Good, then you were actually paying attention. And so, despite Team A winning 58 games and Team B winning 50 games and thus Team A has a higher playoff seeding if they were in the same conference, your stat favors Team B. That means this statement of yours is false:
Actually - at the end of the season, he is right. Home Wins - Rd Losses = Rd wins - Home losses. But that's kinda meaningless. It's just a fact born out by playing 41 games on the road and 41 games at home. And Home wins - rd losses just tells you how many games over .500 you are. Nothing more. It is, essentially, the W-L record. But as a predictor, it's flawed because in mid-season it doesn't account for the statistical noise you see in W-L's. There's a random element to W's and L's. You can lose 4 really close games by 1 point, and win 2 games by 25. Your pt differential would be +7.7 but your record 2-4. If the two wins were at home and the 4 loses on the road, your RDW-HL = 0. That predicts a .500 team. But if the 4 loses were at home and the two wins are on the road, you are at -2. Now, the truth of the matter is that a team that blows out two opponents on the road and loses by a point at home to four teams....is probably isn't a below .500 team. Their pt diffential of +7.7 is probably the more reliable indicator. Seems Durvasa did the numbers and found that pt diff is the better predictor during the season.
Random, but I've been wondering... If its lakers playing clippers, is it home or away?? How do people account for it statswise? Because wouldn't there be more lakers fans even at a clipper 'home' game?
I am so not wrong here. I never said win %. Obviously not. I'm talking +/- wins like Magic numbers? right? You are basing stats on %...when in fact all stats are based on +/- of Wins. So...in your scenario, the 58 win team has a +8 win vs the 50 win team. Right...we all agree on that? right? And they also have a +17 RW vs HL to the +9 of the other team. Note how the difference is 8 in primary Wins to Losses. Are you really going to argue this with me. Just look up every NBA standing ever created. The +/- of Wins ranking is identical to the +/- ranking of RW/HL. It is mathematical.
It is not. PT diff is awesome...but is also flawed for everything you said. Everything about the schedule of the season is tightly packaged in RW v HLs...it is so beautiful, and y'all want to create a more complicated world that doesn't exist. RW v HL is hands down the easiest and best predictor because it converges on reality. It never says the Rockets are #5 in the league....like Hollinger does. The Rockets are closer to #10 in the West than #5 in the league...that's just a fact. y'all are looking for fancy numbers when RW v HL is staring you in the face. Hollinger miserably fails throughout the year at his rankings until he finally converges on reality somewhat (but never 100% like RWs v HLs), and y'all are defending him vs RW/HLs ... really
I know they do, I live in LA, my cousin's bf's a super fervent clipper's fan; but even he doesn't know how they calculate home/away advantage when it's a Lakers game at a Clipper's 'home' game because they're both held at Staples Center... except the logo on the floor's different. It's the same demographics and the Lakers fans outnumber the Clippers fans. So theoretically Clippers are essentially playing a couple more 'away' games than other teams during 'home' games against Lakers unless the whole matter of not having to fly and being able to sleep in your own bed is enough to count for both the Lakers and the Clippers as having more 'home' games than any of the other teams in the league. I mean, hell, if they get another team to be housed in Anaheim that theoretically means three teams with a ridiculous crop of 'home' games when they play in their division.
Actually, I'd like to pose a question; given that the Rockets have vastly improved in December (for that matter several teams like the Lakers and Hawks have improved recently), what are the RW v HL of all teams for December only?
then you might as well just look at win-loss records. why even bother with anything else. fact is, he showed you why it's not a great predictor. Sure at the end of the season it is, but earlier on it's not. that's the statistical fact.
what? we don't look at Wins v Losses because the schedule is skewed. h e l l o No he didn't. He would have to show us RWs v HLs stats, which he got wrong, and can't show....otherwise I would myself. It is a +/- stat, not a % stat. Which I explained mathematically... Wrong again RW v HL eliminates the skewing better that PT differential does, because it actually gives credit to a WIN!!!!!!!!!!!! Do you not see that? That said, I'm a big fan of using PT differential, too. It is the second factor I look at next to RWs v HLs.
OK, just saw your edit. Yes, finally something that makes sense. If we look at it in terms of a plus/minus (wins - losses), then RW - HL will be half of W - L by season's end, so yeah in that case RW-HL will be proportional to total wins. But how does that help us make better predictions? That's the whole point, isn't it? Here's one approach to convert your stat into a projection for the end of the season: PW1 = W + (82-GP)/2 + (82-GP)*(RW-HL)/GP The reasoning is we expect that RW-HL will scale with additional games played (if RW-HL is meant to be some sort of barometer for team performance, that seems reasonable). So if a team's RW-HL was (say) +4 through 20 games, then we expect it to be on average +12.4 in the remaining 62 games. That corresponds to 43.4 expected wins over those 62 games. Add to that the current number of wins, and you get projected wins. Alternatively, I could have completely ignored current wins for the team, and rely solely on the RW-HL stat. Maybe this is what you'd prefer, project final seedings correspond to current RW-HL stat. For this, I can alter the formula to: PW2 = 41 + (82)*(RW-HL)/GP For the current season, the projections are: Rk Tm G W RW HL RW-HL PW1 PW2 1 SAS 33 25 12 2 +10 64.3 65.8 2 NYK 30 21 9 2 +7 59.1 60.1 3 GSW 31 21 11 4 +7 58.0 59.5 4 LAC 31 25 10 3 +7 62.0 59.5 5 DEN 32 17 8 1 +7 52.9 58.9 6 OKC 30 24 8 2 +6 60.4 57.4 7 MIA 29 21 7 2 +5 56.6 55.1 8 IND 31 18 8 3 +5 51.7 54.2 9 MEM 28 19 6 3 +3 51.8 49.8 10 ATL 29 19 8 5 +3 51.0 49.5 11 UTA 32 15 6 4 +2 43.1 46.1 12 MIN 27 14 5 4 +1 43.5 44.0 13 MIL 29 16 7 6 +1 44.3 43.8 15 BOS 30 14 5 5 +0 40.0 41.0 14 POR 29 15 4 4 +0 41.5 41.0 17 TOR 31 11 4 5 -1 34.9 38.4 16 HOU 31 17 5 6 -1 40.9 38.4 18 LAL 30 15 5 6 -1 39.3 38.3 19 CHI 29 16 7 8 -1 40.7 38.2 21 BRK 31 16 5 7 -2 38.2 35.7 22 DAL 31 12 4 6 -2 34.2 35.7 20 PHI 31 14 5 7 -2 36.2 35.7 23 DET 33 11 3 8 -5 28.1 28.6 24 PHO 32 11 2 7 -5 28.2 28.2 25 ORL 31 12 5 10 -5 29.3 27.8 26 CLE 32 7 4 10 -6 22.6 25.6 27 SAC 30 11 1 7 -6 26.6 24.6 28 NOH 30 7 4 12 -8 19.1 19.1 29 CHA 31 8 3 12 -9 18.7 17.2 30 WAS 28 4 1 11 -10 11.7 11.7 The projections look reasonable. Unfortunately, over the same 356 team sample I looked at before, the RMSE for PW1 was 6.5 and for PW2 it was 7.3. So, while its better than using RW-HL% to project performance, it still doesn't appear to be as helpful as using basic W-L record or point differential, which is even better.
No, Rockets are a #10 seed by very easy to gauge +/- rankings. And note that Durvasa's numbers above place the Rockets at a #10 seed in the West. Weird how occams razor works, right? Aren't you being subjective here? Rockets are a #10 seed in the numbers you gave us above. So you confirmed that RW/HL is an easy predictor moreso than pt differential...plus we could just add pt diff ranking to RW/HL ranking very easily....as a new heypartner stat. No need for Hollinger's crap? We are not a #5 team in the league, no matter how you slice it,,,,and certainly not according to a RW/HL measure of any kind.
You're confused. The formula (PW2) was designed to please you. It is based directly on RW-HL, and so it "projected" the Rockets to be 10th in the West (because that's what their current RW-HL indicates). The point is that I took that formula, and I ran it against every team from 99/00 season through the 10/11 season to see how accurately it projected the final records based on results through December. I'm telling you that, over those 12 seasons, it does not do as good as job as even simple W-L record, not to mention point differential. I provided the RMSE. Nothing "subjective" about that. Just the facts.
You aren't really implying that home losses vs road wins should be ignored right? So any measure you create will include that, right? And did any of your measures say the Rockets are the #5 best team in the league? My whole point is you had to run that diagnostic, when RW/HL of mine that I can read everyday in the Standings can tell me exactly what your stats can run and even better than Hollinger. Run other stats. I will bet your stats can't put us at best #5 in the West, right now...(not even #5 NBA) if you include W/L and home losses v home wins.
I'm not understanding the confusion here. You maintained that RW-HL is the best measure of how good a team is. I actually tested that theory by comparing it to some very simple predictors. One based on point differential. One based on current Win%. I also tried CH's suggestion of assuming future HomeWin% and RoadWin% will stay the same (that also "converges" to the final record and accounts for Home/Road games). All three of those very simple predictors were more accurate than using RW-HL, when tested over multiple seasons. So, I have no reason to favor RW-HL over any of the other three, as its not particularly simpler either. Now, I don't know how your stat would compare to Hollinger's in terms of accuracy over several seasons. That would require a lot of time and effort for me to figure out, so I'm not going there.
heypartner seems to put a lot of weight on the fact that the RW-HL metric converges as the season winds down. But Home wins-Road losses and total wins-total losses also converge to the true record as the season ends. If you are focused on separating home/away games but aren't going to account for strength of schedule and you think the home record is a good predictor of future home games and road record is a good predictor of future road games, then why not simply average home win % and road win % (which also converges)? I'd be interested in seeing how each metric performs if you backtest them over past seasons and see how they perform as the season progresses.
