Power Ranking

[EvanZ]

(I think mystic and I are in agreement.)

The ratings are not computed with SRS, though. You know it's a regression:

MOV ~ ATL + BOS + CHA + ... + ATLHCA + BOSHCA + CHAHCA + ...

If Atlanta plays at Boston, then ATL = -1, BOS = +1, BOSHCA = +1, and all other variables are set to zero.

Like I said in my comments, if it were:


MOV ~ ATL + BOS + CHA + ... + ATLHCA + BOSHCA + CHAHCA + ... ATLAWAY + BOSAWAY + CHAAWAY + ...

In this situation (which is redundant), everything would be relative to a neutral court as in your example (e.g. ATL = -1, BOS = +1, BOSHCA = +1, ATLAWAY = -1).

I'm interested to hear what other people have to say.

[mystic]

(I think mystic and I are in agreement.)

If you refer to my first paragraph, then yes.

[J.E.]

I think the team specific HCA needs to be interpreted as "against a team which behaves normally (as in "has an average HCA"), team X has a HCA of:"

Say you lots of data so that you could actually compute "Team Average Point Differential" with double the amount of teams. Each team gets split into a home and an away version of itself: DEN_H, DEN_A etc. Then, when forecasting point differential of a game, you would not add any HCA to the home team, because homecourt advantage should already be accounted for. It is clear that, in this scenario, the built-in HCA is dependent on the home team's strength in home games AND the away team's strength in away games.

Obviously, if you look at Team Average Point Differential the standard way (30 teams instead of 60), then each team's Average Point Differential is just the average of the PD of the home version and the away version of itself. Why should we then only account for HCA of the home team?

[EvanZ]

Obviously, if you look at Team Average Point Differential the standard way (30 teams instead of 60), then each team's Average Point Differential is just the average of the PD of the home version and the away version of itself. Why should we then only account for HCA of the home team?

It's parsimonious. Why introduce more factors when they don't add anything (as far as I can tell). We could have a ACD (away court disadvantage), too. Doesn't make a difference. But why do you need both? It's redundant.

[J.E.]

Say you had 30 teams, all with strength 0.
28 teams have a team specific HCA of 3. Two teams remain, they never play each other until the last game of the season.
Team_A has an HCA of 6, which is equivalent to "their average point differential in home games is 6 points better than their average point differential in away games". At home, they outscored teams that usually get outscored by 3, by 6
Team_B has an HCA of 0. They had they same point differential at home and away. Opponent home teams, which usually won every home by 3, play games with this team to a tie.

Now Team_B plays at Team_A. Should they win by 6? I say they should win by 3! Otherwise you're accounting for the fact that Team_A has been overperforming at home, but you're completely ignoring the fact that Team_B overperformed in every away game

[EvanZ]

Now Team_B plays at Team_A. Should they win by 6? I say they should win by 3! Otherwise you're accounting for the fact that Team_A has been overperforming at home, but you're completely ignoring the fact that Team_B overperformed in every away game

Team A should win by 6, because that's the coefficient you gave. What I don't understand is why you are trying to interpret the value of the coefficients determined from a regression differently than you would, say, with RAPM or any other regression with dummy variables. How would you change the model I gave above?

[DSMok1]

Basically, we want to know what the true, NEUTRAL-COURT strength of a team is. We ASSUME that this is the average of away and home performance.

It has been calculated what the difference between home and away performance is, on average, for the league. This value has been termed HCA. When an actual game is played, adjusting the expected results of a game by adding 1/2 HCA to the home team and subtracting 1/2 HCA from away team is the way I would do it.

I don't think it is possible to solve for HCA (vs. Neutral) and Away-team-disadvantage (vs. Neutral), is it? So, the simplest way to do it is to call the total of the two HCA.

[EvanZ]

Ok, my turn to give a numerical example of what I think is going on.

Say we have DEN and BOS. For the sake of argument, DEN's HCA is +6. BOS is 0. Also, let's say that both teams have an average p.d. of 103.

DEN = 100 on the road
DEN = 106 at home

BOS = 103 on the road
BOS = 103 at home

If BOS plays at DEN, I would say (my model) that DEN should be a 3 pt favorite.

BOS = 103 DEN = 100 + 6 = 106

If DEN is at BOS, then BOS is a 3-pt favorite there.

EDIT: Ok, I think I am understanding what Jerry is saying now. If we assume these HCA and then run a regression without them, those power ratings include the home and away "strength", and we are then overestimating by throwing these HCA in there. Is that it?

[xkonk]

Can't you guys set up your model to do a prediction? Tell them there's a game between Denver and Minnesota (or whoever) but leave the outcome blank or NA, whatever your program prefers. Then see what the model fit for that result is. That should tell you how to work the numbers.

[EvanZ]

I did do that. The question was really how to use those HCA for future predictions. Anyway, I think I get it now.

[J.E.]

Has someone a link for a good website where I can find the Vegas lines also for past games?

http://scores.goldsheet.com/merge/tsnfo ... slist.aspx

I've got most of it in text files with the html stripped away. Hit me up if you want them.

I don't know if they represent opening line or closing line, though. I also don't know where they take the line from, as there are hundreds of different sportsbooks

[EvanZ]

I've got some Vegas power ratings (based on lines from this season) on my website now:

http://thecity2.com/vegas-power-ratings/

[mystic]

http://scores.goldsheet.com/merge/tsnfo ... slist.aspx

Thanks.

I used the data to have another comparison. I have the RMSE for the last 67 games:

with HCA: 13.37
pace adjusted: 13.24
without HCA: 14.32
only HCA: 14.13

Vegas: 11.73

Well, doesn't look good for my Power Ranking at all. At least the ranking predicted the correct winner 54 times, Vegas has "only" 52 times the correct winner. I guess the lower RMSE for Vegas comes from the fact that they have all teams closer to the mean than my ranking.

[Jeff Fogle]

Back from the trip. Thanks to Evan for posting his regressed assessment of the Vegas lines to this point in the season. Here's a new look from me, kind of eyeballing recent numbers as the season continues to evolve.

+7: Miami
+6: Chicago
+5: OKC
+3: Portland
+2: Denver, San Antonio, NY (with both Amare and Carmelo), Boston, Orlando, LAL, LAC
+1: Philadelphia, Dallas
0: Indiana, Atlanta
-1: Houston, Memphis
-2: Phoenix
-3: Minnesota, Milwaukee, Utah
-4: Toronto
-5: Golden State
-6: New Orleans, Detroit, Sacramento, Cleveland, Washington
-7: Charlotte
-8: New Jersey

Generally speaking...give 3 points for home court...1 point for a rested team playing somebody on night two of a b2b...take off 1-2 points if a key starter is out...night three of a b-2-b-2-b can skew in some instances. Will be interesting to chart those over time....to see how many teams are just playing them straight and fading at the end, compared to how many are taking night two "off" in tems of peak intensity and going hardest in games 1 and 3...in essence playing "no" b2b's because they take the middle one off. Think that was behind some of the early G3 successes.

Again, hope anyone following the markets feels free to suggest tweaks or changes. And, these are estimates of how the markets are rating the teams...not personal assesments of those teams. The market had Boston at -5 vs. Dallas tonight...I've got the market showing them 1 better on a neutral court...then 3 for home court...then 1 more because Dallas was in a b2b while Boston was fresh.

[Jeff Fogle]

Best vs. expectations heading into Wednesday Night according to the standings at covers.com:
Philly 7-1-1 ATS (all will be against the spread)
Denver 8-2
Portland 6-2-1

Worst vs. expectations:
Detroit 2-8
New York 2-7
Sacramento 3-7

This page updates daily I think....for those interested in performances vs. expectations for all 30 teams.

http://www.covers.com/pageLoader/pageLo ... dings.html