NFL Def. vs. Off as predictor of score

I'm working on calculating Strength of Schedule based on the season to date Points Scored (PF) and Points Allowed (PA) for each team and the opponents each has played.

People always say "Defense wins Superbowls". What affects a team's score more: their own offense or the defense they are playing against? How would you weight the importance of defense (PA) vs. offense (PF)?

50/50? 60/40?

My guess is it depends on whether a team performs better offensively than defensively but I am interested in hearing thoughts.

I did something similar a few years ago.  I did it for a couple of years but never for a full season because it just didn't work that well.  But what I did was to look at the last 3 games each team played.

I took the total of team A's points in its last 3 games and divided by 3 for an average of what team A scores.

Then I took the total of team B's total points given up in its last three games and divided by 3 for an average of how many points team B gives up.

Then I add team A's average points scored to team B's average points given up and divide by 2 to project team A's score in the upcomeing game.

Then I do just the opposite to calulate team B's average score.

Example:

3 weeks ago Team A scored 21 points
2 weeks ago Team A scored 17 points
1 week ago Team A Scored 10 points.

21 + 17 + 10 = 48 / 3 = 16 points

3 weeks ago Team B gave up 34 points
2 weeks ago Team B gave up 14 points
1 week ago team B gave up 24 points

34 + 14 + 24 = 72 / 3 = 24 points

16 + 24 = 40 / 2 = 20 points as Team A's projected final score v Team B this week

Then you just reverse and add Team B's last 3 week's points scored divided by 3
and add Team A's last 3 weeks points given up divide by 3.

Add Team  A's average score to Team B's average points given up and divide by 2 for Team B's projected score for this week.

In my experience with this system, sometimes it is very accurate and sometimes it is not even close.  Overall, it did not predict enough games close enough to keep my interest.

OK, here's the Green Bay v Cleveland Game

Green Bay points scored last 3 weeks

3 weeks ago GB scored 36 points
2 weeds ago GB scored 23 points
last week GB scored 26 points

36 + 23 + 26 = 85 / 3 = 28.3 points

Cleveland gave up this many points last 3 weeks

3 weeks ago Clev gave up 23 points
2 weeks ago Clev gave up 3 points
last week Clev gave up 27 points

23 + 3 + 27 = 53 / 3 = 17.66

28.3 + 17.66 = 45.08 / 2 = 23 projected points for GB

Cleveland points scored last 3 weeks

3 weeks ago Clev scored 20 points
2 weeks ago Clev scored 6 points
last week Clev scored 14 points

20 + 6 + 14 = 40 / 13.3

Green Bay gave up this many points last 3 weeks

3 weeks ago GB gave up 17 points
2 weeks ago GB gave up 30 points
last week GB gave up 0 points

17 + 30 + 0 = 47 / 3 = 15.66

13.3 + 15.66 = 28.96 / 2 = 14 projected points for Cleveland

Projected score is Green Bay 23  Cleveland 14

I didn't use the entire season as an average because to my way of thinking a team can be totally different in December than they were in September.  Therefore, the more time that passes between the current and the past , the less relevant a game is. 

On the other hand, 1 game could just be a fluke, so I didn't feel comfortable with only using last week's scores for comparison, and 2 weeks did not seem like enough data either, so I went with a 3-week running average.

That being said, the system didn't seem to give enough consistent results.  I didn't try it with just one week or two weeks or four weeks or anything else.  There may be a good average to use along with other filters, e.g., home-field, heads up play, and others.

I always felt that this theory had some potential, but I haven't messed with it in several years.

Oh, with respect to the Green Bay v Cleveland game, the currently posted point spreads range anywhere from 6.5 to 10.  Therefore Green Bay is the pick, especially at -6.5.

Also the projected combined total points is 37 and the posted o/u is 41-42.  The under is the project bet.

Another thought is that this system looks at the total team's performance including offense, defense, and special teams (remember Denver's 2 special teams run backs for touchdowns against San Diego last Monday?).

Thanks for the responses Jeff. It seems I'm going through the same progression you did when you fiddled with this years ago.

I've done almost the exact same math for this season but also broke it down by away/home PF and away/home PA to account for home field advantage.

When I run the model off the first 4 weeks of data for week 5 and then the first 5 weeks of data for week 6 it predicts about 50% of the games correctly. Not from a final score perspective, but from a point differential predicted vs. spread. Its only 2 weeks of testing but I'm intrigued enough to explore more.

My theory is that I can boost my win percentage well over 50% by avoiding games 1-2 games per week where roster changes, injuries or other factors may cause future results to be much different from previous.

I didn't run the numbers for any other game except for the Packers / Browns game.  Of course this little "system" looks very good for the single game posted as it picked the winner, the winner against the spread, and the winning over/under bet.  It may not look so good if all the games for the week were calculated.

Eagles v Skins Week 7

Projected score Eagles 21 Skins 7

Eagles are currently -7 to -9 favorite and the o/u is 37-38

That makes the Eagles and under the plays according to system criteria

OK, hold your horses.  Math error

projected score is Eagles 21 Skins 13

that makes the projected point spread "too close to call"

that makes the over/under a slight lean to the under

sorry if anybody used the above info, I corrected within only a few minutes after posting.  I don't think anybody is paying attention to this thread anyway.

Just me

Week 8 projections (the method for the scores is explained above in posts #2 and 3). This system is unproven, so only use this information at your own risk.

projected score Denver 24 Balt 20

Balt is a -4 point favorite

o/u is 40 ½

My opinion:

Denver is a strong play for this early opening line. This looks like the wrong team is favored.

the o/u is a slight lean to the over; take a pass on this one

projected score Chicago 22 Clev 19

Chicago is a -13 ½ point favorite

the o/u is 40 ½

My opinion:

Even though Cleveland is an early +13 ½ point dog, I would not bet on them on the road against a team who has showed they can play (the Bears beat the Steelers in Pittsburgh) and Cleveland hasn't showed me too much; The Bears projected score does not justify betting on them giving -13 ½ points; take a pass on this one

the o/u is to close to call as the projected final score is 41

projected score is Dallas 22 Seattle 20

Dallas is a -9 ½ point favorite

the o/u is 45 ½

My opinion is that even though Dallas is a home favorite, with the projection as close as it is, I am not laying the points

The o/u is a slight lean to the under but it is too close for me; take a pass

projected score is Indianapolis 29 San Francisco 16

Indianapolis is a -11 point favorite

the o/u is 45

My opinion is that the Colts are putting up a lot of points and the projected score justifies lay -11 on them as a home favorite.

The projected o/u matches the early book so I am going to pass on the o/u.

Projected score is NY Jets 27 Miami 25

The Jets are a -4 point favorite

the o/u is 40

It may be my imagination, but as I recall, these two teams seem to play very tight games. I don't want to bet on either one to win.

The o/u is only at 40 and the projected combined score is a whopping 52 points. The over looks good.

Projected score is Jacksonville 31Tennessee 17

The Titans are a -3 point favorite

The o/u is 43 ½

My opinion:

Even though the Titans are favored by the books as a home favorite, the projected score says Jax will win by 14. These are 2 bad teams. I don't want to bet on either one of them. For those who are interested in John Morrison's system, the Titans are a [C] bet. I don't like to go against JM, but I didn't take the Titans on either of their first two system bets and not betting on the Titans saved me some money. I don't want to bet on a home dog at Tennessee either so I am not going to bet on the Jags.

The o/u looks like a small lean on the over as the projected score is 48. Again, I don't like either of these teams this season; I'll pass

projected score is San Diego 30 Oakland 17

The Charges are a -16 ½ favorite

the o/u is 42 ½

My opinion:

That spread is too darn big to bet on the Chargers who are likely to win this game. The moneyline will probably be too big as well.

The o/u on the other hand looks to be a lean to the over.

Projected score is Packers 25 Vikings 20

The Packers are a -3 point favorite

the o/u is 48

My opinion:

The Pack is a -3 point home favorite, but I wouldn't bet against the Vikings with Bret Favre at QB. On the other hand the Vikings are a projected loser, so I will not bet this game.

The projected o/u is too close to the posted line; I'll pass.

Projected score is New Orleans 30 Atlanta 26

New Orleans is a -9 point favorite

the o/u is 53 ½

My opinion:

New Orleans will probably keep their perfect record in tact this week, but I will not lay -9 points, especially with this projection.

The o/u is a lean to the over, but I will pass.

I like Denver +4

Indy -11

The over 40 on the Jets and Dolphins

As I write this, the rest of the early lines are not out yet. These lines are very early (Sunday Oct 25) and are likely to change. If I get a chance, I'll post the rest of the games.

I just got me some of that Denver +3.  If you want Denver, get 'em while the gettn's good.  I think that the number will drop by Sunday.  I wouldn't be surprised if they don't end up being the favorite.  Hey, Baltimore is tough at home, and Denver may be due for a let down.  But they have been solid, and this is going to be a rare opportunity to get them as the underdog. 

More week 8

Projected score is Arizona 23 Carolina 16

The Cardinals are a -9 ½ point favorite

the o/u is 42 ½

My opinion:

The Cards are at home playing a very mediocre Carolina team. However, the projection rules this game out even though I think Arizona will cover.

The o/u is a slight lean to the under, but I'm going to take a pass.

Projected score is Houston 22 Buffalo 17

The Texans are a 3 ½ point favorite

The o/u is 40 ½

My opinion:

This is one where I wouldn't be surprised if the final score is close to the projection; that is too close for me to pick. I'm going to skip this game. If you must, the Texans are the way to go.

The o/u is a slight lean to the under. Again, too close to call. Pass on this

I'm not comfortable enough to make any picks on these games. If I was just getting my arm twisted and I had to make a pick I would lay the 3 ½ and take the Texans; it would not make me happy though.

I'm still waiting for the lines to come out on the rest of the games.

Actual final score Eagles 27 Skins 17
Projected final score Eagles 21 Skins 13

actual final point spread 10
projected final point spread 8

In my experience, this out come is not unusual.  That is, it projection looks fairly close to the actual outcome.  However, it is also my experience that the projections can be way off.

Yes, the projections don't really factor in strength of schedule. If a team has played pretty weak defenses their PF looks good. And similarly if they have played poor offenses their PA ends up being inflated.

I've been trying to work strength of opponents into the PF / PA prediction. Here's what I've done so far:

1) Calculated the average PF / PA for each of a team's opponents outside of the games played vs. the team being analyzed.

2) Created an index of these opposing team PF / PA values compared to the league average. This is supposed to tell me how good a given team fared against an opponent compared to how this opponent played against other teams.

3) Used the results of #2 to adjust each team's PF / PA up or down. So a team gets +PF if they have played opponents that did not allow as many points against other teams.

4) Calculate final scores for each game based on the same methodology you're using above but using the adjusted PF / PA values instead.

Forgot to add that I'm breaking it all down by home vs. away as well. It makes everything I'm doing twice as complicated but seems valuable since the average PF / PA varies greatly depending on location of the game.

For example, last night's MNF game:

PHI
23.5 PF (away vs. 27.2 overall)
11.5 PA (away vs. 19.8 overall)

WAS
10.3 PF (home vs. 13.2 overall)
11.3 PA (home vs. 16.0 overall)

PHI Opponents (CAR, NO, KC, TB, OAK)
23.1 PA (home since we are looking at PHI PF away)
20.7 PF (home since we are looking at PHI PA away)

Index to league average
1.17 PA (23.1 divided by 19.8 - the league average PA at home)
0.92 PF (20.7 divided by 22.6 - the league average PF at home)

WAS Opponents (NYG, STL, DET, TB, CAR, KC)
29.9 PA (away since we are looking at WAS PF home)
18.2 PF (away since we are looking at WAS PA home)

Index to league average
1.32 PA (29.9 divided by 22.6)
0.92 PF (18.2 divided by 19.8)

Adjustments to PF / PA

PHI
20.1 PF (23.5 divided by 1.17)
12.5 PA (11.5 divided by 0.92)

WAS
7.8 PF (10.3 divided by 1.32)
12.3 PA  (11.3 divided by 0.92)

MNF Prediction:

PHI
(20.1 + 12.3) / 2 = 16.2

WAS
(7.8 + 12.3) / 2 = 10.1

PHI 16 or 17 -  WAS 10

Similar to your math this one looks too close based on the line at kick-off.

Last night's game was a good example because it shows how PHI's PF was actually handicapped by WAS's poor PA. WAS is only giving up 12.3 points a game . . . but against horrible teams with bad offenses. In a way WAS's PA looks good because it's below the league average PA but we know their decent defense is only half of the story. The teams they have played simply cannot put points on the board (either).

My next step will be to look at these adjusted PF and PA values against a common opponent or maybe to look at how each team's opponents fared against their opponents.

What about o/u?  Do you want your projections to predict o/u as well, or for point spreads only?

Even More week 8

Projected score is NY Giants 22 Philadelphia 25

The Eagles are a -1 point favorite

the o/u is 44

My opinion:

Simply put, this one is too close to call under system criteria. I do think the Giants will win, but I will not be wagering

The o/u is a virtual toss up. I'm going to take a pass.

Still no odds on the Rams and Lions. I don't know why not.

I'm mostly interested in which side has the most value. I do appreciate you comments on how the predictions lay out for o/u plays though.

I haven't checked for a few days, but the books finally posted the line for the last game this week.

Projected score Detroit 24 St. Louis 23

Detroit is a -4 point favorite and the o/u is 43 ½

My opinion:

Two bad teams. I don't like betting on these kinds of games. The slight lean goes to the rams and the over, but I am staying away from this game.

...so how is this projections working on the win-loss actual?

 

what is the w-L record so far????

I adjusted my methodology to change the way strength of schedule is factored in for defense:

Instead of indexing performance to the league average (see post #17 above) I'm now comparing a team's PA to their opponents' PF in other games. Essentially, "how good was a team's defense compared to other teams that have played the same set of opponents."

This greatly simplifies the calculations and the results are better too. Since Week 5 the model has gone 33-21 su (61%) and 34-22 ats (59%). Some of the predicted scores are real wacky though just like jeff_kent said they'd be.

Projected score Miami 20 New England 30

NE is -12 point favorite and the o/u is 48.5

no play for me

Projected score Green Bay 28 Tampa Bay 14

GB is a -10 point favorite and the o/u is 43.5

no play for me; However, for those who follow JM NFL, TB is an [A] bet

Projected score Kansas City 21 Jacksonville 17

Jacksonville is a -6 point favorite and the o/u is 42.5

I don't like to bet on bad teams, but you may want to consider a small bet on KC

Projected score Houston 19 Indianapolis 23

Indianapolis is a -9 favorite and the o/u is49

no play for me

Projected score Cincinnati 23 Baltimore 21

Cincinnati is a -1 point favorite and the o/u is 43

no play for me, but it looks like the books used this system to establish their odds

Projected score Arizona 24 Chicago 18

Chicago is a -3 point favorite and the o/u is 45.5

you might consider a small play on the Cards

Projected score Atlanta 22 Washington 21

Atlanta is a -10 point favorite and the o/u is 41

you may consider a small bet on Washington and a small bet on the under

Projected score Seattle 22 Detroit 16

Seattle is a -9.5 point favorite and the o/u is 43.5

I am not going to bet on this game

Projected score New Orleans 32 Carolina 27

New Orleans is a -14 point favorite and the o/u is 52

based on the projection, the Saints is a “no play” but you may want to play them anyway. The over/under here looks like the best, and possibly, the only bet based on this system. The projection is 7 points more than the o/u and the final score could go even higher. Play the over 52

Projected score San Francisco 25 Tennessee 21

The 9ers are a -4 point favorite and the o/u is 41

this is another example of how close this system is to the books, but the over bet is a “lean”

Projected score San Diego 33 NY Giants 20

The G-men are a -4 point favorite and the o/u is 47.5

hmmmm, it looks like the Chargers are the bet given their projection is 13 points higher than the Giants and the Giants are a -4 point favorite; that means between the Bolts being a +4 dog plus their 13-point. projected cushion, that makes them a 17 point play. I guess the system recommends this play. I will only make a small play on the Chargers. Oh, by the way, the Giants are a John Morrison [A] bet this week.

Projected score Dallas 25 Philadelphia 22

Eagles are a -3.5 favorite and the o/u is 49

looks like a lean to the Cowboys and a lean to the over. Actually, I like the over as a small bet.

Projected score Pittsburgh 25 Denver 19

Denver is a -3 point favorite and the o/u is 39

The Steelers are coming off the buy week and Denver is licking its wounds after facing Baltimore's tuff defense. This has the makings of an upset or at least a lean for the road dog.

FINAL ANALYSIS the only play I like is the New Orleans v Carolina over 52