Announcement

Collapse
No announcement yet.

Kyle Lowry and the Raps' come from behind offence

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #16
    Call this index BALLS (Behind/Ahead Logarithmic Label Score).

    Kyle is leading the league in BALLS.

    Comment


    • #17
      BTW, correct me if I'm wrong but wouldn't you have to factor in the fact that players within the same team don't spend the same % of ahead/behind time on the floor? Like for example, Bruno took 100% of his career shots while leading.

      Maybe this tends to be a non-factor for starters or heavy minutes players, but maybe not...

      Comment


      • #18
        Matador defense doesn't win championships

        Comment


        • #19
          This is a great analysis.

          I find Lou Williams numbers to be striking. He is heavily involved in wins, but not trusted when the team is behind? Do minutes distributions in these situations line up with the %TSA? Could be that the role of bench scorer keeps him out of the line-up in tight games.
          Welp, that sucked.

          Comment


          • #20
            c_bcm wrote: View Post
            This is a great analysis.

            I find Lou Williams numbers to be striking. He is heavily involved in wins, but not trusted when the team is behind? Do minutes distributions in these situations line up with the %TSA? Could be that the role of bench scorer keeps him out of the line-up in tight games.
            Unfortunately, I don't have a way to correct of the minutes these guys are playing while ahead/ behind. Instead, I'm assuming that if the team's split for TSA falls one way, that should roughly match the individual That's the biggest weakness of the stat so far.

            Lou's numbers stood out for me as well. The only explanation I really have is he's a bench player who shoots a lot, but happens to have come in to protect a lead.

            / I know no one has asked this yet, but I'm using True Shot Attempts and not USG rate for a couple of reasons. First, I think TSA more accurately captures what I'm looking to show. USG includes Assists and Turnovers. Including those would defeat my purpose. Obviously Assists come when a player looks passes, so if the stat included them that would be including passing in a number that's trying to show a player's tendency to call his own number while behind. The reason I didn't include turnovers is because turnovers are heavily correlated with passing. And second, Basketball Reference doesn't offer ahead/behind numbers for minutes played. I'd need those to calculate USG. So even if I wanted, I can't.

            // Gonna try and do a big update with a bunch more players soon
            @EdTubb - edwardtubb at gmail

            Comment


            • #21
              I like the breakdown, but also disagree with the conclusion. I don't see it as a trust issue, I expect it's how Lowry perceives his own role - in fact, I'd suggest it's how Casey defines both Lowry's and DD's roles. When the game is on the line or a big bucket is needed, those three are all on the same page re: what should happen: DD or Lowry should shoot the ball.
              "We're playing in a building." -- Kawhi Leonard

              Comment


              • #22
                Ok. I did another run at the numbers. Turns out by this metric, Kyle is in the top tier of guys who look to their own offense when behind, but not at the top.

                @EdTubb - edwardtubb at gmail

                Comment


                • #23
                  First of all, great general idea to analyze.

                  I tend to disagree with the 'lack of trust' element, because I think you'd need to show that a player is passing up higher-percentage opportunities to take their own.

                  But that aside, this is all very cool. What's interesting to me is how much traditional big men (Favors, Lopez, Cousins, Randolph, Jefferson) tend to be on the bottom of the list. M. Gasol is really the only post-up heavy guy on the top of the list. The other bigs on the top of the list tend to be more floor-spacing guys. For big men, they don't really have far less opportunity to choose when to be aggressive, since their PG tends to dictate when they get touches. And it seems, based just on this list, that teams may go away from post-up plays when trailing.

                  This could also be a matter of teams that are trailing trying to push the tempo, and thus generating more offence through guards and wings.

                  Comment


                  • #24
                    Love the avatar TeamEd.

                    And love the thought and work put into this.

                    Comment


                    • #25
                      So, I'm going to muse a bit here on a few points, just so I can get it down in writing. Feel free to chime in on anything.

                      - What I think I've got here is a stat that shows a player's situational shooting mentality (that's what I'm going to call it from now on). If the Raps are down in the game, this SSM shows how much Kyle Lowry increases his shooting rate relative to when he plays with the Raps ahead.

                      - I don't know how to include turnovers in the metric. Turnovers are highly correlated with assists. But, I probably need to find a piece that tells me the ratio of how turnovers typically breakdown. That way, I might include an estimate of how many of the player's turnovers apply to each metric. Say, 60 per cent of turnovers from passing, 30% from dribbles, 3% on-ball offensive fouls and 7% off off-ball offensive fouls. Then, I would calculate the SSM using True Shooting Attempts + Turnovers * (0.30+0.03).

                      - In order to make the player's breakdown of shooting attempts relevant, I need to establish how much time he spend on the floor either behind or ahead. Right now, I can't find a breakdown of minutes played in these situations. That's the biggest weakness of this idea. I'm doing it instead by comparing his numbers to the true shooting attempts of the team as a whole. This gives a pretty good sense of how often the team is ahead or behind and I'm using this as a rate to tell us how often we can expect a player would shoot in these situations.

                      But, there are a couple of problems with this: 1. Not all players on a team will have situational splits that match the team's. This is easy to see with bench players. If a teams' starting lineup typically gets ahead early, then the bench lineup will play with a lead relatively more frequently. (I think that largely explains Lou's odd split in my first post) 2. Not all possessions end in true shooting attempts. Some end in turnovers. And, you can imagine that turnovers might also have situational splits.

                      - I'm not sure how to deal with 1. Unless I can find a minutes played breakdown by score, it's a non-starter. On 2., I think the solution is to count turnovers on the team's end, even if I don't count them on the player end. The team calculation is just to figure out what ratio of team possessions occur ahead or behind. Adding turnovers gives me a truer ratio on that end. But, that doesn't mean I necessarily need to include turnovers for the player. I'm not looking at his total situational possessions, just his shooting possessions relative to what we expect.

                      - Am I wrong, or is looking at the teams' ratio of possessions ahead or behind actually better than looking at the minutes breakdown? Surely pace is affected by score. If so, what better measure of the team's situational offence than possessions?

                      - I can do the same with assists using the same metric. That would be a "situational passing mentality" and would achieve another interesting result.

                      - For Kyle, he's recorded 75 assists while ahead and 54 while behind or tied. That means 41,86% of his assists happen while behind. The team as a whole takes 45.24% of it's attempts in that situation. This tells us that his "SPM" is -3.38. He is 3.38 percentage points less likely to record an assist while behind or tied.

                      - I could then combine the two into some kind of "situational attacking mentality". That could be a simple SSM - SPM, or something. Kyle's SAM would be 9.05 (his score on that chart below) - -3.38 or 12.43. To translate that for a second, Kyle's increase in attempts appear not to result from an increase in possessions. He changes his game greatly.

                      - On a nerdy point. Right now, the number I'm using as the final SSM is the percentage point difference between the percentage of attempts the player takes while trailing or tied and the team's. There are other ways I could count this. Let me explain by comparing Kyle, Kyrie and Kobe.

                      So far, roughly 45% of the Raps' attempts have come trailing or tied, that's 48 per for the Cavs and 70% for the Lakers (who have been truly terrible).

                      On the players' end, 53% of Kyle's attempts have come trailing or tied, 55% of Kyrie's and 79% of Kobe's.

                      The way I have been counting up until now simply subtracts the player's number from the team's. For Kyle, we expect 45% of his attempts would happen while behind or tied when actually he takes 53% of his attempts in that situation. With these numbers, that's a difference of 8 percentage points. The number is 8 for Kyrie and 9 for Kobe.

                      Doing it this way gives the impression that Kobe changes his game more than Kyrie and Kyle.

                      But, I'm not sure this is the best way to calculate the final number.

                      Another way is to calculate the percentage change between the player's expected rate of shooting while behind and tied (the team's) and his actual number. For Kyle, that would mean calculating that his 53% attempt rate is 17.5% larger than the expected 45% rate [(45-53)/53*-100]. For Kyrie, that number is 16.5% and for Kobe that number is 12.9%.

                      Doing it this way tells you that Kyle actually changes his game more than Kyrie, and much more than Kobe.

                      The difference here comes from the variance in the base expected rate. The Lakers already take so many more shots from behind that Kobe's increase is less meaningful using this method. Kyle's 8 percentage point increase is more meaningful than Kobe's 9 percentage point increase because the Raptors' spent so much less time trailing than the Lakers.

                      I should note for second here that it's very possible everything I just said here is wrong. Math is hard. Please tell me if what I just doesn't make sense.
                      @EdTubb - edwardtubb at gmail

                      Comment


                      • #26
                        So. I've been thinking - and yes, I'm using this thread as a diary for this - do the team's splits actually matter when interpreting the player's splits look like? I ask because I've found player minutes played by scoring margin splits on NBA.com.

                        So, my choice is to simply use player minutes as the baseline or to continue the way I've been doing it and compare player attempts to team attempts.

                        The way I've been doing it so far, if a team splits its true shooting attempts 60/40 that's the standard I use to set the expected shooting rates of the player. I see a few problems with still doing it this way. The biggest is that this compares the player's rates against a standard that includes his own contribution to team attempts. The team as a whole might shoot more than expect while trailing, but if the player is contributing to this trend it becomes a double counting problem.

                        The other way is to simple assume that a player should be equally likely to take a shot in any given minute regardless of score. That way, if the player splits his minutes 60/40 that becomes the ratio I expect to see in his shooting attempts. The problem with this is that it looks like there's an NBA wide trend in teams shooting more and scoring better while trailing. It seems like on average, every NBA player is more likely to shoot while trailing or tied.

                        The big question for the purposes of this is, does that trend even matter to what I'm trying to analyse? Do I need to correct for this at all? I'm leaning no on that question. Which means I'm pretty confident that I the best number I can come up with uses just the player's own stats with no reference to his teams at all.

                        And, this makes this whole project a lot easier. I can now look at player's situational shooting without first calculating his team's. It doesn't matter now if a player has been injured for long periods of time. It now becomes relatively easy to for me to pull stats for a bunch of players at once. And, for the first time now I can actually write down a straightforward formula to come up with what I'm going to call my Situational Shooting Mentality metric.

                        That formula is this: (MP<= ÷ MP - TSA<= ÷ TSA) ÷ (MP<= ÷ MP) x -100.

                        It's the same for a Situational Passing Mentality metric. There the formula becomes: (MP<= ÷ MP - AST<= ÷ AST) ÷ (MP<= ÷ MP) x -100.

                        From these formulas, I see that Kyle takes 54.9 % of his shots behind or tied, but only plays 49.1% of his minutes in that situation. Using the formula above, that makes his SSM 12.0. ie: He is 12% more likely to look for his oen shot while trailing or tied against an equal distribution. Meanwhile he records 45.6% of his assists when trailing or tied. Still comparing to his minutes played, that puts his SPM at -6.9. ie: he is 6.9% less likely to record an assist in the same situation.

                        There. Simple.
                        @EdTubb - edwardtubb at gmail

                        Comment


                        • #27
                          OK. *taps mic* Is this thing on?

                          I mentioned something in the last post I want to explain a little here. Before I do, I'll recap up to this point.

                          - I've created a metric using NBA.com's ahead/behind splits that shows how much a player's shooting attempts increase against expectations when the team is trailing or tied.
                          - I'm calling this number a player's "situational shooting mentality."
                          - The formula is (MP<= ÷ MP - TSA<= ÷ TSA) ÷ (MP<= ÷ MP) x -100, where MP is minutes played and TSA is true shooting attempts.
                          - A higher SSM indicates that a player is more likely than expected to register a true shooting attempt when their team is behind or tied. A lower number means the opposite.
                          - The goal is to create a number that compares players' by their tendencies to seek their own offence in comeback situations. That way, when we say "Kyle Lowry puts the team on his back," I have a number to say exactly how he does that.

                          Back to the problem I noticed in the last post. Right now, I have a elegant little formula anyone can use. But, the numbers it's producing are skewed. Why? Well, as it turns out the entire NBA has a positive SSM.

                          Since my formula works just as well for teams as it does for individuals, you can see this if I simply chart the shooting mentality of all of the teams.



                          As you can see. There's a problem here. It turns out basketball teams take more attempts when trailing than you would expect from looking at their minutes played in those situations. Weird.

                          I wish I had a good explanation for this, but I don't.

                          Nevertheless, this tendency affects the usefulness of my stat. Currently, I'm giving a team like Cleveland a positive SSM. This says the Cavs shoot more often than you would expect when trailing. But, against the average that conclusion is simply wrong. They don't shoot more than you'd expect, they shoot much less.

                          The only solution I have is to fix this is to adjust for the league average. I do that by subtracting the league average SSM from the player/team's SSM. The final number certainly makes for a more useful chart:



                          The problem with doing it this way is a matter of labour: If I adjust against the league average, that means every time I have to pull the stats to compare a set of players I also have to pull the numbers on each team in order to calculate the league average SSM. That adds almost a half hour to every run. What a bummer.

                          Anyway. For the sake of completion, here is a current run of the SSMs for a set of the top 30 NBA players by FGAs, as well as a handful of starting point guards who didn't otherwise qualify:



                          Next. Here is a current run of situational passing mentality (SPM). Remember, my formula works just as well with Assists in the place of True Shooting Attempts, with the caveat of a smaller sample size this early in the season. With that in mind, here's a chart of those players in the above list who have recorded at least 80 assists so far this season. This chart is corrected against the league average SPM, which currently is is 3.38. (J. Smith is Josh not J.R., by the way.)



                          Finally, here's a run of the SSMs of the Raps' rotation players. I'll leave it up to you to interpret these numbers, with the warning that I'm pretty sure there are some sample size issues going on here.

                          @EdTubb - edwardtubb at gmail

                          Comment


                          • #28
                            Really? Lou would pass up as opposed to shoot?

                            Comment


                            • #29
                              Zainab wrote: View Post
                              Really? Lou would pass up as opposed to shoot?
                              Not exactly. This is saying he shoots less often when trailing than you'd expect if you distributed his shots evenly. That could also mean he's taken a ton of shots with a lead. My guess is that this is a sample size issue coming from the fact that his insane shooting games have come with him piling on an already big Raps lead. Most or maybe even all of his 20+ shots from those Bucks and Cavs games came with the team protecting a lead. I don't see that number staying like that for long.
                              @EdTubb - edwardtubb at gmail

                              Comment


                              • #30
                                TeamEd wrote: View Post
                                OK. *taps mic* Is this thing on?
                                This is going to take a while for my simple brain to absorb.

                                Comment

                                Working...
                                X