Hello, board. I'm not yet sure who the resident stat geeks are. But I'm sure there're a few of you. I've been starting to get into advanced analytics. And I had a few questions. #1) Perhaps the main question I have is this: What is the difference between PER and Win Shares/48. Based on the general descriptions of the two stats, it sounds like it should be just about the same thing. But, no. ...Win Shares/48 (where .100 is league average): 1) Mitchell is first on the team, with a whopping .205. Then 2) Drummond (.184), 3) Monroe (.115), 4) Harrellson (.114). Then as we move into below league average, 5) Singler (.093), 6) Jerebko (.086), 7) Jennings (.084), 8) Villanueva (.059), 9) KCP (.058), 10) Bynum (.056), 11) Stuckey (.055), 12) Smith (.028), 13) Billups (-.036), 14) Datome (-.038), 15) Siva (-.168). ...But the PER ratings are way different: 1) Drummond (22.2), 2) Mitchell (20.9), 3) Monroe (18.1), 4) Jennings (16.3), 5) Villanueva (14.8), 6) Bynum (14.6), 7) Smith (14.3) 8) Stuckey (13.6), 9) Harrellson (13.4), 10) Jerebko (13.4), 11) Singler (11.8), 12) KCP (9.2), 13) Datome (7.3), 14) Billups (5.3), 15) Siva (-2.2). What gives?
The PER sums up all a player's positive accomplishments, subtracts the negative accomplishments, and returns a per-minute rating of a player's performance. All calculations begin with what is called the unadjusted PER (uPER). The formula is: uPER = (1 / MP) * [ 3P + (2/3) * AST + (2 - factor * (team_AST / team_FG)) * FG + (FT *0.5 * (1 + (1 - (team_AST / team_FG)) + (2/3) * (team_AST / team_FG))) - VOP * TOV - VOP * DRB% * (FGA - FG) - VOP * 0.44 * (0.44 + (0.56 * DRB%)) * (FTA - FT) + VOP * (1 - DRB%) * (TRB - ORB) + VOP * DRB% * ORB + VOP * STL + VOP * DRB% * BLK - PF * ((lg_FT / lg_PF) - 0.44 * (lg_FTA / lg_PF) * VOP) ] Most of the terms in the formula above should be clear, but here are a few of the less obvious ones: factor = (2 / 3) - (0.5 * (lg_AST / lg_FG)) / (2 * (lg_FG / lg_FT)) VOP = lg_PTS / (lg_FGA - lg_ORB + lg_TOV + 0.44 * lg_FTA) DRB% = (lg_TRB - lg_ORB) / lg_TRB The terms that start with "lg" are for league stats. Terms that start with "tm" are for team stats. This helps normalize the data so that every year the average PER for a player is 15.0. This allows you to make comparisons between players of different eras or track performance as the NBA makes rule changes. The top 5 career PER leaders are: 1. Jordan 27.91 2. LeBron 27.65 3. Shaq 26.43 4. D. Robinson 26.18 5. W. Chamberlain 26.13
Right. I read up on all the stats and I get them. I just don't understand why PER is so much different than WS/48? Not the actual numbers-I know they're on different scales. But how the rankings of one against the other are so different? It seems like they're attempting to accomplish the exact same goal--a single, all-inclusive ranking statistic that isn't a counting statistic--but coming up with really different results. FWIW, from what I've read, the stat-junkies seem to be entirely lined up against Hollinger in the WS/48 vs. PER battle.
I actually read that article when it was first published Slippy. Just read it again, more closely. It's exciting stuff for sure man. The stats available online in a decade will be really exciting (though I wish I could get my hands on what those guys are calculating now!).
I have used PER for quite a few different projects, so I shared that. I am not as familiar with the Win Shares. Perhaps someone else could explain them.
As unpopular as it might be, the advanced stats almost always shows that James is on par or just below Jordan. Stats don't care about narrative.
I don't think anyone has ever argued that PER is a more complete statistic than WS/48, including Hollinger. However, it is important to note that both suffer from small sample sizes. I personally like PER because it something like a normal distribution, but recognise that WS is likely a better stat as it combines OWS and DWS. Both rely on arbitrary weighting systems to converge points, rebounds, steals etc.
PER certainly does, but it is normalised each year, so long term comparisons from different eras is a serious misuse of the statistic
I don't value PER or win shares but I really think that the other stats are really interesting. Like how likely is a guy to score off the bounce vs. the catch and shoot? In what areas? For example, Joe could have used advanced stats to see that KCP shot in rhythm and shot a lot and so what do you expect if he just camps at the 3 and launches standing still? Cheeks could be like: hey, this guy can curl around a screen or run a staggered baseline and catch and shoot...OR we could have Jennings dribble the air out of the ball, then kick it to KCP because he ran out of options.
I agree. I don't think I'll ever know what a win share is. That situational stuff is uber useful though which probably means the Pistons are at least 5 years from using it.