Didn't they have points per post-up possession - it was quoted a lot for Dwight. Wouldn't they have an aggregate one for an individual covering all possession types? That is what you seem to be asking for and what I was thinking of when I said "points per possession " I have heard of points off assists from driving. I don't know if a drive where you get an assist counts as a possession though. That is why I was thinking of something like usage as a denominator, though I confess I don't know that calculation. Im basically looking for an ROI on the time the player has the ball in his hands. With return being points not rebounds or another peripheral stat. You could have points from his actions (FGM, FTM, Assists) or just team points. Id be interested in both. The difference from ORTG as I understand it would be it would measure return on the ball being in the player's hands vs time on court. Im just spit balling. I really just want that stat where the and 1 counts the same as a 3PM
@Easy @JayZ750 @oakdogg I was being somewhat disingenuous in previous post by saying TS% is awarding all free throws more than normal. While that's true, the formula merely wants to treat AND1s the same as 3ptrs, so they don't hurt your TS% if you miss them. So, my post was meant to get you thinking about the overall effect of using an league average multiplier to simulate AND1s as 3ptrs and ignoring Techs altogether. To make AND1s score the same as 3Ptrs, and If you were to use the PBP and do it by hand, you just don't include that FTA in denominator like this: 3ptr made using the normal .5 math: 3 / ((1 FGA + .5 * 0FTA) * 2) = 3/2 = 150% AND1 made using the normal .5 math: 3 / ((1 FGA + .5 * 1FTA) * 2) = 3/3 = 100% AND1 made but leaving the FTA out of the denominator like PPS does: 3 / ((1 FGA + .5 * 0FTA) * 2) = 3/2 = 150% And conversely, if you miss the AND1, the score counts same as the 2ptr you made. So, since they aren't scraping the PBP after every game to get the true TS% for a player, by manually subtracting AND1 FTAs from the denominator (and removing Techs entirely from the equation), they then need an average/estimate multiplier to replace .5. So I believe they simply counted up all the AND1s/Tech league-wide and ran a league-wide TS% with AND1 FTAs in the denominator and ran the same calculation without those FTAs in the denominator, the latter allowing AND1s to count identical to a 3ptr. Then they applied a multiplier different than .5 until the two league-wide results were equal. @durvasa -- correct me if they came up with the .44 differently. ------------------------------------------- Slight problem with this is ALL free throws are given an extra value in order to treat AND1s like 3Ptrs....including hack a shaq fouls and other non-shooting fouls leading to free throws while in the bonus. This leads me to ask, hell, what's so wrong with PPS if TS% effectively deletes some FTAs from the denominator. And as JayZ750 says, when are we going to be able to just run a true TS% that treats AND1s and Tech differently, but keeps the natural .5 multiplier for other shooting fouls. Further, should we just ignore all non-shooting fouls, like Techs? -------------------------------------------
I still believe this is the purpose of Ind ORtg. It includes total points produced and all parts of usage, resulting in ORtg = 100 * (PProd / TotPoss) PProd includes stuff like Assists and the value of second chance points from offensive rebounds. TotPoss sub-formula covers everything that Usage covers. I don't know the two formulas very well (both quite complex), but I'm assuming they are statistically very similar. Here's what the inventor says: http://www.basketball-reference.com/about/ratings.html In Dean's words, "Individual offensive rating is the number of points produced by a player per hundred total individual possessions. In other words, 'How many points is a player likely to generate when he tries?'" The basic building blocks of the Offensive Rating calculation are Individual Total Possessions and Individual Points Produced.
@heypartner Thanks for the explanation of the possible reason of the .44 factoring. The average FTA per game for a team is slightly more than 23. 14% of 23 is slightly more than 3. Does a team get an average of about 3 And1's per game (assuming techs are fairly rare, less than one per game)? Seems about right to me. Treating And1 as a 3pt shot makes sense. You don't get penalized for missing a 3pt shot by getting .5 FGA more, so you should not be penalized for missing an And1 FT. Counting FT shooting for non-shooting fouls also makes sense. You "earned" the trip to the foul line by drawing a foul, shooting or not. I do not see how FTs drawn from shooting or non-shooting are different. I do think intentional fouls (not just the hacking fouls) is a problem. Players get those FTs by virtue of the other team's management of the game situation, rather than their skills.
I agree wrt intentional fouls. Regarding non-shooting fouls, did you put the other team in the bonus by yourself? no. You're benefiting from other fouls. The stat then is deciding to award some fouls and not others that are identical, but just not in the bonus. But the man point is TS% is supposed to be about shooting. You might not even have the ball. You could be fighting through a screen without the ball ... or setting a pick. Even loose-ball fouls are awarded FTs when in the bonus. Seems more of a role for ORtg than something called True Shooting %. Are these minor points, sure. But it is possible to decide and make a true TS% with PBP. Why not? It's just computer code. And who cares if it alters history for years that data isn't available. We have a lot of stats suffering from that, right?
Man, heypartner, I envision you as some genius investment banker who made millions by a young age and whose early retirement freed you to unleash your intellectual horsepower on your true passion - basketball.........I will definitely have to make time to digest these great posts. I always appreciate your great contributions! I appreciate everyone's insights.
Many thanks heypartner. I think we're on the same page. Not 100% sure why the number is 0.44 but have to assume it gave the best fit / r-squared in some way probably like you noted (league or such). And yes, at this point it seems like they could just use the data to use actual numbers for every player instead of the formula approximation. Still have your question as to how to calculate the actual number wrt AND1s, technical fouls.... and even common fouls as you note. Alas there's tons of advanced stats out there now and I suspect the better organizations ARE using more detailed play by play analysis.
You can get true shot attempts from NBAWOWY (bases on scraping the PBPs rather than using the TSA estimation formula). I searched my posts from a few years ago where I grabbed the true shot attempt stats from that site and compared how close "true" TS% compared to the quick and dirty TS% formula everyone uses. Code: Player pts fga fta tsa (est) TS% (est) FT_coeff (act) tsa (act) TS% (act) TS% diff James Harden 1450 958 540 1196 60.6% 0.42 1186 61.1% -0.5% Chandler Parsons 831 674 113 724 57.4% 0.43 723 57.5% -0.1% Jeremy Lin 720 609 180 688 52.3% 0.44 689 52.2% +0.1% Omer Asik 588 432 227 532 55.3% 0.45 535 55.0% +0.3% Patrick Patterson 536 450 49 472 56.8% 0.45 472 56.8% +0.1% Carlos Delfino 474 421 30 434 54.6% 0.40 433 54.7% -0.2% Marcus Morris 450 391 75 424 53.1% 0.45 425 52.9% +0.1% Toney Douglas 389 344 66 373 52.1% 0.41 371 52.4% -0.3% Greg Smith 253 159 76 192 65.7% 0.45 193 65.5% +0.2% Patrick Beverley 90 70 19 78 57.4% 0.47 79 57.0% +0.5% Daequan Cook 55 59 3 60 45.6% 0.33 60 45.8% -0.2% Donatas Motiejunas 65 48 15 55 59.5% 0.47 55 59.1% +0.4% Cole Aldrich 50 43 9 47 53.2% 0.44 47 53.2% +0.0% James Anderson 52 41 7 44 59.0% 0.43 44 59.1% -0.1% Terrence Jones 34 34 6 37 46.4% 0.50 37 45.9% 0.5% Scott Machado 8 6 2 7 58.1% 0.50 7 57.1% +1.0 That's pretty good. We're talking fractions of a percentage point off in all cases for players who got significant playing time. I see no reason to be concerned over "and-1's" not counting the same as a made 3-pointer. I think you're fretting over the TS% awarded for a single play, rather than TS% aggregated over several games. The latter is what we care about, and TS% is (IMO) good enough. A fraction of a percentage point difference really shouldn't be that important in the overall player evaluation of a player.
@durvasa So would you agree (or know for sure) that they determine the coefficient based on running true TS% calculation (with .5 coefficient) for the entire league, by deleting And1 FTA from the numerator and erasing both Tech points and FTA from the formula completely ... then comparing that to a league-wide TS% where all FTA are used in the formula, then adjusting the coefficient in the later result to equal the true one thx, I vaguely remember you posting that. I'm not so sure I'd call Harden's .5 difference insignificant, though. If you look at FG% at ESPN for shooting guards right now...that difference can drop you from 3rd to 7th. It's a rather tight race. They'd never allow that for FG%. Also, for TS% at nba.com (filtering out the >55% FG shooters, ie the Bigs), looks like that difference can drop you from 3rd to 8th...and it's not like everyone drops. Your data shows plenty of people rise, too. Might as well be true and perfect, if the data is there. You know how people love to argue over who is ranked higher than whom. I'll have to check out that site. I would be curious to see if it can scrap out all non-shooting fouls. I found the hack-a-shack bigs drop pretty significantly too, as I recall. I did something last year, based upon finding data of their times going to the line intentionally. I think Harden was actually able to pass Howard due to that. But it probably wasn't completely clean data.
I'm pretty confident they go way beyond anything the scorer's table records. Afterall, even the old school approach did via having your own scouts charting things for you. Then they'd do it in the video room, too. Anything the coach desired. Also, something tells me when Morey says, "Such and such player was an outlier or he jumped out at us" for specific criteria, he was originally using his own formulas from Boxscores (or PBP, too), but has tweaked those to include stuff from videos, not in the PBP. With all the video technology, and Big Data algorithms they can run on video streams, they can enter and query so much more data now, than they ever imagined. I like to explain this by explaining other uses of Big Data video. There is a company in Boulder who installed video security cameras for retail companies. These are stationary, right. They improve their services to customers by not only providing security footage, but they further upsell them with Business Analytics to help them see where their customers are walking, and where they stop, based on time of day, etc, in order to improve product placement decisions. That's all computer algorithms sorting through terabytes of video data. In bball terms, the Sports-VU company can run algorithms too (and customize them to coach parameters or teach the computer to look for a play). So, you can get stuff like, "Show me who runs the High PnR best with Harden." "How open is the right corner three vs left, when Harden goes Right around Capela, vs Left." Also, stuff like "How often is SAS running their Hammer Play." "Who defended it the best." Then the defensive coach pulls that up. And "Show me Curry % on reset threes, where he has to take one dribble to his left, and then again for one dribble to his right." It's mindboggling to think what they can do with the video data being captured by those ceiling cams in stadiums now.
nbawowy, at least, computes TS% as: TS% = (2 * 2fgm + 3 * 3fgm + ft2m + ft3m + and1m) / ( fga + ft2a + ft3a) where: 2fgm: 2-point field goals made 3fgm: 3-point field goals made ft2m: number of FTs made during 2-shot foul trips ft3m: number of FTs made during 3-shot foul trips and1m: number of FTs made for and1s ft2a: number of 2-shot foul trips ft3a: number of 3-shot foul trips You can then compute the coefficient using league-wide totals: LeagueTS% = LeaguePTS / (LeagueFGA + K * LeagueFTA) and solve for K: K = ( LeaguePTS / LeagueTS% - LeagueFGA ) / LeagueFTA Assuming I didn't screw up the formula there, I think it would be determined in that manner. It might be an interesting project to see how close to 0.44 the computed K would be over the last few seasons using the nbawowy data.
Looked up a few players to see what difference it makes: James Harden nbawowy TS% - 62.2 BBall Ref TS% - 61.7 Russell Westbrook nbawowy TS% - 53.7 BBall Ref TS% - 54.1 Steph Curry nbawowy TS% - 63.9 BBall Ref TS% - 63.6 Minutes tied b/w sites except 4 minute difference for Westbrook Clearly, the common TS% calculation was invented by the mainstream media to prop up Westbrook and screw Harden!!! Actually.... .44 must be an alright approximation. Though Harden's lead over Westbrook in TS% increased by 12% when the calculation changed. Don't really understand why the nbawowy TS% would be lower than the classic TS% for Westbrook. Still, I guess they are all close enough that all my concerns were for naught. In theory though, nbawowy is still right to me.
hmmm. That's a very different formula. It doesn't look like base 100% to me. Don't we still have to multiply the denominator by the expected value of the normal FGA, ie 2. Just for the sake of simplifying that formula, if TS% didn't include free throws at all, that formula above is pts / fga vs pts / (2 * fga) -> base 100% In your formula posted, a made 2 pointer scores 200%. And a made 3 scores 300%. Likewise, making both free throws scores 200%, and making all three in a 3fta situation scores 300%. If you multiply the denominator by the expected value of a 2, then it works. What am I missing here? I find that you are usually right, and I'm wrong ... so I must be missing something. Same goes with this. Don't we need to multiply the denominator by 2. LeagueTS% = LeaguePTS / ((LeagueFGA + K * LeagueFTA) *2 ) solve for K K = ( LeaguePTS / (2 * LeagueTS% - LeagueFGA ) / LeagueFTA
You're right. I forgot that. The 0.5 factor doesn't make a difference for analytical purposes. It's merely to scale it down to a form that looks like FG%, which our brains are used to.
^^ agree. One thing that nbawowy formula does remind us is that trips to the ft line on a fouled three ptr need to be treated differently than two point trips. Each free throw needs a .33 multiplier vs .5, when using FTA in the denominator. The nbawowy formula accounts for that by making all trips equal 1. So, the .44 multiplier is probably also accounting for triple free throw trips, as well a single ft trips. If it isn't, then that could explain why Harden scores less, since he's the NBA leader in getting fouled on 3ptrs.
