I have calculated MLS Power Rankings based on 2003 regular season and play-off results, and plan to update them weekly as the season progresses. The ranking scheme assumes goals are Poisson-distributed, and takes into account team strength, home field advantage, and strength of schedule differences. Pre-Season Power Rankings: Code: [size=1] Rank Team PPCT GD 1 NE 0.593 0.50 2 SJ 0.593 0.44 3 CHI 0.537 0.39 4 KC 0.463 0.17 5 CLB 0.444 -0.06 6 DC 0.426 0.00 7 LA 0.407 -0.06 8 COL 0.389 -0.22 9 MET 0.370 -0.06 10 DAL 0.074 -1.11 [/b][/size] One advantage of this ranking scheme is its simplicity and ease of interpretation. PPCT is the predicted percentage of points each team would win in a complete home and away round-robin tournament, and GD is the predicted goal difference per game. These results may be interpreted to mean that NE would be favored over DAL by 1.6 goals if they were to meet on a neutral field.
Interesting use of a basic statistical distribution. It would be of more interest if your system somehow incorporated current differentials such as LA's not having to start with 8 road games, the loss of Euro signings, retirements, draftees, new signings, stars guaranteed to be injured for a length of time, venue changes, Nat team commitments, and this years' schedule.
Re: Re: 2004 MLS Power Rankings [R] I agree, it certainly would be more interesting. To do what you are suggesting would require a model that relates individual player performance to team results. I have some definite ideas how to do that, but the relevant data on individual player performance are lacking, and without automation would be tedious to collect.
Re: Re: Re: 2004 MLS Power Rankings [R] I'm only imagining that first post as being recited by the nerdy scientist from the Simpsons. As you can see by your rankings, computer generated sports guidelines are useless (e.g. The NCAA BCS system). It's easier and much more honest to just take the stats from last season and then see who got stronger and who got weaker. The brain is the best computer for all of that "tedious" data.
Thanks for the invitation, mellon002. I have posted in Stats and Analysis before, but am aiming for a wider audience with this thread.
2003 Over- and Under-Achievers It is interesting to compare each team's actual point percentages with the predicted power rankings: Code: [size=1] Rank Team ACTUAL PRED DIF 1 DAL 0.256 0.074 0.182 2 MET 0.448 0.370 0.078 3 CHI 0.608 0.537 0.071 4 COL 0.427 0.389 0.038 5 KC 0.465 0.463 0.002 6 LA 0.406 0.407 -0.001 7 SJ 0.588 0.593 -0.005 8 DC 0.406 0.426 -0.020 9 CLB 0.422 0.444 -0.022 10 NE 0.495 0.593 -0.098 [/b][/size] DAL, MET, and CHI top the list of over-achievers, while NE, CLB, and DC top the list of under-achievers. As bad as they were in 2003, DAL seemed to win more games than their horrific goal differential would imply they should. Bradley and Sarachan deservedly get props for their coaching, but Mike Jefferies for coach of the year, anyone? Interesting that Hudson and Andrulis were under fire most of the season, but Steve Nicol seems not to have suffered unduly for NE's relative under-achievement. Guess that's what a strong finish will do for you.
Re: 2003 Over- and Under-Achievers I think the predictions are for 2004, not 2003. In other words, the good ones are the teams who have higher power rankings than where they finished (like NE).
Re: Re: 2003 Over- and Under-Achievers That all depends on your point of view. If I look retrospectively at NE's 2003 performance, they under-achieved in the sense that they won ~10% fewer points than their goal differential would indicate they should. Alternatively, I can look prospectively at NE's 2004 performance and say that based on their 2003 goal difference, they are a better team (should win ~10% more points) than their 2003 record would indicate.
Re: 2003 Over- and Under-Achievers I don't think that this is the right way to look at this at all. Margin between Poisson-expected performance and actual performance isn't really a measure of how good a coach is - it's a measure of how lucky a team is (2002's Fire had the worst win%-Poisson% in the league - was Bradley the worst coach in the league). A good coach doesn't coax a few extra % points from his goal differential, he coaxes a superior goal differential out of inferior players, something which this setup can't really account for. (Could you give us the strength of schedule numbers you computed - this isn't the NCAA or even the A-league - teams play very similar schedules, and I wouldn't expect these to have much of an impact on teams' numbers.)
Re: Re: 2003 Over- and Under-Achievers I said this tongue planted firmly in cheek - who knows what accounts for the difference? Random luck, injuries, coaching... the list of possibilities is endless. I haven't performed the calculations without accounting for strength of schedule so I can't quantify the impact for this particular set of data. In MLS the potential for significant strength of schedule differences IS present due to the unbalanced schedule between East and West conferences, but as a practical matter this effect is mitigated somewhat by the parity prevelant in the league. At the end of the season I saw a calculation of strength of schedule differences (by Sagarin at USA today?) for the 2003 regular season, and IIRC the differences were pretty small (+/- 5%), in line with your intuition.
Re: Re: Re: 2003 Over- and Under-Achievers Well, yes obviously; nevertheless, it's a valid question (and I don't see how injuries could factor into this at all, maybe you could explain). I don't have your complicated numbers, but if you want to compare win% and a poisson whatever (using 1.5 as the exponent), you get, where a negative number means you were worse than expected: Bob Bradley -0.009 -0.034 0.024 0.002 -0.082 0.017 Bob Gansler: -0.017 -0.033 0.134 0.055 -0.016 Sigi Schmid: -0.047 -0.026 -0.000 0.019 -0.050 Ray Hudson: -0.033 0.046 0.005 -0.037 Doesn't seem indicative of much to me.
Re: Re: Re: Re: 2003 Over- and Under-Achievers Jeez, Chris, you're right, OK, it's 100% luck - random variation. Tongue planted firmly in cheek = joking Maybe another reason I didn't post in the stats and analysis forum is that I wanted to be able to post some simple rankings without necessarily having to undergo a colonoscopy with a scanning tunneling microscope.
Re: Re: Re: Re: Re: 2003 Over- and Under-Achievers Excuse me for being a humorless son-of-a-bitch. I'll now take my leave of this thread as discreetly as possible... (back to the original topic)
Re: Re: Re: Re: Re: Re: 2003 Over- and Under-Achievers Awww, c'mon, I'm just being a smart-ass as usual, please don't take it personally. As the original responder noted, by themselves the pre-season rankings don't mean much - what will be interesting is how they change as the season progresses (which, I think we can all agree, can't start soon enough!)
Re: Re: Re: 2003 Over- and Under-Achievers Meaning, since they blew some teams out, that made them overall underperformers?
Unless there are fairly significant roster moves to come, I would rank DC no higher than 9th right now. As much as it pains me to say so - I think it will be a huge hill to climb.
Re: Re: Re: Re: 2003 Over- and Under-Achievers After reflection (I can't say after further reflection, because I obviously wasn't thinking before that post), ChrisE has it right. The proper interpretation of the table is probably not to think of NE as the most under-achieving team, but rather the most unlucky. That said, I still hesitate to characterize DAL as the luckiest team in 2003. I wonder if the rather dramatic difference in performance under Jeffries vs under Clark might be inflating the difference between actual and predicted point percentage.
Between #6 and #8, we've almost got our midfield covered... Not that this is reflected in the rankings, but under Nowak I expect to see significant improvements, in (1) team cohesion and morale, and (2) better use of young talent. Not to mention that Freddy will be starting by May. Don't listen to all the nattering nabobs of negativism over on the DC boards. DC won't contend for a title this year, but they'll be significantly improved.
Strength of Schedule Adjustments For those of you interested in delving into the gory details of the rankings, a comparison of rankings with and without strength of schedule adjustments is posted over on the stats and analysis forum: strength of schedule
Week #1 Power Rankings This week's Power Rankings, LW is last week's ranking: Code: [size=1] LW RANK TEAM PCT GD 1 1 NE 0.537 0.39 2 2 SJ 0.537 0.33 3 3 CHI 0.519 0.39 6 4 DC 0.519 0.17 8 5 COL 0.481 0.06 9 6 MET 0.463 0.11 4 7 KC 0.407 0.06 7 8 LA 0.407 -0.11 8 9 CLB 0.315 -0.28 10 10 DAL 0.074 -1.11 [/size] COL and MET are the biggest gainers this week (up 3) after their away draw at DAL and away win at CLB, respectively. KC is the biggest loser this week (down 3) after their home draw to CHI.
Why did LA drop from a #7 ranking last week to #8 this week after tonight's convincing win over New England?