, and so it will still be useful. No reason to throw out the baby with the bath water.

[mysticbb]

Regarding the first point, why not look at usage rate and assist rates? If your usage rate is high and the fraction of shots which you are assisted on is low, then you are likely creating a lot of your own offense.

That might be an idea, but the assisted rate isn't an really accurate thing for that. The problem is the way assists are credited to players. I like to use the "Dirk Nowitzki" example here. The guy is usually operating in the mid post area without much movement. Thus everytime he makes a basket with just a fake (head, shoulder, pump) and takes only one step for a fade away shot the guy who dumped the ball to him (which isn't exactly the most difficult thing to do taking his height and catch abilities into account) will get an assist. Here is an example:

http://www.youtube.com/watch?v=4wbt9sbz6j4

That is the boxscore entry: 0:00 Dirk Nowitzki makes 19-foot jumper (Jason Kidd assists) 115-113

Jason Kidd didn't create anything in this situation for Nowitzki. And in this situation Nowitzki doesn't even has help to get open due to a screen, in fact he is setting screens for other players before.

And especially for bigger players who are operating in the post the assisted rate is screwed up. The entry pass alone is most times enough to get that assist, even though the player with the ball in his hand has to work for his shot via post moves.

It is also most times not the case that a player really creates for himself from the perimeter. Most times the player gets an effective screen to get open, but in that case no assist is given out. Assisted rate might be a starting point, but due to the rather subjective way the assists are given out and due to the fact that effective screens aren't counting as an assist, that stats is probably the most misleading one of all advanced stats.

[Ripp]

Hrm, that definitely should not have been credited as an assist...
Hrm, so I wonder how systematically this sort of thing happens. If not too much, then it will not affect things too much...you can just view it as random noise.
But if assisted rate is biased against certain positions, then that is a sort of systematic bias that you cannot really treat as random noise.
Kind of interesting...I'm not sure what the best way to handle this is. I guess you could distinguish between different type of assists, as hoopdata.com sort of does...assists at the rim, 16-23 foot jumpers, etc. Maybe also track assists to wings versus bigs? I'm not really sure.

[mysticbb]

Unfortunately it seems like that is not just random noise. I guess there is an bias in the assists stats which favours the perimeter players. And Nowitzki isn't the only example, you can see similar things for a lot of players. It seems like the scorekeepers are pushing assists numbers in the NBA, especially for smaller guards (e.g. point guards).

You can probably reduce the "error" or "bias" by tracking different kind of assists from different players. I guess the teams are tracking already the way a player scores, how the shots are creating, they need those informations anyway to make a proper game plan. It is maybe just not accessible for us.