|
Author
Recent Articles: Recent News Posts: 
CONTRIBUTORS
Articles:
News:
Join Us:
Archives
|
by KnickerBlogger | permalink | trackback | comments |
In addition to their weekly power rankings, maintained whimsically by Marc Stein, ESPN is now publishing a statistical power ranking. The formula, created by John Hollinger, relies on point differential and recent performance. It does account for strength of schedule, but unlike OTTER, it doesn't factor which games are won. For this session of OTTER, I'd like to compare the two statistical systems using the stats from January 22nd.
Here's how ESPN ranks the top 5 (OTTER ranking in parenthesis): PHO (2), DAL (1), SAS (3), HOU (6), CHI (10). Hollinger's system and my system disagree over who's #1. In John's rankings, Phoenix takes the top spot over Dallas with a superior point differential and a flawless last 10 games. On the other hand, OTTER likes the Mavs over the Suns for 3 reasons. First is Dallas' superior overall record (34-8 to 32-8). Second is their tougher schedule. ESPN's calculates the Mavs strength of schedule at .502 to the Suns .492. Finally OTTER saw the Mavs beat the Suns twice: 119-112 in November and 101-99 in December. How does a ranking system see a game? OTTER, based on Dr. Colley's matrix, calculates a baseline win% ([1+wins]/[1+games]) and then adjusts for each game played. If a team won, its win% gets a bonus depending on the opponent's win%, the margin of victory, and the location (home/road). In other words your win% would go up more if you beat the Mavs on the road by 20 than the Celtics at home by 1. After it has done this for every game, OTTER goes through each game again, this time using the new win% to calculate a newer win%. It does this over and over again like a monkey with a miniature cymbal until it regresses to its "true" value. Hence why OTTER knew that Dallas is 2-0 against the Suns, since those 2 games are factored into the result. Another big discrepancy between our top 5 lists is the Lakers' position. Hollinger ranks them 12th, OTTER 5th. Los Angeles has a great record, but only has the 10th best point differential in the league. Hence's John's formula, which is largely reliant on point differential, paints them as the league's 12th best team. But the Lakers have the league's 5th best record, they've faced the 3rd toughest schedule, and they've compiled an impressive 4-2 record against the league's top 3 teams (PHO, DAL, SAS). OTTER is founded primarily on the strength of each game (accounting for the result, opponent, location, score) and pushes the Lakers up to the 5th spot.
Hollinger's 6-10: UTA (4), ORL (13), DET (8), WAS (9), TOR (19). The real eye-opener in ESPN's rankings is the 10th ranked Toronto Raptors. Under each team, ESPN chooses a fan to write in a team description. Toronto's is from Andy (from Millbrook):
The surging Bobcats? I always thought surging meant "a strong movement forward", not "beat up on crappy teams." The Bobcats beat the Eastern flowerhouse Hawks twice, the Sixers who were bad even with Iverson, and the no-need-for-a-funny-description Knicks. As for that Utah-Toronto game, the Raptors would have won if the game ended with about 8 minutes left in the 2nd quarter. That was the latest Toronto held a lead. ESPN's high standing of the 20-22 Raptors is hard to justify, even for uber-blogger Scott. Toronto has faced one of the easiest schedules thus far (21st in ESPN's SOS), and they don't have a particularly good point differential (14th). So what gives? The Raptors faced two teams with great records, Dallas and Utah. Even though they lost both games, the matchups increased Toronto's strength of schedule. Meanwhile they beat a bunch of bad teams in Atlanta, Charlotte, Boston, Philly, and Sacramento. Combining the inflated strength of schedule with the Raptors strong play against the hapless gave Toronto a fortunate set of circumstances, at least in the ESPN ranking system. Hollinger's formula would show Toronto to be improved, because it doesn't account for which opponents they beat in that 10 game span. Oddly enough if the Raps destroyed the Mavs instead of the Bobcats, it wouldn't have changed their ESPN ranking at all. OTTER ranks Toronto a more conservative 19th.
Hollinger's 11-15: DEN (12), LAL (5), NJN (16), CLE (10), IND (11) Up to this point, here's how both of us have the Eastern teams ranked: ESPN-RANK OTTER-RANK Chicago, ESPN's highest ranked Eastern team, is only 4th in OTTER's Eastern team ranking (and 10th overall for OTTER). The Bulls have had some huge blowout victories this year. You probably remember the 42 point destruction of the Miami Heat to start the season. In December the Bulls pounded the Bobcats by 39, and just recently flogged the Grizzlies for 45 points. In fact, the Bulls have had 5 games we're they've won by 20 or more, but they haven't had a single 20 point loss. Compare to the Magic, who have the same record, but are 2-2 in games decided by 20 points or more. Generally point differential is a solid way to gauge a team's effectiveness, but there are exceptions. Having a disproportionate amount of blowout wins or losses is more likely to be a sign of luck, rather than skill. I'm more comfortable with Chicago's 10th standing in OTTER, than 5th ranking on ESPN.
Hollinger's 16-20: LAC (14), MIA (17), MIN (16), SEA (24), NYK (25) No huge differences in the bottom half. Seattle, New York, and Boston are notably higher on ESPN, Portland and Milwaukee are lower. And anyway does it really matter if Boston is 24th or 30th? No one is going to brag that their team is 28th in one ranking, as opposed to 30th in another. In the end there seems to be a fundamental difference between the two rating systems. I created OTTER to gauge the strengths of each team, while John's formula is to enhance discussion about the league. OTTER tries to objectively rank the teams, while ESPN wanted to emulate "human power rankings". Each was made different by design. Published on Tuesday, January 23rd, 2007 at 10:35 pm 13 Responses
KnickerBlogger said on January 24th, 2007 at 9:20 am :
Trev, OTTER takes into account each opponent's strength. When it's calculating the new win%, if a team has faced a lot of tough opponents, it's win% will increase. For example if Team A beats Team B. In the first set of calculations, Team A gets an increase for Team B in their win%. Team B gets an increase in their win% for every team they've beaten due to their tough schedule. In the second go around, Team A will get an increase based on Team B's win%, but Team B's win% has increased due to their tough schedule. Hence Team A is accounting for Team B's opponent's SOS. Got it? Nelson said on January 24th, 2007 at 11:58 am :
A comment on the Bulls' tendency to blow other teams out: this is due to the unusually even distribution of talent on their roster; when a game gets out of hand and both teams clear their benches, the level of play for the Bulls doesn't drop very much, which leads to 20-point leads becoming 40-point leads. But, unfortunately for Bulls fans such as myself, this phenomenon doesn't mean that the Bulls' starters are that good when compared to the league's top teams. Their high average margin of victory is a reflection of the relatively high quality of their deep bench, which doesn't help very much if the starters aren't strong enough to beat the elite teams. Rob L said on January 24th, 2007 at 1:35 pm :
You point out something important with the Bulls, Knickerblogger. Their point differential has been affected by outlier games. I think we need to have an Equivalent Point Differential. Although that seems to be incorporated into OTTER already, I'd love a formula broken out of your larger scheme. Patrick said on January 24th, 2007 at 2:20 pm :
Outstanding. Thanks. National writers and strong bloggers should read and cite this and add to the discussion about the topic. Hope you and Hollinger will have friendly dialogue about the differences as well. The ratings are different and could stay that way or maybe can lead to even stronger systems. Patrick said on January 24th, 2007 at 2:48 pm :
Take simple average of Hollinger and Otter "Rankings" and top 3 becomes Phoenix/Dallas co-leaders and Spurs a clear 3rd. Utah and Rockets tied for 4th. A pretty large gap to the next set of whom order probably can be argued until the cows come home. KnickerBlogger said on January 24th, 2007 at 2:50 pm :
Patrick, I emailed John about the article, and he said it was a "good read". I think the important thing to remember is that we both had different intentions. John wanted to emulate how a human would make rankings, and (I'm guessing) was limited by a single equation that ESPN's stat guys could easily program. I don't have the same goal or restrictions. Don Babbitt said on January 24th, 2007 at 2:52 pm :
You must have missed this article from roughly one year ago. http://www.courtsidetimes.net/articles/337/ "If good teams don't necessarily win the close games, then what do they do? Simple: They blow people out." The same is true in the NFL. Patrick said on January 24th, 2007 at 3:02 pm :
Ok. Live and let live. Several methods to read and the differences set off good questions. Combining them maybe isnt everybody's impulse but I think that can be useful too. On Lakers difference I think I lean toward Otter's ranking. Ultimately since a lot of people look at these rankings are gauges for future playoff perfromance (maybe moreso than for the regular season win producing strength assessment they are designed for), I wonder if another verion of these rankings designed for this playoff strength interest could give much greater weight to performance against top 10, beyond what the strength of schedule features already do. Gabe Farkas said on January 24th, 2007 at 3:30 pm :
One way to deal with outliers (aka blow-outs) is to use a more robust analysis method that isn't as sensitive to outliers and other skew in the distribution. Perhaps using median point differential, rather than the mean point differential, would help to deal with this. Patrick said on January 24th, 2007 at 4:21 pm :
If current strength rankings are a sum of what has happened and a heavily weighted version could be useful for looking ahead to the playoffs, in between these it might be possible to look at the data from the part of the season that has been played and apply the strength information contained within it to the remaining schedule - which is not the same as the previous schedule- so simply extropolating from the overall strength number from the season to date to estimate expected wins during the remaining schedule isnt as rigorous as is technically possible. If anyone wanted to go that far. Patrick said on January 24th, 2007 at 5:03 pm :
Regression to the mean to damp season to date highs and lows some may be a proper part of the extremely detail method. KnickerBlogger said on January 24th, 2007 at 5:23 pm :
I sure read that article. But it's main point was to compare winning close games with blowouts. Not to say that blowouts are more important than expected win percentage, or winning games, or anything that OTTER takes into consideration. "Let me wrap this up by summarizing the two key points of the article: 1. A team's winning percentage in close games is not a particularly reliable predictor of their overall winning percentage. Leave a Reply
|
