weighting the NFL schedule

[jets96]

Looking for ideas on how to weight the the season. I used to use week one as one point, week two as 2 pts ,and so on, as the season progresses to weeks 14 , 15, 16 and 17 are they that much more important then week 1 or 2. They say if you dont win in Dec , you dont play in Jan, am thinking of weighting them by month , say sept games I weight as 1 point, oct 2 pts ,nov 3pts and dec 4 pts.
Any ideas ,thank you in advance.

[KVB]

In my opinion the best way to define the difference between early season games and later games is to incorporate a recency effect.

Similar to exponential moving averages you want to find a way to give more recent performance a heavier rating. I jumped ahead though, making sure you give simply equal weighting can be important.

Until you start talking about playoff implications, be careful treating games too differently.

When you create your NFL line, you likely use season to date stats. In the NFL, it takes four weeks to really get something meaningful, but that's all it takes, and, in fact, you may not have to look past the last 4 games of any two teams to come up with something reasonable.

In basketball, which has more even scoring, I would tell you (and this isn't rocket science here) to make multiple lines... season to date, last 10 games, last 5 games, or something like that, then average them so that you are giving equal weightings to each segment, and not minimizing recent performance because of a long season.

For football I would sugggest the same, 4 weeks, 8 weeks, 12 weeks, season to date, for example.

This allows you to emphasize a teams recent performance and you can also see the trend of that performance as time passes. Knowing where a team has been, and where they are now, can give you an idea as to where they are headed.

Obviously we need the season to progress to fill in these blanks, but I think it's better in this case to address each team and their opponent individually in these different stages of the season as opposed to trying to find one stage that has a different set of importance than others.

You can do that, but your talking more about something like NCAA Basketball, where there becomes a preseason that settles into conference games later on. Those are different animals, or can be.

In the NFL important division games for teams can happen at any time during the season it only seems more urgent as playoff pictures develop. But we can't forget all those earlier games help develop that playoff picture.

The NFL is nojt the NBA, and can be volatile and really vary from one week to the next, and that is why I am shying away from suggesting caluclations based only on 2 week intervals, the NFL really does need 4 weeks.

[KVB]

...They say if you dont win in Dec , you dont play in Jan, am thinking of weighting them by month...

This is entirely team and situation and opponent specific. Teams will win in Dec with no hope of playing in January and some teams might not need to win and can cruise for some of the late weeks to get into the playoffs.

It's so situational, that it's tought to put a blanket number on the importance of those weeks for everyone.

Many decent NFL models work best when the best teams are involved and everyone is performing as expected, like good teams often do, so you may find that when the good teams are motivated later in the season that you will get more statistically predictable results, and less noise, but in the end all of that can just be noise in the NFL.

If you want to assign numbers or a weighting mulitplier to some situations, like true must win scenarios, that is reasonable, but late in the season only so many teams are in that scenario.

[jets96]

I appreciate the advice, in the nba i do break it down as you say, in the nfl not so much , but am going to have to start because it makes sense. In the NFL though , is my thinking completely wrong when i treat every week different as i stated above? Every week becomes more important the way i was doing it, but like i said, i dont think week 17 is 17 times more important then week one, even though its the same number am using for every team...i use a bunch of other variables with those week numbers.

[semibluff]

Be aware that home field advantage isn't a constant. In a pure heads-up data point rather than against the spread home field is a bigger league factor earlier in the season.

[jets96]

yes i think that is true, trust me i dont use 3 pts for every team , i think there is an advantage even for the worst teams in the nfl but it comes from the other team traveling and how they play on the road.

[KVB]

Be aware that home field advantage isn't a constant. In a pure heads-up data point rather than against the spread home field is a bigger league factor earlier in the season.

I hear what you're saying there. In fact, that can be true for a lot of adjustments.

It's like it's wide open early in the season then the competition and games slowly tightens down those metrics, making them less to work with.

It comes with "parity" for sure and without a doubt is part of the reasons lines often get "sharper and sharper" as the season progresses. Obviously not every line and game, but in general that is the way so many factors trend.

Some cappers thrive in the early part of the season, some cappers wait for the info and then don't even start playing until it gets tighter.

It's like a consolidation of metrics as the season progresses. It's true for the books too, I posted some charts last year showing the average error the books have on totals and spreads compared to the actualy scores and results. I'll try to dig that up here in a bit.

It's good info for the think tank.

Good thread, SBR could use this kind of stuff right now.

[jets96]

Taking KVB's advise ,am incorporating two model into one, using my model that uses a combination of last years stats and this years ,and the last three weeks.
First number will be for last years and this years stats, 2nd number will be just for the past 3 weeks.

cinn 9/4 ... cinn -6.75
n.o 10/4 no -7
tenn 7/1 ...tenn -4
wash 1 atl 3 ....atl -1
buf 15/3.5 buf -9.25
chi 6.5 /3.5 chi 5
cle .5/ minn 2 ... minn -0.75
mia 3/3 ...mia -3
lar 6.5/3.5 ...lar 5
sf -3.5..sea .5 ...sf -1.5
balt 3/pk ...bal -1.5
gb 7/5 ...gb -6
t.b 1.5/1.5 ....tb -1.5
lac 7 /2.5 ...lac -4.75

Now ill see how they close ..

[Waterstpub87]

You might consider something like exponential weighting https://en.wikipedia.org/wiki/Exponential_smoothing

I tend to, for most sports, generate weights so that by 50% of the season, the past 2 years data and the current data are roughly equal. I weight last year twice the weight of two years ago.

I would be cautious of how much you weight things in the short term. Why is week 17 more important than week 1? how much more? When I've incorporated like "Last 10 games" as a weight factor in baseball, it actually reduced the predictive power. Checked it again over the offseason, excel's regression module agreed. Have to check the same with NBA for the coming season.

Best of luckeither way. We are rooting for you Jets. Always good to see people trying to use their mind to get better.

[jets96]

waterstpub , i stopped reading after the first paragraph , way above my pay grade , i have great thoughts with a no math background. So am going to work with the above and see how it works out ...ill do what kvb said ...i keep doing what ive always done , but now ill add the last 4 games , last 8 games etc and divide them to find a better number ...
and thank you for kind words

[jets96]

dallas 5/3.....dallas -4
k.c 8 .....phila 3 .....k.c -2.5

[semibluff]

Taking KVB's advise ,am incorporating two model into one, using my model that uses a combination of last years stats and this years ,and the last three weeks.
First number will be for last years and this years stats, 2nd number will be just for the past 3 weeks.

cinn 9/4 ... cinn -6.75
n.o 10/4 no -7
tenn 7/1 ...tenn -4
wash 1 atl 3 ....atl -1
buf 15/3.5 buf -9.25
chi 6.5 /3.5 chi 5
cle .5/ minn 2 ... minn -0.75
mia 3/3 ...mia -3
lar 6.5/3.5 ...lar 5
sf -3.5..sea .5 ...sf -1.5
balt 3/pk ...bal -1.5
gb 7/5 ...gb -6
t.b 1.5/1.5 ....tb -1.5
lac 7 /2.5 ...lac -4.75

Now ill see how they close ..

dallas 5/3.....dallas -4
k.c 8 .....phila 3 .....k.c -2.5

What do these numbers actually mean and what are you using them to do?

[jets96]

the first number is MY LINE using part of last season and current season stats, the 2nd line is current season ...divided by 2 which gives me my lines for this week.

[jets96]

i teased cinn with n.o ...no look ahead for cinn here , they need this game vs jags, they cant let the jags come into their house being 0-3 with gb packers up next , ill take my chances with them laying the 1.5 ...n.o , i just think for the gents to come into n.o beat up playing the saints in their first home game is going to be a tough task for jones and company .....

[jets96]

kvb or anyone else ,if i were to use the difference from the closing line into my model, were would you think i should cap it at to alleviate garbage time ...example n.o beat the close by 38.5 pts .... I could use that number but i could cap it say at 21 for any game or i could use the 38.5 as well as cap it, not sure ...any ideas
thank you guys in advance and as this ideas come to light am going to post my lines just so you can follow along.

[semibluff]

the first number is MY LINE using part of last season and current season stats, the 2nd line is current season ...divided by 2 which gives me my lines for this week.

You can't average out NFL handicaps that way. If you had 1 line at +0 and another line at +4 the average would be around +3. This would be much easier to see if you were working with expected win percentages rather than handicaps. As it is you're counting all handicap points equally. In reality they're not.

[mtneer1212]

The biggest flaw in schedule weighting is injuries. If I don't have my starting corner because of injury for 8 weeks, or my QB is constantly hurt (i.e. Wentz) or my QB degrades over the season due to age (i.e. Roethlisberger), etc...... the league is dynamic and change of personnel and injury status make this type of modeling somewhat irrelevant.

[jets96]

i adjust for qb only , i dont think any one player other then the qb will impact the line that drastically, if multiple skill players are out , ill have to take a look at it more closely.

below is what kvb said ....do you not agree

In basketball, which has more even scoring, I would tell you (and this isn't rocket science here) to make multiple lines... season to date, last 10 games, last 5 games, or something like that, then average them so that you are giving equal weightings to each segment, and not minimizing recent performance because of a long season.

For football I would sugggest the same, 4 weeks, 8 weeks, 12 weeks, season to date, for example.

[jets96]

and schedule weighting might be different for everyone....how do you schedule weight ???? tell me , give me an example..
And i dont think its schedule weightings, its more week weighting in combination with other factors , like what was the teams record that they played ...last 8 , last 4 ...also in combination with pts off the closing line...i have a great model that wants to get better.

[mtneer1212]

i adjust for qb only , i dont think any one player other then the qb will impact the line that drastically, if multiple skill players are out , ill have to take a look at it more closely.

below is what kvb said ....do you not agree

In basketball, which has more even scoring, I would tell you (and this isn't rocket science here) to make multiple lines... season to date, last 10 games, last 5 games, or something like that, then average them so that you are giving equal weightings to each segment, and not minimizing recent performance because of a long season.

For football I would sugggest the same, 4 weeks, 8 weeks, 12 weeks, season to date, for example.

Vegas oddsmakers say that QB injury is the biggest adjustment, Loss of a center is the second biggest adjustment, with offensive and defensive linemen next, and adjustments are rarely made for RBs, and most WRs

[jets96]

Maybe its who you talk to at the time, one of my dearest friends is a sportsbook manager in Vegas and has been for many years, i think you can look at teams that win and am sure ALL OF THEM have a monster WR...the question is how much is he worth to the line...in the nba i might adjust 4 or 5 pts with lebron out ,in the nfl if the center is out and he is a great one ,how much would you adjust a line .5 ,1 pt ?

[jets96]

So am waiting on N.O -1.5 ....i think i have value in a bunch of games but am going to hook up minn and Indy in a 2 teams 6 pt tease.
waiting to see where atl goes but i still think ill tease them up with someone else once i start to see lines move.

[jets96]

combination last year and this year seatttle -4.5 ....last 4 games seattle -2.5
too much value on seattle ill take them +2.5 ....i guess ill buy the hook for the push which am sure isnt a good idea, but its better then a loss.
I think cle and lac are a solid 3 with all my ratings so ill throw a 6 pt teaser in with seattle and cleve as well.

luckl

[OldBill]

THE schedule is your gold mine this is where teams are are in trap games

1st one i spotted was Bengals 3rd str8 road @ jets and off biggest win in history @ ravens with div rival up next got hammered by the jets

as big fav -7.5

then this one Ravens hammer vikings home net game @ miami on TNF omg i told every one i could take miami

22-10 waxing by dolphins

Raiders omg - 5.5 VS bears trap game Broncos up next s/u loser 20-9

thats all i recall look at other teams schedules predict a win huge then loss ATS home or away if conf rival or div rival is up next