APBRmetrics

Last season, there were 115 games (RS) decided by 1 or 2 points.
Dividing the league into thirds -- the 10 youngest teams, avg age, and 10 oldest -- we get these rates of success in these 'one possession' games. Age range, W-L (and W%) in the closest games, RS Win% and PythW%

Older teams tend to be better; to win more than their MOV would predict; and to win more than their share of the closest games.

In the last 5 years, team age correlates more closely with actual Win% (.57) than to PythWin% (.51)

One could imagine a PythWin% formula that accounts for team age; and in fact I have found a closer fit between W% and PW% with an age factor.
The traditional PW% = tmPts^14/(tmPts^14 + oppPts^14)
Since 2011, avg 'error' between W% and PW% is .030

The alternative formula is:
P2W% = tmPts^13.37/(tmPts^13.37 + oppPts^13.37) + (tmAge-26.55)/160
The avg error here is .028

An improvement from .030 to .028 may not seem like much. But one win is .012 of 82 games. Without fractional wins happening, even a 'perfect' fit is off by an avg of .006 per team.
In the 150 team-season sample, 12 teams' errors are made worse by at least .010, with the alternate p2W%.
Meanwhile, 28 errors are reduced by at least .010

With a given MOV, a team that's 2 years older tends to win one more game (in 82).

Since I have them, here are some correlations with team age (2011-2015) p2W% is the pythagorean expected W% with age factor (1/160 of age-26.55) and with the exponent at 13.37
Younger teams tend to win by faster pace and O-Reb.

[see the opening post]
While it seems extraordinary that a 58% win team should win 54% of 1- and 2-point games, I confess I don't know what success rate should be expected.
Since better teams seem to be older than weaker teams, year in and year out, we may be seeing (partly or mostly) that better teams win more often at every level of point spread.

Going back to 1979, I find an NBA season in which there's zero correlation between team age and win %.
With just 22 teams in the league, still there are 143 1 or 2 point decisions in the season.
Again dividing the league into thirds (roughly) with 7 oldest, 7 youngest, and 8 middle-age teams: Note the final column -- RS pythW% -- indicating the 7 youngest teams were presumably the best of these groups. Yet they floundered in (93) one-possession games.
These closest games probably account for most of the "error" between W% and pythW%

The distribution is very similar when dividing the league into halves or fourths.

EDIT: For that youngest-team 1/3 of the 1979 NBA, their drop from 50% (or 52%) in all games, to 42% in the closest games, is actually understated.
If 7/22 of closest games are between two young teams, that's about 22 games going 22-22.
Their remaining record would be 17-32, or just 35% wins vs older teams.

Older teams tend to be better than younger teams.

In a past post, I found that better teams play better in the clutch than worse teams.

http://www.apbr.org/metrics/viewtopic.p ... 75&p=19593

I looked at both of these topics in the same column a few years ago:
http://basketballprospectus.com/article ... icleid=573

Eyeballing that chart, you'd expect a team that won about 60% of its non-close games to win around 53% of its close games, so that's not unusual.

I didn't find any correlation between team age and record in close games when I looked at it there, so my first thought was that this was a more recent effect. And it has been stronger lately, though as you note it exists over a longer period. Separating into five-season chunks:
So now I'm wondering whether the correlation might be because of causation the opposite direction: teams that win more close games are likely to play their veterans because they're either in the playoffs or in contention for it, whereas teams that don't get those breaks slip out of the playoffs and go young.

One reason I looked at 1- and 2-point decisions is that about half the time, it is literally one possession at the end of the game which determines the winner. Hit the game winner, or prevent it.
A team which wins 51% of its possessions will win about 70% of its games; so yeah, it's pretty good for a 60% team to win 53 or 54% of the most critical ones.

There are parallel explanations: Rather than experienced players actually playing better basketball, refs may favor them. There's something "better" about working the refs, I suppose. Certainly it's better than antagonizing the officials, or putting them on the spot.

When a team is formed, or modified, with players who have 'known' plus-minus values, we can theoretically add them up (weighed by minutes) to predict a team MOV. But when these players are older, they may be expected to add more wins than that MOV stipulates.

The 2007 Celtics were the youngest team in the league, at 23.5 yrs avg. Their PythWin% indicates they should have won 31 games, but they won 24. Next year, they were 27.9 yrs old and won their expected 66.
Gaining 4.4 years in avg age on the court, one could add 2.2 wins to their predicted wins based on player RAPM or whatever.

For the 1978-79 season, correlation between team age and MOV (and PythW%) is .002; for actual Win% it's .156
Given the consistency over (thanks, Kevin) several decades -- W% > pW% -- it seems the evidence is consistent that older teams win more than their MOV says they should. Even when they are not particularly the better team.