[SOXROCK]

i would think a team's pace could differ in the NBA in a B2B, or 3+ days rest, as well as the pts per possession

I think you are right. Last night was an example of that with the charlotte vs. okc game. I had that game going over 190 by about 10+ points.

[nouseforaname72]

hey i was wondering if you could help me out? i've been using this and watching it but is keeps giving me "over" for almost every game :S i was wondering if mabye i'm doing something wrong? i tried it manually and with your excel document. I am getting the stats from
http://www.nbastuffer.com/2010-2011_...ced_Stats.html

if you think of the offensive and defensive efficiency as percentage of points scored per each possession, then to incorporate an opponents efficiency factor into a team's expectation you would multiply the teams respective efficiency numbers by the average pace

For example

Team A: Pace 95.1
Team B: Pace 90.1

Expected Pace = (95.1+90.1)/2 = 92.6

Team A: OE = 1.01
Team B: DE = .91

Expected Team A points = 1.01*.91*92.6 = 85.11

Team A: DE = .85
Team B: OE = .99

Expected Team B points = .85*.99*.92.6 = 77.92

Final score

Team A: 85.11

Team B: 77.92

Here is a spreadsheet I made about 3-4 years ago, one for NBA one for college with the formulas already filled in, you just need to enter the pace and efficiency numbers in the appropriate cells (for NBA I use 82games.com for stats, NCAAB I use kenpom and bbstate)

THe files are in excel. Make sure you DO NOT fill down the formulas for they are inconsistent from one cell to the next. Each cell corresponds to its preceding crossover, so if you need to add cells do it manually. I have pre-filled a ton though so it should be plenty to get you started

[nouseforaname72]

what formula are you using to get these numbers? im using the one mentioned but it keeps giving me "over" for every game.

[Meestermike]

A site I use immensely is HERE. It has many calculations already done for you. You can set up a web query to download as you wish.

[dirk93]

I have all of the data in excel, does anyone have any ideas on how to incorporate league averages into the formula? I have ORtg, DRtg, Pace, and SOS. I just don't know how to put it all together correctly.

[Samzilla]

Bump.

Let's say we have adjusted Pace, OEff, and DEff (so they're already taking league averages into account). How does Defensive Efficiency work into the equation?

Let's look at a hypothetical Team A:
pace: 90
OEff: 105
DEff: 100

We can divide OEff 105 by 100 (to get 1.05) and then multiply 1.05*90 to estimate Team A would score 94.5 pts. We can also divide DEff of 100 by 100 to get 1. But we wouldn't multiply 1 by Team A's 90 pace. Would we simply multiply the DEff by the opponent's pace?

[mcduggly]

Bump.

Let's say we have adjusted Pace, OEff, and DEff (so they're already taking league averages into account). How does Defensive Efficiency work into the equation?

Let's look at a hypothetical Team A:
pace: 90
OEff: 105
DEff: 100

We can divide OEff 105 by 100 (to get 1.05) and then multiply 1.05*90 to estimate Team A would score 94.5 pts. We can also divide DEff of 100 by 100 to get 1. But we wouldn't multiply 1 by Team A's 90 pace. Would we simply multiply the DEff by the opponent's pace?

This is what I was wondering as well.

[Justin7]

There are two ways to attack this problem. The quick way is to figure out the league average OEF and DEF. Your projected points scored is then: Pace * oef / def * league oef /100.

A better way is to estimate a team's oef versus this individual opponent. One approach is to use a regression. If you do this, I would exclude all historical matches where a team won by more than 15 to get a better fit on competitive or semi-competitive matches. You could use either margin of victory, or closing game spread.

Another approach is to use a log-based team OEF projection based on the two teams. You'll have to try them all, and decide for yourself what you think works best.

[Balco10]

http://youtu.be/2UZCGEjSBqA

[mathdotcom]

There are two ways to attack this problem. The quick way is to figure out the league average OEF and DEF. Your projected points scored is then: Pace * oef / def * league oef /100.

A better way is to estimate a team's oef versus this individual opponent. One approach is to use a regression. If you do this, I would exclude all historical matches where a team won by more than 15 to get a better fit on competitive or semi-competitive matches. You could use either margin of victory, or closing game spread.

Another approach is to use a log-based team OEF projection based on the two teams. You'll have to try them all, and decide for yourself what you think works best.

I fear for your BR

[Juret]

mathy, would you exclude an outlier in a poisson (GLM) regression that is more than 4 standard deviations away from its, by the model, predicted mean value? and then refit

[mathdotcom]

Depends

Arbitrarily monkeying with your data is the #1 sign you've gone down the wrong path

[yak merchant]

So I'm glad this thread was dragged out of the basement. It has lot's of interesting topics. Log5, League Averages, Medians, Normalization etc. So in baseball Log5 works like a champ (Sabermetric dorks got that one right), I've never done a study but I would assume league averages would work fine in the NBA, but let's talk about normalizing data for a something a bit more difficult such as College Football. So with 11 games (really 9 or 10 if you take out non D1 opponents), outliers all over the place, and "divisions" where league averages don't mean much, what strategies are you guys using to normalize? Let's take YPC for instance. First game of conference play, Team A plays Team B.

Team A is averaging 5.0 YPC On offense and allowing 4.0 YPC on Defense.
Team B is averaging 3.0 YPC on Offense and allowing 6.0 YPC on defense.

So Joe Six Pack's answer is....

Team A will average 5.5 and Team B will average 3.5

Given the data in a vacuum I guess that is all you can do, but I'm scared to think of the percentage of handicapers that go forward making decisions with that calculation (if they are doing calculations at all)



Problem 1. Averages suck. throw in one crazy game for Team A where they have 2 90 yard runs and then what?

Possible Solution 1. Use Medians. But then what if each game has one big play it in? Anyone use medians from the play by play data? Does this factor out a team's big play ability? Think Barry Sanders hey day Lions. 3 runs for a loss and one for 50 yards. Rinse Repeat.

Problem 2. Strength of Schedule. Team A has played cream puffs while Team B has played the hardest non-conference schedule in the nation. Then what? Can't use league averages as the league average might be 4.00 but they've played the best and the worst not the league "average".

Possible Solution 2 Adjust Team A's numbers using the "averages" or "medians" of the teams they've played. However this then starts the "infinity mirror effect" , what about the teams that their opponents played, and those that their opponents played and on and on.

Problem 3 Home Field. Some teams (especially in passing stats) just can't get it done on the road, others in makes little difference.

Possible Solution 3? Don't really have one. Not really enough trials to segregate data to home/away just using one season data. By the time you do get enough data the teams that are going to give up have given up. Hard to discern between sucky road teams and true "Strength of Schedule" adjustments without home/away data. Anyone having any success using previous year home/away adjustments?

Problem 4 Other unquantifiable variables. Weather, Turf/Grass, Field Configuration. Weather is minimized because drainage is also better these days and they are building fields with much less "Crown". (I stopped tracking it, but certain fields had such a crown I would never bet a road team in those stadiums as the visiting quarterback would have hell throwing. One example was Autzen at Oregon. I believe it used to have over a 2% grade which mean that it was over 2 feet drop from the center to sideline).

Field turf has also helped miniminze anomalies in turf/grass, but interesting to watch in the future as Natural Grass is going the way of the Dodo. Outside of the SEC and ACC Natural grass is getting scarce.

Possible Solution 4. Possibly "exclude" data in rare instances. If it's raining side ways with a 50 mph wind I don't think passing stats are going to be valid. I usually don't worry about it as the "median" usually takes care of most of these instances.

So.... Would love to hear how other people are "normalizing" data like college football stats, or any Log5 type methods for non-percentage stats like YPC. I don't have any elegant methods, and use some scary brute force adjustements by looping through schedules and comparing medians. I would guess it is much better than J6P's handicapping methods, but I'm sure there are plenty of methods used by the posters on this board that are much better.

[yak merchant]

Completely forgot probably one of my biggest problems of all.

Problem 5. Team A who played cream puffs was up so much at the half they pulled their starters and the stats are skewed . This is why I have over the years ended up not playing Big Dogs as half the time crappy team's stats reflect better than reality because they get to gain yards in mop up time.


Possible Solution. Use play-by-play to determine when a game is "over" and only use stats before that point in the game.


Crickets.....

[mathdotcom]

It's not easy yak

The modeler who controls for all that best makes the big money

[357vegas]

Maybe 2ND halfs are the best bet?

[yak merchant]

I've had good success with 2nd halfs in College Basketball. Not so much on CFB. Line makers are way ahead of me on which coaches do what in the 2nd half. Also the "taking the 2nd half off" doesn't often reflect in the score as much as in the stats. Good team screws around until bad team get's in red zone and then clamps down, result is super inflated stats for bad team in 2nd half. I then use full game medians in my computer program and over-rate the bad team going forward.

[degengamble]

Great stuff here - may have to
Model it in excel and do some testing

[Goat Milk]

of course you always have to account for league averages for every single one of your elements in your sets or else your numbers mean nothing. Numbers don't mean anything unless they are compared to something.

[illfuuptn]

Idk if it's been mentioned but you also need to be wary of "expected" outcomes.

just because the expected total given all the data is 195 doesn't mean that 195 is the median total. Think about the fantasy football props. A player's expected score might be 15 but the line would be somewhere around 9.5 or something. You have to keep in mind how the scores are distributed.

[James Marques]

Bump. One of the more interesting true modeling discussion threads I've seen in the HTT.

[lamichaeljames]

Bump. One of the more interesting true modeling discussion threads I've seen in the HTT.

agreed. I found it informative and interesting.

[LandryAZ]

Old thread, wondering if poster (or anyone else) has expanded on these models.

Obviously I DO NOT recommend playing any totals based on these results only, but I think it would be interesting to see how this plays out for a few days and maybe see if we can throw up some filters. Who knows until we try, right?

Game 1:Indiana Pacers vs. Milwaukee Bucks
Est. Score:
IND: 95.88
MIL: 95.9
Est. Total Score: 191.78
Total Line: 195.5
Play: Under 195.5

Game 2: New Jersey Nets vs. Orlando Magic
Est. Score:
NJ: 83.3
ORL: 104.1
Est. Total Score: 187.4
Total Line: 194.5
Play: Under 194.5

Game 3: Cleveland Cavaliers vs. Philadelphia 76ers
Est Score:
CLE: 91
PHIL: 100.8
Est Total Score: 191.8
Total Line: 188
Play: Over 188

Game 4: Charlotte Bobcats vs. Detroit Pistons
Est Score:
CHA: 99.5
DET: 96.4
Est Total Score: 195.9
Total Line: 183.5
Play: Over 183.5 Note: The difference between the estimation and line is more than 10 points.

Game 5: Washington Wizards vs. New York Knicks
Est. Score:
WASH: 96.6
NY: 110.4
Est Score: 207
Total Line: 208
Play: Under 208

Game 6: Chicago Bulls vs. Boston Celtics
Est. Score:
CHI: 97.5
BOS: 105
Est Total Score: 202.5
Total Line: 195
Play: Over 195

Game 7: Miami Heat vs. N.O. Hornets
Est. Score:
MIA: 100.6
NO: 88.3
Est Total Score: 188.9
Total Line: 188
Play: Over 188

Game 8: Memphis Grizzlies vs. Phoenix Suns
Est. Score:
MEM: 108.9
PHX: 113.4
Est Total Score: 222.3
Total Line: 221
Play: Over 222.3

Game 9: Utah Jazz vs. Golden State Warriors
Est. Score:
UTA: 113.1
GSW: 115.4
Est. Total Score: 228.5
Total Line: 223.5
Play: Over 223.5

Game 10: Toronto Raptors vs. LA Lakers
Est. Score:
TOR: 102.6
LAL: 119.4
Est. Total Score: 222
Total Line: 212.5
Play: Over 212.5 Note: 9.5 Point difference between Estimation and Line

We'll see how this goes but at first I'd only put small plays on games 4 and 10 since those had the greatest difference in points between the estimated score and the line. So yeah, these are today's predictions based off that basic formula. Let's see how it does and look for how we can fine tune it. (if possible)

Regardless of if it works or not, it'll definitely be a fun little project!

[JIBBBY]

For setting lines I believe theses types of formulas work..

For betting I recommend looking at both teams last 5 game totals, revisit their last match up totals if recent as well.. Look at how both are scoring home/road as well..

In hoop and hockey it's all about current play, numbers and trends IMO.. That's my opinion for what ever it's worth.. It works for me and has for many many years..

[gojetsgomoxies]

grunching to a certain degree here. we had a similar discussion in a handicapping thread i.e. finding winning % on games given each team's winning % (and a bunch of simplifying assumptions)_

isn't a team's winning % or offensive/defensive rating vs. the league average (or it's strength of schedule but let's assume it's the same)?

the equation that gave you a 70% winning % team vs. a 40% winning % team wasn't linear at all.......... there was link to bill james article for it in baseball but i think it was identical to elo chess.

so how do you adjust a good offensive team with a poor defensive team? or all kinds of combos of such?

[TheMoneyShot]

First time I read this thread.

If only everyone had an Albert Einstein brain.

[gojetsgomoxies]

i agree with the one poster that there is alot of "art" to this and that it's based on the "here and now"

you have historical data. there'd lots of it but it's generally stale and based on many conditions that no longer exist.

you have very recent data. informative but very limited in amount and very volatile (just because phoenix upset a top team doesn't mean they are now a top 5 team)

so you need to combine the two..........

[A4K]

This thread is on the right track but I would suggest adding referee data to the mix. I found that certain refs had a home dog bias at one point. I'm sure their actions affected the pace of play.

[KVB]

...I found that certain refs had a home dog bias at one point. I'm sure their actions affected the pace of play.

This may go a step further. Perhpas it's not the referee witht he bias but the situation and market environment.

In other words, the ref does his job, regardless of his name. Any ref could be in a particular spot and still show the bias.

There could be a coincidence that certain refs are put in certain spots, having little to do with the actual referee and much more to do with the spot.

[A4K]

This may go a step further. Perhpas it's not the referee witht he bias but the situation and market environment.

In other words, the ref does his job, regardless of his name. Any ref could be in a particular spot and still show the bias.

There could be a coincidence that certain refs are put in certain spots, having little to do with the actual referee and much more to do with the spot.

I apologize, the perceived bias seems to point towards away teams as dogs when
Scott Foster, Tony Brothers or Joe Crawford (now retired) were part of the officiating crew. 57.6% winning percentage over 960 contests. This led me to believe that the referees in question may have been up to something. This data was pulled 2 seasons ago so I don't know if anything has changed.

[KVB]

I apologize, the perceived bias seems to point towards away teams as dogs when
Scott Foster, Tony Brothers or Joe Crawford (now retired) were part of the officiating crew. 57.6% winning percentage over 960 contests. This led me to believe that the referees in question may have been up to something. This data was pulled 2 seasons ago so I don't know if anything has changed.

No need to apologize it's a good find.

I was just adding to the discussion and pointing out that there could an underlying factor that acutally takes the ref out of it. But the research to find the bias is the first step for sure.

Joey Crawford was always a little shady...lol.

[A4K]

No need to apologize it's a good find.

I was just adding to the discussion and pointing out that there could an underlying factor that acutally takes the ref out of it. But the research to find the bias is the first step for sure.

Joey Crawford was always a little shady...lol.

No offense taken. I enjoy your input and posts.

Crawford was around during the transition period for the NBA from a basketball league to an entertainment juggernaut in terms of revenue for the league. I'm almost positive it was stressed upon him the importance of keeping certain TV games close. Just my .02 lol