Stat question - scoring margin "ranks"

[andalusian]

We have seen that many stat heads consider scoring margins as "important" to define a team's projected success.

What I have not seen - is these scoring margins normalized for pace.

Since in every game the difference of the number of possessions of each team is 1 or smaller - it makes sense that scoring at large, and scoring margin should be normalized to the team's average pace - to really compare them.

Yes, no? What say you?

 

[RR7]

How would pace have an effect on the scoring margin, though? Or maybe I am misunderstanding. If we play at an extremely slow pace, and only have 85 possessions per game or something, our opponent will have the same. So if we win by 10, our scoring margin is that +10. If GS gets 100 possessions, their opponent will as well, or close to, and so if they win by 10, they will also be at the same +10. Their faster pace didn't necessarily help them to achieve that margin as far as I can see, unless you are thining that a team that plays at a higher pace ultimately will have more blowouts and higher scores, and thus effect that number.

 

[andalusian]

How would pace have an effect on the scoring margin, though? Or maybe I am misunderstanding. If we play at an extremely slow pace, and only have 85 possessions per game or something, our opponent will have the same. So if we win by 10, our scoring margin is that +10. If GS gets 100 possessions, their opponent will as well, or close to, and so if they win by 10, they will also be at the same +10. Their faster pace didn't necessarily help them to achieve that margin as far as I can see, unless you are thining that a team that plays at a higher pace ultimately will have more blowouts and higher scores, and thus effect that number.

The issue is, that there is a discrepancy between offensive and defensive efficiency and scoring margin - so one of those is out of whack when it comes to predicting wins.

Take an extreme case, let's say that there is a team that if it slows the game to 2 possessions is guaranteed to win by 2 points, no matter what. Compare it to a team that has 10 possessions per game, wins 50% of the time, but their scoring margine is +4 - the scoring margin stat gives you a prediction that the faster team that does not win as much is going to do better than the other team. Seems rather silly to me.

If Portland does better by keeping possessions low and forcing it's pace - but because the pace is low - the margin is low (not that many chances to score) - it seems silly to look at raw margins.

For example, let's look at PHX - their scoring margin is +3.42, the Blazers are at +3.16 - but if you look at the offensive efficiency per 100 possessions minus the defensive efficiency per 100 possessions - you will see that the Blazers and PHX are exactly at +4 per 100 possessions. In this case, it seems that PHX's faster pace gives it an artificial advantage as far as scoring margin is concerned - simply because they play an extra 7.5 possessions per game, on average.

 

[Minstrel]

How would pace have an effect on the scoring margin, though? Or maybe I am misunderstanding. If we play at an extremely slow pace, and only have 85 possessions per game or something, our opponent will have the same. So if we win by 10, our scoring margin is that +10. If GS gets 100 possessions, their opponent will as well, or close to, and so if they win by 10, they will also be at the same +10. Their faster pace didn't necessarily help them to achieve that margin as far as I can see, unless you are thining that a team that plays at a higher pace ultimately will have more blowouts and higher scores, and thus effect that number.

If you have an advantage or disadvantage in ability, more possessions should translate to a bigger difference in score. For example, if a team is slightly better than another...over 20 possessions, there may be little to no difference in their scoring against one another, but over 1000, you'd expect the slightly better team to have a solid lead...growing even larger after 1,000,000 possessions (if fatigue and loss of interest weren't issues ).

Greater pace creates more possessions, and thus more chance to separate yourself if you're better.

I think technically point differential should be normalized to pace, but I'm not sure the differences in pace add up to enough over a game to make a big difference in point differential. But maybe it does over a season's worth of games.

 

[bobf]

I don't think they should be pace adjusted. Take an exteme exaggeration illustrate.... Suppose two teams A and B would be equal at +3.0 if they played at the same slow pace. But suppose team A actually plays at 2x the pace. So towards the end of the game team A will typically have a much larger lead (average of nearly +6.0 vs +3.0). That larger lead will clearly cause a much higher win %. If you pace adjusted then they'd be considered equal.

EDIT: Hmmm... a second thought is what would happen if these two teams played each other... then they would be equal!

 

[Minstrel]

I don't think they should be pace adjusted. Take an exteme exaggeration illustrate.... Suppose two teams A and B would be equal at +3.0 if they played at the same slow pace. But suppose team A actually plays at 2x the pace. So towards the end of the game team A will typically have a much larger lead (average of nearly +6.0 vs +3.0). That larger lead will clearly cause a much higher win %. If you pace adjusted then they'd be considered equal.

I think you're begging the question with the bolded assertion. Would having a better point differential due to greater pace actually lead to superior winning percentage? To me, in your example, you have two equal teams in "true talent," but one appears better because they play faster and so their talent advantage is more apparent in the score.

 

[RR7]

Take an extreme case, let's say that there is a team that if it slows the game to 2 possessions is guaranteed to win by 2 points, no matter what. Compare it to a team that has 10 possessions per game, wins 50% of the time, but their scoring margine is +4 - the scoring margin stat gives you a prediction that the faster team that does not win as much is going to do better than the other team. Seems rather silly to me.

But if a team is winning by two points every time, theoretically, their "odds" or chances or whatever of winning WOULD be less than a team winning by 4 points every time. It's not meant to show who IS winning, but who is LIKELY to win. If your scoring margin is 2, then there's the easy chance the opponent hits a shot, and then it is a tie game, giving you the potential to lose. At 4, that doesn't exist. So if you continually outscore opponents by 4 points per game, you are more likely to win a game than if you "only" outscore them by 2.

 

[hasoos]

There is no need to normalize it for pace. It is a bottem line stat, how much do you win by. Nothing more, nothing less. IMO most teams that run a lot also play shitty defense, so it is already normalized.

 

[Minstrel]

But if a team is winning by two points every time, theoretically, their "odds" or chances or whatever of winning WOULD be less than a team winning by 4 points every time. It's not meant to show who IS winning, but who is LIKELY to win. If your scoring margin is 2, then there's the easy chance the opponent hits a shot, and then it is a tie game, giving you the potential to lose. At 4, that doesn't exist. So if you continually outscore opponents by 4 points per game, you are more likely to win a game than if you "only" outscore them by 2.

What do you think is the consequence of that then?

That every team with a talent advantage should be playing at the fastest pace possible, in order to create a larger score disparity and thus leave less to luck?

Or that playing at a faster pace is actually part of talent...there is no such thing as "equally talented" teams that play at different paces...players who play at greater pace are superior to equally good players who play at slower pace?

Or is there another option that I'm missing?

 

[bobf]

I think you're begging the question with the bolded assertion. Would having a better point differential due to greater pace actually lead to superior winning percentage? To me, in your example, you have two equal teams in "true talent," but one appears better because they play faster and so their talent advantage is more apparent in the score.

The two teams in my example are equal strength, i.e. equal point diff per possession. So when they play each other, it's 50/50. But when they play lesser teams (and average teams are lesser), the faster pace team will have more possessions over which to exaggerate their advantage. So they will have a larger lead near the end of the game. Clearly, the team with the larger lead near the end of the game will win more (to me this needs no proof).

Here is what my method would be for comparing two teams in a head-to-head matchup. You can pick holes in it.

1. Normalize both point diffs to per possession. This gives the absolute strength of each team.
2. Subtract to get the advantage per possession. This gives the relative strength of the two teams. If this number is 0 it matters not what pace the teams play. If it's positive then team 1 wants to play more possessions.
3. Estimate the number of possessions when those teams play*
4. Multiply (2) x (3) to get the expected game point diff.
5. Predict win % from (4)

For the two teams in my example, this would come out to 50/50. But apply this method for each team against the league average, the faster pace team would have a higher win %.

*Estimated possessions = Team 1 Average Poss + Team 2 Average Poss - League Average Poss

 

[bobf]

That every team with a talent advantage should be playing at the fastest pace possible, in order to create a larger score disparity and thus leave less to luck?

Yes.... as long as playing at a faster pace did not hurt their advantage per possession.

 

[andalusian]

But if a team is winning by two points every time, theoretically, their "odds" or chances or whatever of winning WOULD be less than a team winning by 4 points every time. It's not meant to show who IS winning, but who is LIKELY to win. If your scoring margin is 2, then there's the easy chance the opponent hits a shot, and then it is a tie game, giving you the potential to lose. At 4, that doesn't exist. So if you continually outscore opponents by 4 points per game, you are more likely to win a game than if you "only" outscore them by 2.

That is not the example I gave you. I gave you an example of a team that always limits the game to 2 possessions - one they have, one their opponent has, and because they have "perfect" offense for 1 possession, they always score and because they have "perfect" defense for 1 possession - they always stop - so they win every game by 2 points.

At the other extreme, you have a team that plays .500 or .600 ball - but because they play fast, they have a higher win margin - of 4 points per game.

In this (absurd to illustrate a point) example - the win margin is not a good prediction - because it ignores efficiency.

My opinion is, that if you are going to use win-margin as a predictive stat - it really needs to be corrected to possession - because some teams excel in a faster pace (See PHX, when Porter tried to slow them down they did not do well, their biggest mismatch is Nash running very fast and making good decisions) and some teams excel in a slower pace (See Portland, where minimizing possessions and turn-overs and using their advantage of rebounding and superior half-court execution is their ace in the hand) - but, in reality, the difference between the offense/defense of both those teams, per possession - is the same.

Seems to me like the use of that "element" in predictive formulas is biased when you ignore pace.

 

[blazerboy30]

Make me down for one that think's point differential should be adjusted by pace. Team's are likely (or should be likely) to play at a pace that is suited to their players to maximize their winning percentage. The point differential, adjusted for pace, should be an indicator and help estimate for the future, how well they do so.

I'd be curious to see (since I don't have time to do it myself) what the average point differential for the top teams was in the early 90's, when the scores were routinely in the 120's, compared to the top teams today. If there is a significant difference in average point differentials, this would add to the case to adjust by pace.

 

[andalusian]

1. Normalize both point diffs to per possession. This gives the absolute strength of each team.
2. Subtract to get the advantage per possession. This gives the relative strength of the two teams. If this number is 0 it matters not what pace the teams play. If it's positive then team 1 wants to play more possessions.
3. Estimate the number of possessions when those teams play*
4. Multiply (2) x (3) to get the expected game point diff.
5. Predict win % from (4)

I have a problem with this in general, and the problem is as follows. I am using Hollinger's description as a starting point: (From http://insider.espn.go.com/nba/insider/columns/story?columnist=hollinger_john&page=Rankings-Intro )

One of my goals was to create a system that told us more about a team's quality than the standings do.

So instead of winning percentage, these rankings use points scored and points allowed, which are better indicators of a team's quality than wins and losses.

This might not sound right at first, but studies have shown scoring margin to be a better predictor of future success than a team's win-loss record. Thus, scoring margin is a more accurate sign of a team's quality.

That explains why, for instance, three seasons ago the Spurs ranked ahead of the Mavericks even though they had won nine fewer games -- San Antonio's scoring margin was superior. That ultimately proved to be prophetic, as Dallas lost in the first round of the playoffs while the Spurs won the championship.

I am willing to accept that there is enough data and evidence to support the idea that scoring margin is a better predictor than record, and his formula uses both scoring margin and SRS for normalization (I have no idea what the exact "constants" he uses for the formula are, but that seems to be the basis of it) in addition to recent performance (scoring margin and SRS of the last 25 games) as well as upcoming schedule (including home/away games).

Because the pace in the NBA is not that different from the slowest team to the fastest one - about 11-12% between the fastest (GSW) and the slowest (POR) - I can understand that ignoring it will not make a huge difference, but still, it seems that a way to "predict" the outcome of any game based on what we know so far (Pace, Scoring-margin, efficiency) would be rather easily done by using a formula that calculates a predicted pace

pP = (pA + pB) / 2.

Assuming that for each team X we know the defensive efficiency (dX) and offensive efficiency (oX) - we can assume the score to be

Score per possession A = (oA +dB) / 2
vs.
Score per Possession B = (oB + dA) / 2

We can now calculate the estimated score for the game by applying these numbers to pP possessions (with some constants that estimate home court advantage, I guess).

So - the idea that a higher scoring margin for a team whose offensive/defensive efficiency is not better just because it has more possessions is rather artificial.

I can understand the argument that if a team has a positive offensive/defensive efficiency allow it to build a bigger margin and avoid "close games" - but likewise, the opposite is true - where you play a team that is better than you - the close game allows you to steal some games because of luck/being hot at the right time etc...

So, in general, seems to me that a scoring margin normalized for pace makes a better parameter for these prediction formulas.

 

[bobf]

In this (absurd to illustrate a point) example - the win margin is not a good prediction - because it ignores efficiency.

Yes, but it's also wrong to only look at efficiency and ignore pace. If you can wade through the details, my "method" addresses both parts.

Basically I claim that win % for a two team matchup is a function of relative-strength x combined pace. Relative strength (efficiency) has to be pace-adjusted as you point out. But once you have relative strenth you have to factor pace back in, because more possessions enhances the winning chances of the stronger team (except in your example where the team is perfect and one possession is enough to guarantee a win).

 

[andalusian]

Basically I claim that win % for a two team matchup is a function of relative-strength x combined pace. Relative strength (efficiency) has to be pace-adjusted as you point out. But once you have relative strenth you have to factor pace back in, because more possessions enhances the winning chances of the stronger team (except in your example where the team is perfect and one possession is enough to guarantee a win).

The method I presented above includes pace as well, and I think I addressed that claim that more possessions enhance the winning chances when we discussed what happens in close games against teams whose relative strength is slightly better.

So, overall, it seems to me that the use of offensive-defensive efficiency and multiplication by pace to predict a score is much better than using the raw scoring margin to plug into the predictive formula.

My specific argument against your method is with step 5 - it is very vague. Basically, your steps 1 to 4 agree with me that scoring margin needs to be adjusted for pace - as you basically do it in #1 .

Also - it seems to me that basically, if team A has a higher relative strength - this wants more possessions, but if team B has a lower relative strength - it wants to limit the possessions, but again, if these teams have the same relative strength (as Portland and PHX do, this year) - I am not sure that the team with the higher pace is really more likely to end with a better win%

It would be interesting to see if this really happens like that by checking large volumes of data.

 

[bobf]

I have a problem with this in general, and the problem is as follows. I am using Hollinger's description as a starting point: (From http://insider.espn.go.com/nba/insider/columns/story?columnist=hollinger_john&page=Rankings-Intro )



I am willing to accept that there is enough data and evidence to support the idea that scoring margin is a better predictor than record, and his formula uses both scoring margin and SRS for normalization (I have no idea what the exact "constants" he uses for the formula are, but that seems to be the basis of it) in addition to recent performance (scoring margin and SRS of the last 25 games) as well as upcoming schedule (including home/away games).

Because the pace in the NBA is not that different from the slowest team to the fastest one - about 11-12% between the fastest (GSW) and the slowest (POR) - I can understand that ignoring it will not make a huge difference, but still, it seems that a way to "predict" the outcome of any game based on what we know so far (Pace, Scoring-margin, efficiency) would be rather easily done by using a formula that calculates a predicted pace

pP = (pA + pB) / 2.

This formula isn't right. I thought that should be it too but it's not. It should be

pP = (pA + pB - pLeagueAverageExcludingTeamsAandB)

Suppose there are three teams in the league with these averages:
pA = 100
pB = 90
pC = 80

In head to head matchups, your formual yields
A vs B = 95 possesions
A vs C = 90 possessions
B vs C = 85 possesions

Aggregated this does not come out right:
Team A averages only 92.5 which is way too low.
Team B averages 90 possesions as they should
Team C averages 87.5 which is way too high.

My formula yields
A vs B = (100 + 90 - 80) = 110
A vs C = (100 + 80 - 90) = 90
B vs C = (90 + 80 - 100) = 70

Aggregated comes out right:
Team A averages 100
Team B averages 90
Team C averages 80

Assuming that for each team X we know the defensive efficiency (dX) and offensive efficiency (oX) - we can assume the score to be

Score per possession A = (oA +dB) / 2
vs.
Score per Possession B = (oB + dA) / 2

We can now calculate the estimated score for the game by applying these numbers to pP possessions (with some constants that estimate home court advantage, I guess).

I agree.

So - the idea that a higher scoring margin for a team whose offensive/defensive efficiency is not better just because it has more possessions is rather artificial.

I agree. But once you got the two team's effeciency you have multplied by the expected number of possesions. So pace matters.

I can understand the argument that if a team has a positive offensive/defensive efficiency allow it to build a bigger margin and avoid "close games" - but likewise, the opposite is true - where you play a team that is better than you - the close game allows you to steal some games because of luck/being hot at the right time etc...

Yes. A fast paced team (even a good one) is hurt by their pace when they play an even better team.

So, in general, seems to me that a scoring margin normalized for pace makes a better parameter for these prediction formulas.

I think both numbers are flawed. I think you've spelled out the ideal method, execpt for your combined pace calculation.

What kind of prediction are you talking about?
1. Who will win a two team matchup like a playoff series
2. What will a team's record be against the league

In the first case I'd go with the normalized number because you really need to know the true relative stength of those two teams.

In the second case I'd go with the pure point differential. Because on average that team is playing an average team. And your formula against an average team boils down to point differential.

 

[bobf]

The method I presented above includes pace as well, and I think I addressed that claim that more possessions enhance the winning chances when we discussed what happens in close games against teams whose relative strength is slightly better.

So, overall, it seems to me that the use of offensive-defensive efficiency and multiplication by pace to predict a score is much better than using the raw scoring margin to plug into the predictive formula.

Our methods are the same except for how to compute combined pace of two teams.

My specific argument against your method is with step 5 - it is very vague. Basically, your steps 1 to 4 agree with me that scoring margin needs to be adjusted for pace - as you basically do it in #1.

Yes, I left it vauge to avoid further complication. What's clear is that the higher the expected point diff between two teams, the higher the win %. The exact relationship would not be linear though, it would be some kind of S curve.

Also - it seems to me that basically, if team A has a higher relative strength - this wants more possessions, but if team B has a lower relative strength - it wants to limit the possessions, but again, if these teams have the same relative strength (as Portland and PHX do, this year) - I am not sure that the team with the higher pace is really more likely to end with a better win%

Since both teams are equal stength and above average strength, the team with higher pace should win more. If both teams were below average strength, the team with slower place would win more. Against elite teams (better teams) the Blazers with slower pace should fare better. But against lesser teams the Suns should fare better. There are more teams worse than the Suns/Blazers than better. So overall the Suns should have a higher win % due to their pace, not their strength.

It would be interesting to see if this really happens like that by checking large volumes of data.

It would.

 

[andalusian]

Since both teams are equal stength and above average strength, the team with higher pace should win more. If both teams were below average strength, the team with slower place would win more. Against elite teams (better teams) the Blazers with slower pace should fare better. But against lesser teams the Suns should fare better. There are more teams worse than the Suns/Blazers than better. So overall the Suns should have a higher win % due to their pace, not their strength.

You know what, given enough consideration - I am willing to accept this argument logically. Still would love to see the data to find out if it really translates to this kind of behavior in "real life".

 

[bobf]

You know what, given enough consideration - I am willing to accept this argument logically. Still would love to see the data to find out if it really translates to this kind of behavior in "real life".

You know, I think it's not going to make a whole lot of difference either way. Take your Portland vs Phoenix matchup at (made up) +3.0 vs +3.3 where they are equal in true strength but Phoenix is inflated by a 10% faster pace. If you fail to normalize for pace, you end up with Phoenix having a spurous +0.3 advantage over Portland in a head to head matchup. That amounts to less than 1 win per season. So a 1% error. You predict Phoenix has 51/49 advantage instead of 50/50. Not much difference. Maybe that's why Hollinger didn't worry about pace adjusting or not.

 

[andalusian]

You know, I think it's not going to make a whole lot of difference either way. Take your Portland vs Phoenix matchup at (made up) +3.0 vs +3.3 where they are equal in true strength but Phoenix is inflated by a 10% faster pace. If you fail to normalize for pace, you end up with Phoenix having a spurous +0.3 advantage over Portland in a head to head matchup. That amounts to less than 1 win per season. So a 1% error. You predict Phoenix has 51/49 advantage instead of 50/50. Not much difference. Maybe that's why Hollinger didn't worry about pace adjusting or not.

I can see it - but if we expand this argument to talk about half a season (so we have some "relevant data" to begin with to get your initial data), even a 0.1 difference per game - can translate to 4 wins over-all... might still be worth doing it.

 

[bobf]

I can see it - but if we expand this argument to talk about half a season (so we have some "relevant data" to begin with to get your initial data), even a 0.1 difference per game - can translate to 4 wins over-all... might still be worth doing it.

A long time ago I did a study of the relationship between team point differential and seasons wins minus losses. I took about 10 or 15 seasons of data and did a scatter plot in Excel. The plot looked farily linear so I had Excel fit a line to the data. It turned out that each point was worth about 2.5 extra wins (something around there don't remember exactly). So 0.1 points is not going to result to 4 wins over a season in normal cases.