Changing Teams: Part 2
Earlier in the offseason, I looked at how individual shot selection changes when a player switches teams midseason. Most players who moved during 2014-15 saw an improvement in their shot selection, especially when going to a better team. No one saw any sort of outrageous swings in shot selection, however.
But why focus only on shooting, what about other stats?. Do these other indicators stay constant after switching teams, or are there big changes? Given players’ production is to a large degree a product of their environment, it would be natural to expect there to be some sort of change.
As an example, let’s examine Reggie Jackson. After moving to Detroit, he saw an increased role[1. 8.2 minutes of possessions time per game in Detroit vs. 4.7 TOP in OKC, 17.2 assist chacnes per game in Detroit vs. 9.3 in OKC (both via SportVU) and 2.1 Bad Pass TOVs per 36 mins in Detroit vs. 1.3 in OKC (via play-by-play data.]. This exapansion of opportunities led to the largest increase in AST% for a single player switching teams since the NBA/ABA merger[2. minimum 200 minutes played on both the first team and second team]. By definition, this makes Jackson an unusual example.
To know just how unusual, we need to understand how much these sorts of stats[3. Even rate-based stats.] tend to hold constant between stints on different teams within a season. To that end, I examined correlations between the first and second teams across a variety of statistics.[3. I looked at all player seasons where the player switched teams within the season since the merger. This analysis does not include any third, fourth or fifth teams that a player played for within the season. It is worth noting that the players who played for 3+ teams within a season are generally very poor players. Players who appeared for 3 or more teams had a weighted BPM of -4.2 as compared to -3.3 for players who only played on two teams. Even requiring a minimum of 500 total minutes played on the season, players who played for 3+ teams had a minutes weighted BPM of -2.6 versus -1.9 for players who only played for 2 teams.]
Data from nyloncalculus.silk.co
Most of the individual stat correlations[4. Comparing the stat of the 1st team to 2nd team.] are pretty strong[5. Looking at either a minimum of 200 minutes or 500 minutes.] with the exception of TS%, which is probably an indication of shooting percentages being highly volatile. We know that three-point percentage can take years to stabilize and so it’s not surprising that TS% is relatively weak correlated between player-teams. We could split TS% out into eFG% but we’ll save this for another post when we’ll look at the correlation of actual shooting percentages when switching teams.
Looking at some other individual correlations, two of the more surprising are Usage% and Assist%. One might expect these statistics to be more team dependent[6. More on this later.] and yet there’s a strong correlation – that is, they tend to hold steady when moving from what team to another? Why is this the case? Possibly, asking a player to drastically change roles midseason is largely unreasonable. In that case, even if a player were traded to a worse team, they aren’t asked to do[8. or able to do?] too much more than they had before they were traded.
To put a point on things, how well players do most things seems to stay largely constant even when changing teams. However, how much of each skill the player is asked to employ might change a great deal. This can be seen through an examination of some all-in-one metrics, which roll all of those individual contributions together and estimate the value produced.
Data from nyloncalculus.silk.co
On these measures, the correlations were generally smaller than for the individual skill metrics. There remain some decent degree of correlation[5. It it still the same player, after all.], with the exception of Win Shares. Given how team dependant the formula for Win Shares statistic is,[6. And in a following post, I’ll expand on this.] this isn’t really surprising. The strong correlation between Defensive BPM is notable, though perhaps explainable, as DBPM uses box score stat to measure individual defense. The correlation between both Steal% and Block% from old to new teams as shown above likely greatly influences that relative constancy. However, there is a team adjustment in DBPM and unlike Defensive Win Shares, this team adjustment doesn’t appear to be as strongly correlated with the strength of the teams involved. Still, the generally much weaker correlations in the one-number metrics might indicate the difficulty in predicting how much a player’s role will change, and how much of an effect that change will have on his perceived value.