Blatsheet

Old title: Where careers go to die

Addendum

No team's new arrivals play worse than chance allows

Updated : new title; addendum, no clear team; the table opens on how they played. long-formnba

A player joins a new team. Against his age and new teammates, over 2,920 moves since 1996-97, Philadelphia's arrivals play worst, but within chance. Playoff rounds in the Move score measure the new team's quality.

This was a long bow to begin with. The first and second passes made it clear there were a few teams in the running. Since being set straight by some statisticians, I can see it was not statistically significant and relied on chance. Now it is only Philadelphia, and even that is not enough to make a story.

  • Counting every season equally, Philadelphia stays lowest (p = 0.68).
  • Across 10 minutes-floor settings, a single team passes once.
  • With a higher floor before the move, teams differ in 2 of 8 settings.
  • Movers score 7 points below stayers.
  • Split by position, role or arrival, one test in 11 passes.

How we set this up

From 1996-97: 30 seasons.

  1. Every player since 1996-97

    • 2,859 players, 14,569 player-seasons. Each dot is one of them in one season.
    • The number above each column is how many played. Fewer played early on, with 29 teams until 2004-05 and no two-way contracts until 2017-18.
  2. Rookies out

    • Rookies have no team before to compare with. 2,443 rookie seasons go, leaving 12,126. 9 rookies traded in their first season stay, because they had a team before.
  3. Seasons he stayed put

    • We want the seasons he arrived at a new team. 7,176 seasons where he stayed on the same team go, leaving 4,950 where he joined one.
  4. Under 500 minutes

    • He needs 500 minutes in his two seasons before and his first two there. 2,145 seasons go. The 2,805 left hold 2,920 moves by 1,100 players.
    • Up is how well he played; across is the playoff rounds his team won.
  5. The teams by how they played

    • Each row is a team, lowest first on how its arrivals played. The dot is their average, the grey bar its 90% range, the ring the Move score.
    • No dot passes the test across 30 teams.

TL;DR No single team's new arrivals play worse than chance allows. Teams do not differ as a group either (p = 0.127), as the Harvard Sports Analysis Collective found in 2015. Philadelphia's are lowest, at 41 (range 35–46), with p = 0.135 across 30 teams. Gaps in the full Move score are playoff rounds, which follow the new team's quality.

Counting those rounds, the six lowest are Charlotte (42), Chicago (43), New Orleans (44), Orlando (42), Philadelphia (41) and Sacramento (43). That score follows team results. It tracks winning percentage at r = 0.68; how they played tracks it at r = 0.35.

Where careers ended

Of 4,390 moves, 1,062 (24%) were to a player's last team, 648 for under 500 minutes. Against age, no team ends more careers than chance allows (p = 0.56). Fading, under 500 minutes in each of the next two seasons, differs between teams in 4 of 6 settings, by 2 to 4 points. One team's arrivals fade less often in 2 of 6, about what 24 tries give.

Teams ranked by how their arrivals played

We rank each of the 2,920 moves since 1996-97, by 1,100 players, active players included, from 0 to 100, so 50 is typical. The table opens on how he played, against what his age and new teammates predicted. A team's score is the average of its arrivals, recent moves counting more. The filters add playoff rounds, for the full Move score.

Based on
What counts
Arrived
Recent seasons count more

A move 10 seasons ago counts half as much as one last season, 20 seasons ago a quarter.

Teams ranked by how their arrivals played, 1996-97 to 2025-26. Recent moves count more. Each move needs 500 minutes either side.
How they played, 50 typical
PHI +$4.7M44−$6.9M39
DEN +$5.4M49−$4.6M30
DAL +$6.8M48−$11.3M50
SAC +$4.9M55−$7.1M33
ORL +$5.5M60−$11.7M32
NOP +$5M61−$4.9M29
TOR +$5.6M59−$4.6M30
MIL +$3.7M56−$7.3M43
CHI +$4.9M55−$5.5M39
LAL +$4.6M43−$12.8M63
CLE +$5.3M63−$6.3M33
CHA +$4.5M43−$6.6M39
ATL +$4.5M45−$7M41
POR +$6.8M48−$7.4M29
MIN +$5.1M54−$7.5M32
BKN +$5M59−$9.2M52
MEM +$3.8M52−$7.1M26
NYK +$6.3M61−$9.6M43
WAS +$5.9M48−$7M42
LAC +$3.6M41−$9.5M59
GSW +$4.9M39−$8.2M33
PHX +$4.8M58−$7.8M44
SAS +$3.4M33−$10.8M25
HOU +$8M37−$9.8M44
BOS +$5.9M41−$10.5M31
UTA +$4.7M36−$8.1M32
OKC +$4.5M36−$6.9M24
IND +$6.2M44−$6.4M27
MIA +$4.3M41−$17M48
DET +$5.7M52−$9.1M45
How this was worked out

A move is a player's first stint with a franchise after playing for another. He needs 500 minutes for other teams in the two seasons before, and 500 for the new team in its first two. A mid-season trade counts the rest of that season. A team's arrivals are the players who moved to it.

Relocated franchises are one team. Seattle is Oklahoma City, New Jersey is Brooklyn and Vancouver is Memphis. The 1997-2002 Hornets are Charlotte, as the NBA records them, so players who moved with them to New Orleans did not change teams.

The parts are each the change from before to there, minutes weighted. By default they are Box Plus/Minus (35), on-off per 100 possessions (15) and playoff series won a season (30). Minutes a game (25), usage (10) and pay (15) can be added. We rescale the weights to 100% across the parts switched on, so the defaults are 44%, 19% and 37%.

Pay is salary as a share of that season's cap. The table shows a change in pay as that share of this season's cap, in dollars. We count pay where we found a salary (2,663 of 2,920 moves).

What playoffs add: how he played is 63% of the default score, playoff rounds the rest. Everyone who joined a team in one season shares its rounds. So that part moves with team results, and its 90% range is too narrow. The default score tracks winning percentage at r = 0.68; how they played tracks it at r = 0.35.

On how they played, one team is wholly below 50 and three wholly above. On the default reading, the counts are six and six. With 30 teams and 90% ranges, chance alone puts about 1.5 at each end.

We take what we expected off each change. First comes his age, the average change from one season to the next, by age, for players with 500 minutes in both. A 22-year-old usually gains 0.58 Box Plus/Minus a year; a 33-year-old loses 0.53.

Then come his level before and his new teammates, which we fit over every move, for Box Plus/Minus, on-off, minutes and usage. For each point of Box Plus/Minus above zero, a player usually gives back 0.30. For each point his new teammates are better, his on-off usually falls 0.41.

We do not adjust playoff rounds for teammates. That part measures how good the new team was, so the TL;DR rests on how they played alone. We divide each change by how widely it varies across all moves, then weight and add them.

Move score is the move's rank among all moves, 0 to 100. A team's score is its arrivals' average, a move counting half as much for every 10 seasons back. The range is that average plus or minus 1.645 × 28.9 ÷ √moves. 28.9 is the standard deviation of one move's rank, and moves is the effective number.

Allowing for 30 teams takes the lowest team's two-sided p on how they played and multiplies it by 30. At 500 minutes this matches the Harvard Sports Analysis Collective (2015). It ran the same kind of test on Box Plus/Minus by team and found no team effect.

The group test asks whether teams differ at all. It uses Cochran's Q over the 30 team averages. Its p comes from 5,000 seeded shuffles of whole team-seasons within each season, because arrivals to one team in one season share its luck. The team share is Q less the shuffles' average Q, scaled as DerSimonian and Laird do, over one move's variance.

After leaving scores his first two seasons after the team against his seasons there, ranked against every player who left a team. Era scores need 8 arrivals.

Out injured comes from his game logs. Inactive, Did Not Dress, Not With Team and suspensions count as out. Did Not Play, the coach's choice, counts as available. A season counts against him when he played under half the team's games and was out more often than benched. That includes a season on the roster without a game.

Careers that ended: 160 players arrived aged 31 or younger at the team where their career ended. 28 of them dropped on minutes, usage, Box Plus/Minus and team. The other 132 did not fall on all four. Joining a worse team usually raises a player's usage (Nylon Calculus, 2017), so a fall in both is rare by design. With so few, the team counts are luck.

414 retired players ended their careers at a team they had moved to, 188 of them aged 32 or older on arrival. Players peak at about 27 (Pelton; Vaci and others put it at 28 or 29). We cut arrivals aged 32 or older, the age Dave Berri gives for a nosedive, quoted by Tom Sunnergren in Bleacher Report.

We also leave out 8 who won a title there. Miami won two titles with Chris Bosh before blood clots ended his career. We leave out 51 with a season out injured, as defined above. Danny Granger had played five games two seasons before Miami. We leave out 7 who went home to their first team, which leaves the 160.

Lifting the limits. A minutes floor is the fewest minutes a move needs. With 750 or more on both sides, teams pass the group test in 8 of 8 settings. Minutes at the new team depend on how the move went. With the floor on seasons before only, 2 of 8 pass, and team explains at most 1.1% of the spread. We tried higher floors after 500 found nothing.

The single-team pass is Philadelphia at 750 minutes (p = 0.046), about what ten tries give. Part of the movers' gap is who gets moved: teams keep players who are going well. EPM's past seasons are for subscribers only, so we kept on-off.

Splits and died rules. We split arrivals by position, role before the move (usage and minutes thirds) and how he arrived. One of 11 tests passes (p = 0.043), about what 11 tries give. Within a split, a team's gap would need 16 to 31 points to show. Traded-in and signed players differ by 0.5 points (p = 0.66). Looser died rules stay within chance (p ≥ 0.44).

Front office. We credit each arrival to the executive in charge at its game, offseason deals at game 1. That gives 168 executive spells, a median of 12 arrivals each. We measure each arrival against its franchise's average, then slide spell boundaries along that franchise's arrivals. The joint test has p = 0.36, and no single spell passes. A spell would need 26 points to show.

Windows, pay and coaches. Scoring only the first season, Philadelphia stays lowest (p = 0.080). We tried three-season windows, pay before and on arrival, and each coach's guards, wings and bigs. One of 9 team tests passes, about what 9 tries give. Pay on arrival uses offseason arrivals only, as in-season pay tracks how long a player stays. Counting form, top-third earners beat bottom-third ones by 0.3 to 2.8 points, within chance.

Where careers ended counts a career as ended at the team when he never played for another. The floor is 500 minutes on his two seasons before only, and we leave out players still with the team.

Fading means under 500 minutes in each of the two seasons after he left, against his age alone or his age and form. The group test passes in 4 of 6 settings, from p = 0.004, on a base of 46 to 49%. The single-team test passes in 2 settings, each time for a team whose arrivals fade less often.

Regular seasons only. Stats, game logs, plus-minus, salaries, contracts and the salary cap come from Basketball Reference. Team records come from NBA.com. Built 10 Nov 2026. The whole ranking is a table at NBA Insights.