What happens if you normalize against playoff teams only? Does the correlation become stronger? Beating the Grizzlies by 20 may not represent future playoff performance well, but beating NO by 20 might. Nice work. You should try JMP, it's so far superior to Excel for statistical analysis.
Statistics are like miniskirts, if you know how to unravel it you will reach the ground truth. Fixed!
Some of that can be cured if you include the standard error. If the error or overlap is large then the sample size is too small. Otherwise, it may be O.K.
guys, doesn't this just say, "if you have score more points than most opponents, you have a better chance of having a high winning percentage?" pardon my simpleness, but it seems that yes, you would win, if you outscored an opponent.
hah, another way to look at it is that the Lakers and Celtics beat their guys down just as much as we do in 2008. But during this last 10 game stretch there isn't anyone even remotely close to us. So yeah, bring on the Lakers and the Celltics! We'll give them a beatdown to remember! lol.
See the other teams are pretty much staying consistent with the Celtics raising their level of play somewhat. But the Rockets..they are blowing out their opponents and getting better and better at doing it.
As much as I claim to understand all the numbers mumbo jumbo as a web developer (computer programmer lite), I don't get it! But, wouldn't it make sense to judge point difff & strength of schedule?!
Great stuff! But.. we should consider a Chow Test from probably after our 6 game losing streak in December to what we see right now. Definitely two different regressed performances. We'd see the same in that insane 2004-05 season.
It is a good link, and I did see a nice discussion. Didn't those guys try to use point differential(PD), win%, rWin%, Pyth^14, Pyth^16, or whatever estimators to predict the future (future win%)? I don't any problem with that. But, where is the conclusion? Which estimator is better? Why is PD better(more powerful) than win%? Good point. If PD really is so powerful, imagining that those "stats geniuses" are able to predict the current 19 winning streak by the PD of 6 game losing streak in December, wow, how great it is! Even a caveman can do it.
I think the general consensus is that PD works very well as a current win estimator, and the evidence seems to indicate that it can work at least as well as win % as a predictor for future wins. Some think it works better as a predictor than current win %, but others argue that while the evidence seems to point to that it's not conclusive (i.e. "statistically significant"). I'd look at it this way. Suppose two teams (A and B) have a more or less similar win-loss record, but team A tends to win more in blow outs and therefore has a high PD, while team B wins more close games. If I had to choose which was the better team with just that information, I'd favor team A. Winning close games has more to do with chance then winning blowouts. I think that's generally a good rule of thumb, but of course there are exceptions. Other information worth considering is record and/or PD against good teams.
If you plotted the Power Distribution as a Moving Average, that would give you the tendency of a team to win. A ten game moving average would probably give you a pretty decent predictor. What would be more meaningful is if you plotted the Power Distribution and Moving Average against the Power Distribution and Moving Averages of the the team they were playing against on any particular night. This would give you a good look at our opponents and factor in the strength of schedule. Course..that's a heck of a lot of work. Any takers?
It is interesting, but need to look both sides before any conclusion. How about A and B have a similar PD, but A has a better win%, which team is better? And based on the chart in OP, there is a perfect linear correlation between PD and win%. Doesn't it mean PD <=> win% are statistically equivalent? How can PD be a better predictor?
Our Points differential would be higher too I reckon. I can think of two games off the top of my head where blew them out by half time or 3rd Q (New Jersey and Atlanta) and our starters didn't play much in the 4th. I mean we were winning by 30 at some points there.
I don't know if its a better predictor, but apparently that's what the past suggests (though not conclusively). And I don't see the contradiction in one being a better predictor than the other, even though they are highly correlated. They are not equivalent stats.
