NPPD stat shows the 2010 team as the worst-ranked Skins team in the Snyder Era

[Oldfan]

MSF ~ What does ranking by NPPD really demonstrate?

If you took (theoretical situation disclaimer) two Redskins teams with identical records yet disparate NPPD and compared them, what would the NPPD ranking actually tell you?

The rankings would tell you nothing in that case, but a comparison of the NPPD stats would indicate that the one with the better stat was probably the better team.

Also, as NPPD is a better predictor of a team's ranking (defined as...? - see first question), than wins, then why claim it is a valuable stat based on a 0.92 correlation to wins? Correlating a tool's accuracy to a reference standard that has a less accurate measurement precision doesn't make sense to me. What am I missing?

You are thinking that "wins" in a 16-game sample is equal to "wins" in a much larger sample. The fact that it's the same word does not make them equally good as standards of measurement.

Think of it this way. The NPPD is a measurement tool forged in the hot fires of a large sample size of 3,000+ wins to grade an NFL team within .92 accuracy for the purpose of ranking it. The alternative, using wins to grade the team, is a poor tool because the small 16-game sample size involves too big a luck factor.

BTW - Happy Birthday OF!!! I'm glad it was a good one.

Thank you, Sir.

[MassSkinsFan]

Think of it this way. The NPPD is a measurement tool forged in the hot fires of a large sample size of 3,000+ wins to grade an NFL team within .92 accuracy for the purpose of ranking it. The alternative, using wins to grade the team, is a poor tool because the small 16-game sample size involves too big a luck factor..

So the 0.92 correlate is to wins over a span of years then. I see. I'm still not convinced that this makes sense mainly because I don't understand how the correlation works (that you can relate one measurement, based on one season, to a standard that is based on many seasons; this is a unit of measure issue for me). Looks like I have a bit of reading to do at some point...

[Oldfan]

Not surprising.... I didn't step in until you started contradicting yourself, silly!

What were the contradictions? Is it possible for you to explain them so I will understand?

It's a revisionist stat, dude. It's like saying, the guy who won the school election for class president got the most votes. Therefore....useless.

The statistic measures the team's performance. That's like saying that Charles won the election with 63% of the vote. Except that grading the teams is more useful.

Ah ha! Use your PRICELESS stat to pick winners against me next year. Then we'll see who is incomprehensible! LOL

I don't know about picking winners in football games, but I'll concede I'm no match for you when it comes to blowing wind.

---------- Post added July-7th-2011 at 01:31 PM ----------

So the 0.92 correlate is to wins over a span of years then. I see. I'm still not convinced that this makes sense mainly because I don't understand how the correlation works (that you can relate one measurement, based on one season, to a standard that is based on many seasons; this is a unit of measure issue for me). Looks like I have a bit of reading to do at some point...

I found the link for that .92 correlation for you to look at. However, it doesn't describe their method.

http://www.twominutewarning.com/correlations2.htm

[DC9]

Dude, this is a study that says the team that scores the most points wins. How does that help measure anything other than state obvious? How does that help you analyze your team for the future? How does that help you pick who the best team will be next season? What good is the stat, other than saying....yeah.... New England was pretty freaking good....they scored more points than a lot of guys...?

It doesn't take change into account. Changes in rules, penalties, opponents. Strength of schedule is a huge variable. This isn't something that you can use to measure how bad a team was compared to other teams in different seasons.

Really.....it's useless. Cool that you dig it, but I'm not rubbing peanut butter all over myself.

Enter: Your one-liner defense.....

[Oldfan]

Dude, this is a study that says the team that scores the most points wins.

That's just not true.

It doesn't take change into account. Changes in rules, penalties..

. Non-factor. The study covered years subsequent to major rule changes.

Strength of schedule is a huge variable.

As I said earlier, the stat could be improved by a SOS adjustment, but it isn't a huge factor in the NFL where parity reigns. There's only a 3-4 percent swing from the easiest to the hardest schedule.

[Est.1974]

the .92 correlation is pretty damn good. It certainly beats wins and losses. Although your comment suggests to me that some sort of adjustment using standard deviation might improve it.

I don't know what that adjustment would be, but there is just something about points differential that I find can be misleading.

Take the Packers, they ranked #2 is points diff' at +148. But, much like the Redskins, that stat is driven largely by only 4 games. Victories of 27,38, 28 and 28 again. And those are the games that will dictate their positive NPPD. They had ten games decided by a touchdown or less.

So, whilst points diff' and NPPD are may look at certain sample sizes, much of the sample is not making a meaningful contributing to the end result / Or more like a small element of the sample has too great an influence on the end result.

But maybe that is meaningless when applying this stat - Is it ?

[Oldfan]

I don't know what that adjustment would be' date=' but there is just something about points differential that I find can be misleading.

Take the Packers, they ranked #2 is points diff' at +148. But, much like the Redskins, that stat is driven largely by only 4 games. Victories of 27,38, 28 and 28 again. And those are the games that will dictate their positive NPPD. They had ten games decided by a touchdown or less.

So, whilst points diff' and NPPD are may look at certain sample sizes, much of the sample is not making a meaningful contributing to the end result / Or more like a small element of the sample has too great an influence on the end result.

But maybe that is meaningless when applying this stat - Is it ?[/quote']No, I don't think it's meaningless. I think your concern is valid. The stat need adjusting, but I'm not sure how to go about it.

One idea that crossed my mind is that, after the teams are ranked 1 - 32, the net points number should be dropped completely to allow the ranking to determine the average wins expected for that position. Example: The average wins for a team ranked #16 would be 8.00 --- the average wins expected for a team ranked #4 would be 12.50 (a guess).

Such a chart could be compiled by taking the average wins for a team ranked #1, and the average wins for a team ranked #32 over the past 25 years, then calculating the average wins for #2 through #31 as equal steps between the extremes.

This would eliminate the effect of those big numbers (I think).

[Est.1974]

No, I don't think it's meaningless. I think your concern is valid. The stat need adjusting, but I'm not sure how to go about it............

This would eliminate the effect of those big numbers (I think).

Sure is a tricky one, OF

Coming from another angle, do you want to totally eliminate the big numbers?

Whilst they distort the stats, perhaps teams than post 4 or 5 significant wins should see some recognition for that, as it demonstrates their ability to hang it on someone every 3 or 4 games. Something we're not used to, as I don't think the Redskins have posted 40 points in a game since putting 52 on SF about 6 years ago .

Maybe, as such a high % of games are 'close', and therefore not influencing the stat, you have to eliminate some of the closer results first, just to get some of the parity out of the way. Then you maybe exclude the top one or two larger margins, to get the freak result out of the way. Much like the Raiders 50 point win over Denver last year. Then even after taking those top ones away, teams like the Patriots & Packers still get some dues for their ability to post those numbers on a one in four basis.

I think their is a balance somewhere, between taking the marginal stuff away, eliminating the out of the ordinary results, yet still retaining enough variation to make the stat more than just a marginal number.

Don't know, that's just a ramble of mine probably headed into a cul-de-sac in afraid

But someone with a brighter mind may take something from it....

[Oldfan]

UKSF74,

Click on this link and check out a method already being used based on net points.

http://www.footballoutsiders.com/stats/teameff

They call it...

PYTHAGOREAN WINS represent a projection of the team's expected wins based solely on points scored and allowed.

They show the actual wins in another column to the left.

I don't know... those big numbers bother me. I'm not inclined to use a direct computation from the net points numbers.

[Est.1974]

OF - will check that out over the coming days, time is short at the moment but will get around to it.

Have those big numbers always bothered you, or was it something I said ?

EDIT -

Had a quick look at this - How the hell do they end up with the 5.9 Pyth Wins shown against the Redskins ? :dunce:

What am I missing from this principle when it is being being applied to the NFL. What else is in the calc ?

Pythagorean Theorem: The principle, made famous by baseball analyst Bill James, that states that the record of a baseball team can be approximated by taking the square of team runs scored and dividing it by the square of team runs scored plus the square of team runs allowed.

Also, can't help but think that if you square the totals of points for / against, you inflate the effect of those bigger numbers ? maybe not - but if you do it game by game, wont the end result change, increase / reduce ? Its got me confused, , but the logic of the statement above suggests to me that larger margins of victory or defeat will distorte the Pyth' wins calc if you look at the points as a 16 game aggregate, as opposed to 16 individual calcs averaged out at the end.

[Oldfan]

UKSF74 ~ Had a quick look at this - How the hell do they end up with the 5.9 Pyth Wins shown against the Redskins ?

I don’t know. I applied the formula to the Skins entire season, 302 PF, 377 PA, and came up with 6.3

What am I missing from this principle when it is being being applied to the NFL. What else is in the calc ?

When I did my calculation, I got a .391 result. I presume that represents the percentage of wins expected which has to be multiplied by the 16-games played to get the expected total wins. Is that what you missed?

Also, can't help but think that if you square the totals of points for / against, you inflate the effect of those bigger numbers ? maybe not - but if you do it game by game, wont the end result change, increase / reduce ? Its got me confused, , but the logic of the statement above suggests to me that larger margins of victory or defeat will distorte the Pyth' wins calc if you look at the points as a 16 game aggregate, as opposed to 16 individual calcs averaged out at the end.

I don’t think that squaring the numbers would magnify the effect of the big numbers, but it would not help us in solving the problem either.

Here’s how I’m seeing the problem:

At the very extreme, team A could go 16-0 with 16 one-point wins. Team B could go 1-15 with 15 one-point losses and one win by 31 points and teams A and B would both show +16 in net points.

Team A in this case shows zero deviation from the one-point win average. Team B’s 31 shows a huge deviation from its one-point average. So, the adjustment is one involving deviation from the average. However, that’s where my thinking stops. I don’t know which formula to apply to make an adjustment.

[Est.1974]

OF, also got 6.3

(91,204 / 233,333) x 16

Read somewhere that the NFL calc may not be an exact square, maybe applied at 1.82 or 2.37 - Are you familiar with that ? Also need to work out what the 0.4 variance is as a %.

I still like the idea of taking out some of the results - maybe reduce the 16 game sample to 14 by eliminating the bigger margins. Takes out the fluke results. That would help in your scenario, right ? Seems fair to me if applied to all teams.

Will give this some more thought later.

[Oldfan]

...Read somewhere that the NFL calc may not be an exact square' date=' maybe applied at 1.82 or 2.37 - Are you familiar with that ? Also need to work out what the 0.4 variance is as a %.[/quote'] Can't help with either.

I still like the idea of taking out some of the results - maybe reduce the 16 game sample to 14 by eliminating the bigger margins. Takes out the fluke results. That would help in your scenario, right ? Seems fair to me if applied to all teams.

I'd like to see inconsistency made a negative, so that approach wouldn't be ideal. Although, it does make an improvement.

[Going Commando]

One wonders if we're going to be any better this season. Our QB situation next season will probably be worse than the Shane Matthews and Wuerffel era--the worst of the Snyder era.

If Shanahan gives us the two worst teams of the Snyder era in his first two seasons as HC, does he deserve to keep his job for year 3? Zorn was canned for much less.

[Oldfan]

One wonders if we're going to be any better this season. Our QB situation next season will probably be worse than the Shane Matthews and Wuerffel era--the worst of the Snyder era.

If Shanahan gives us the two worst teams of the Snyder era in his first two seasons as HC, does he deserve to keep his job for year 3? Zorn was canned for much less.

A couple of posters in the free agency thread made some astute observations on this point, I thought.

Strictly as a career move, Mike might be safer going young in free agency and selling this as a rebuilding year. If he makes a big splash with big name vets, and builds hope as he did last season, and still loses, he might not make it to year three.