I'm just trying to get my head around the predictive validity of the two methods at various points in the season. So let's say there are 5 games to go, and you have a .500 team that has won many games by huge scores, but lost most of the close ones. Why would net rating be a better predictor of the team's record for the season at that point? Wouldn't that team tend to exhibit the same behaviors (i.e. win by alot, lose by a little) going forward? Now go back to that same team with 10 games to go, or 20 or 30 or 60 and ask the same question. I'm not implying that I know the answer, but if I were a gambler I'd love to know it.
I'll try the following hypothetical:
Team A is 0.500. All of their wins have been by 15 points or more, while all of their losses have been by 2 points.
Team B is 0.500. All of their losses have been by 15 points or more, while all of their wins have been by 2 points.
Both teams have played 60 games, all against the same opponents. So Team A's differential would be +390, while Team B's would be -390.
Team A and Team B are scheduled to play each other in a neutral site tomorrow night. Which team do you think would be most likely to win that game?
Similarly, looking ahead to the remaining 20 games, which team do you think would perform better, assuming similar schedules, injury luck, etc.
The reality is that there is a large degree of randomness in very close games. While the aggregate of blowouts tends to illustrate the underlying talent on the roster (or lack thereof). This phenomenon is true in nearly all sports: the record in 1-run games in baseball is considered to be largely a fluke and therefore not predictive going forward. Same with 3 point wins in football.
Of course, the hypothetical above is unlikely to ever happen. Just pointing out that studies have shown that point differential, while imperfect, is a better predictor than record alone in terms of predicting a team's future. If you consider 2 point games to be a 50/50 proposition, or even a 60/40 proposition, then it seems very unlikely that Team A would continue to go 0-whatever in their remaining 2 point games.
The other "of course" is that Team B could still win the above game, and the model would not be invalidated by that outcome.
EDIT: Back to the real world: The Celtics point differential standings will be cold comfort if the Celtics find themselves having to play 4 games in Philly or Indiana in the opening round.