Writers of Pro Football Prospectus 2008

29 Aug 2005

Third Down, Red Zone Efficiency are Significant

No, really? Golly, John, if only one of those "sabermetric converts of baseball" could figure out a method of rating NFL team efficiency that really did judge teams by more than just total yardage. Maybe somebody could measure teams by yards per play, just like you suggest, and give more weight to third downs and red zone performance, and maybe adjust based on game situation and even opponent strength, so that they ended up with a statistic that did a better job of predicting future performance than wins or points or total yards. Wow, man, that would be awesome! If only somebody could come up with something like that. Maybe they would stick it somewhere on ESPN, like Page 2 or something.

Seriously, I know some people out there accuse me of being a self-promoting media whore, but apparently the problem is that I'm not enough of a self-promoting media whore. I'm gonna have to set aside some extra time each day for shameless media whoring.

Posted by: Aaron Schatz on 29 Aug 2005

44 comments, Last at 01 Sep 2005, 5:53pm by Pat

Comments

1
by sublime33 (not verified) :: Mon, 08/29/2005 - 2:20pm

He could have simplified it all by using "points per possession". This is the quickest reliable metric of all - provided that one can find the number of offensive and defensive possessions.

2
by Ray (not verified) :: Mon, 08/29/2005 - 3:45pm

I saw this on ESPN and I wondered if it would make it up here. Fire him an Email, Aaron. Maybe he just doesn't know? ;^)

3
by karl (not verified) :: Mon, 08/29/2005 - 3:51pm

Does anyone know Clayton's email address? When i saw the word "efficiency" I thought maybe FO was getting some good ESPN credit...The article is frustating. Seriously, what's Clayton's email? I want to send him a link.

4
by Phil P (not verified) :: Mon, 08/29/2005 - 4:48pm

Aaron, don't be too discouraged. FO has the best football analysis anywhere (web, print, radio, TV, etc.) and there are a growing number of people "out here" that know that.

Do post his email address when you find it so we can help spread the word.

5
by zach (not verified) :: Mon, 08/29/2005 - 5:05pm

this reminds of a headline that was on philly.com this morning: "Shore a hot spot for dolphin sightings".

6
by beedubyuh (not verified) :: Mon, 08/29/2005 - 5:06pm

I think one of the obstacles to Aaron's work being embraced by Big Media is that the underlying process that spits out the final number is too complicated for the layman to understand. Just interpreting the numbers that are spit out (e.g., when is a negative number good, when is it bad) takes more effort than most writers, much less reades want to expend.

I have been preaching the FO gospel for over a year now and have been unable to get any friends or family excited beyond a mumbled, "Hmmm, interesting." They see the equations and acronyms and their minds go blank.

It's really not that hard, but it does take some effort at first. This learning curve is anethema in a business that puts a premium on quick, compact, if not necessarily accurate, statements.

7
by Dan Babbitt (not verified) :: Mon, 08/29/2005 - 5:28pm

Hi Aaron, et al:

I clicked on espn.com/nfl and thought for a moment that Clayton had spilled the beans....

I am going to promote this site like crazy after my fantasy draft on Sunday. Until then, can we please keep the wraps on?

Just kidding about the secrecy, although not among my friends and co-owners until Sunday.

8
by Matt B (not verified) :: Mon, 08/29/2005 - 7:39pm

Did anyone pay any attention to the "Aikman efficiency Ratings" that suspiciously appeared shortly after TMQ moved to NFL.com and looked like a pathetic(and unsuccesful) attempt to copy DVOA? Call me a cynic, but I find it more likely Clayton has heard of FO and is trying to take credit for the idea just like Mr. Concussion. Next week...the premier of the Clayton Efficiency Ratings.

9
by Aaron (not verified) :: Mon, 08/29/2005 - 8:47pm

While it is true that I am always working to get my stuff out there and build my career (not to mention Mike's and Mike's and Russell's, etc.), I hope that my joking references to media whoring made it clear I'm not seething with anger or anything. Just a little annoyed.

And I don't think Aikman was ripping us off at all. Just a matter of great minds think alike. It's not shocking that a guy who quarterbacked three Super Bowl champions has some unconventional -- yet accurate -- ideas about why teams win and lose. What's shocking is how many former players not named "Troy Aikman" are completely dependent on conventional wisdom -- not to mention wrong -- when they talk about why teams win and lose.

10
by Larry (not verified) :: Mon, 08/29/2005 - 9:12pm

With more teams and coaches making moves towards this type of analysis (Patriots, Eagles and Titans come to mind), I suspect we'll see more broadcasters pay attention to this sort of deeper analysis in the future. It is probably a few years before these guys start to retire and get cushy booth jobs, and there's always the possibility the producers stop them from trying to talk about it, but we can hope.

Aikman's ratings seemed a bit ad hoc to me (in terms of the wieghts), and strangely normalized to 75=average. Better than yards, though, which is a start.

11
by Sean D. (not verified) :: Mon, 08/29/2005 - 9:34pm

I can't see Belicheck or Reid getting into the announcer's booth after the careers and spilling the beans on the sort of analysis NFL teams do. But, I could definitely see Fisher getting up there and talking about more then just how fat the lineman's butt is. That is assuming the Titan's don't make him sign a non-disclosure agreement.

12
by carl s (not verified) :: Mon, 08/29/2005 - 9:39pm

I franky don't understand why some people are so tied to their old notions of points and yards and TD receptions. I'm most familiar with this phenomenon in baseball (try telling a die-hard White Sox fan that Scott Podsednik sucks because SB don't matter all that much), but whenever me and my friends argue basketball or football its the same thing. Someone says "Michael Vick is the best QB in the NFL" and I am just left with my mouth agape, trying to figure out where to BEGIN. I seriously think not that much is going to change until some of the more sabermetric guys are able to move in the mainstream. In baseball it was Neyer, basketball Hollinger, and hopefully Aaron and Co. for football.

13
by Pat (not verified) :: Mon, 08/29/2005 - 10:10pm

I think one of the obstacles to Aaron’s work being embraced by Big Media is that the underlying process that spits out the final number is too complicated for the layman to understand.

I don't agree. In fact, it's extremely simple. It's been tweaked massively, but the basic idea underlying it is simple.

If you want to explain it to someone, you could say this. Suppose you have a team (Denver) who's got 1st and 10 at their own 20 yard line against another team (Miami), and M. Clarett rushes the ball. He gets 2 yards. On the average, for the entire league, on 1st and 10 at their own 20 yards, the average rusher gains 3 yards. 2 yards is 33% less than 3 yards, so Clarett gets a -33% for that play. This is how a "brain dead stupid" version of VOA would work. Now get smarter: OK, we can consider the average for the entire league against Miami on 1st and 10 at their own 20 yard line, which is 2 yards. So now Clarett would get 0% for that play. Now you've got a brain-dead stupid version of DVOA.

This is much nicer than yards for a lot of reasons. First off, suppose that the average on 1st and 10 is 5 yards rushing, and on 3rd and 5 it's 2 yards rushing. Well, a guy who faces a whole bunch of 1st and 10 situations will rush for more, but that doesn't make him better than a guy who faces a whole bunch of 3rd and 5 situations.

After that, the main, #1 correction is to switch to "value points" rather than "yards", for a whole bunch of reasons - first, fumbles/interceptions don't get included in yards, and the game is won and lost by points, and not yards. Aaron will probably yell at me saying that this is the entire point, and not a correction, but hey. :)

DPAR is pretty simple: it's basically an integral stat of (DVOA+18%), renormalized back to points. It's like saying someone passed for 3000 yards this season, as opposed to 300 yards/game.

But even besides that: it's not like the NFL doesn't have an arcane silly rating system already - the QB rating system, and that caught on fine.

I think the more important thing is to explain to people what it means, because it's not like the QB rating system. It's not arbitrary. It means something. DVOA correlates to a point difference. If the Rams have a +33% DVOA defense, that means that if a 0% DVOA offense plays them, if they played a bazillion times, the Rams would tend to give up 33% more points than the league average (from 2002, I think). So if the league average was 25 ppg, a 0% DVOA offense would tend to score 33 points per game against the Rams.

So if we say "Player X has a low DVOA", that means something. It means that if you replaced that Player Y with a high DVOA helped his team score more points than player X. Who cares if you gain 3000 yards, if ultimately you didn't help your team score points?

OK, my "DVOA advocacy" is now done. Aaron can feel free to scold me if I'm wrong. :) Note that I know that the brain-dead version of DVOA doesn't fix a huge number of the problems DVOA was designed to fix. But it's a basic example everyone can understand. The conversion to points can just be mentioned as a caveat, in my opinion.

14
by MarkB (not verified) :: Mon, 08/29/2005 - 11:22pm

It's so "extremely simple" that it took you paragraphs to explain it in simple terms? Think about it.

15
by Miles (not verified) :: Mon, 08/29/2005 - 11:32pm

Nice explanation of the (D)VOAs. Need a lot of work on the DPAR.

DPAR is pretty simple: it’s basically an integral stat of (DVOA+18%), renormalized back to points. It’s like saying someone passed for 3000 yards this season, as opposed to 300 yards/game.???
I have no idea what that means, or how the 2 sentences relate.

How about, 'The team should get DPAR more points this season using Player X than his backup'

16
by Pat (not verified) :: Mon, 08/29/2005 - 11:37pm

It’s so “extremely simple� that it took you paragraphs to explain it in simple terms? Think about it.

It took me one paragraph to explain it. The rest was commentary.

Again, though - what, is QB rating easy to explain? Takes something like 4 paragraphs there as well. And it also sticks in random numbers like "if result is greater than 2.375..."

I think if it was easy to calculate, there wouldn't be calculators for it over the Web.

17
by Larry (not verified) :: Tue, 08/30/2005 - 2:17am

Carl s:
I wasn't saying the coaches would take it into the booth, more their players who will have won with the new thinking. You think when Brady goes into the booth he'll spout TDs and yards as the things that are important? I tend to doubt it.

Ahhh, now is the time where all of us readers explain Aaron's stats and he sits at home saying, "Shit, a whole bunch of people are sending me money and they don't even understand what the hell I'm doing..." More importantly, we all get to have our misconceptions corrected. Anyway, I'll try my hand at shortening Pat's FO elevator spiel:

Every play has a goal. On 3rd and 1, it's 1 yard, on 1st and 10 it might be 4 yds. VOA is Player X's ability to reach that goal relative to the league average. DVOA is the same thing, relative to the average against the specific defenses Player X faced. PAR/DPAR adds up the value (converted to actual scoreboard points) of every play made over the season, and compares not to the league average, but to the Beer truck driver who'd take his place if, say, he was suspended for smoking too much weed.

I think that's a little shorter than Pat's, but perhaps not quite as clear. Also, it isn't strictly true, since VOA/DVOA are not average value of a play, but weighted averages. Partly because some plays are deemed more important (Red zone, your own 20-39). In addition, plays that have higher average success values are more highly weighted, if I understand the description properly. It's actual success points on each play that go into the numerator and denominator, not the value relative to average. At least, that's how I understand it.

Also, try explaining the rules of baseball someday. Not at all simple. But, nobody says people don't play because it's too complicated. I wouldn't use the QB rating example, because, well, QB rating is stupid.

18
by jimmo (not verified) :: Tue, 08/30/2005 - 2:34am

The day will come when the average/casual/ armchair football fan will believe Ladell Betts is better than Clinton Portis and that Sammy Morris is better than both of them or that Chester Taylor is more valuable than LdT.
I'm sorry but the numbers here are not "extremely simple." I love football, I love this site, I love the attempt to dig more meaning out of the simple stats we've grown up with, and also long walks on the beach and sunsets..oops, sorry... but my eyes glazed over three sentences into Pat's explanation, and I understand it!

Flawed as it is, QB rating matches up fairly well with DPAR/DVOA as far as rankings go in that nine of the top ten are the same in both, with a bit of a different order. So is Kyle Boller so truly horrible as conventional wisdom and his passer rating of 70.9 suggest, or is he just average or slightly below, as his DPAR/DVOA rankings of 20/21 show?

I really have no point, except that its very late any my babies will be up very, very soon, and that even though I've cherry-picked from the RB rankings, no can ever convince me that Chester Taylor is more valuable than LdT.

19
by Aaron (not verified) :: Tue, 08/30/2005 - 8:08am

Remember there's a difference between "Chester Taylor is a better player than LaDainian Tomlinson" and "Last year, though it was a small sample size, Chester Taylor had more value than an injury-hobbled, not-100% LaDainian Tomlinson." The stats on this site are meant to be used as a tool, not as a replacement for your own judgment...

20
by jimmo (not verified) :: Tue, 08/30/2005 - 10:24am

Aaron, that's part of the/(my) problem; the stats are supposed to replace judgement, or at least a large part of it, and stand as an objective measure of a guy's value. The use of stats is to dehumanize the evaluation process, that is remove to a degree the human element, the "Ryan Leaf just looks like a quarterback" branch of scouting.
The general public already relies on their own judgement in forming opinions; my belief is that FO is attempting to minimize the error inherent in much of that judgement, debunking the conventional wisdom as it were. If we're told, I know our stats say Chester Taylor was more valuable than Tomlinson last year but obviously he isn't because your judgement says so, then why believe the stats?

I say this of course as a supporter, more devil's advocate than true critic.

21
by Aaron (not verified) :: Tue, 08/30/2005 - 10:59am

Well, you need to look at a few other things besides DVOA or DPAR:

1) Sample size
2) Trend over multiple years
3) Possible mitigating circumstances

Now, there are some cases where the stats totally overwhelm 1, 2, and 3, like Ricky Williams over his career, where he had just the one strong year in 2002 and otherwise was never the player people thought he was.

On this specific issue, it isn't that the numbers are wrong about Chester Taylor, but rather that we know that Tomlinson was injured last year and was the most valuable back in football in 2003. So I don't think the numbers prove that Taylor is a better player. There is a difference between "had more value in a specific season" and "is an intrinsically better player." Taylor was more valuable than Tomlinson in 2004, but that doesn't mean he's better, period. Tomlinson just had an off year.

When it comes to the conventional wisdom about which players are better, I'm a believer in "innocent until proven guilty." I don't want to emphatically insist that everyone else is wrong unless the statistical evidence is *very* strong. Sometimes it is, of course.

By the way, the "adjusted points per game" number from PFP is an attempt to make DVOA easier to understand, and we'll be listing it on the site this upcoming season. I'm just trying to fix a bug in my equations that seems to make the adjusted points O-D+ST not rank the 32 teams the same was as DVOA for O-D+ST. Did people like the adjusted points per game?

22
by Ray (not verified) :: Tue, 08/30/2005 - 11:02am

Pat, from what I've seen of your other posts here you're a statistician. Take my (and the others here) words for it when we say that DVOAs not as simple for most people as it is for you. ;^)

Personally, I think one thing that could be immediately done to help(although it's probably long too late to do so) is to get rid of the silly "negative is positive for defence". C'mon Aaron, haven't you heard of the absolute value? ;^)

I know it sounds like a small petty thing, but if I get annoyed by having to double-check whether 34.5% is good or bad for a stat, then other people do too. Why make people have to reverse their mind-sets depending on the column of the chart?

If you really want to make DVOA more accessable, then I think swapping the sign of the defencive DVOA would be a good place to start.

23
by Aaron (not verified) :: Tue, 08/30/2005 - 11:51am

Actually, every time I've asked about that, the majority of respondents tell me they prefer it this way. (The negative sign for defense, that is.)

24
by RowdyRoddyPiper (not verified) :: Tue, 08/30/2005 - 11:54am

"I know it sounds like a small petty thing, but if I get annoyed by having to double-check whether 34.5% is good or bad for a stat, then other people do too. Why make people have to reverse their mind-sets depending on the column of the chart?"

Don't project your hang-ups on me Daddy-O ;)

The reason is so comparing an offense for a given team versus the defese they are playing is relatively simple. You just add the two numbers, if they're positive you'd expect the offense to have a better game than normal, if they are less you'd expect them to have a worse game, vice-versa, etc. inter alia kaput.

I guess you could just subtract defense from offense, if you made the signs equal to compare matchups (like you do for finding overall team DVOA) but where's the fun in that?

As for the Clayton piece...I like how he brings up ohhhh I don't know, the situation where a team can finish very high in pass defense (as measured conventionally) and not really have a good pass defense...I remember reading something like that somewhere recently...can't really remember where though.

25
by pawnking (not verified) :: Tue, 08/30/2005 - 12:04pm

I've been reading this board off and on for a while now, and I've continually made this point: football is such a complicated game with so few plays compared to other sports (specifically baseball) that it is extremely difficult to quantify it in statistics. Things like coaching decisions, game plans, opponents' decisions, injuries, mistakes, and so on severely skewer any data we can gather.

In baseball, there are a relatively few possible plays and outcomes which are repeated thousands and thousands of times, with participation limited mostly between two players (the pitcher and the hitter) and we are still evolving our methods of evaluating players. Each play in Football involves not only 22 players, but also the coaches, the field position, wind direction, field condition, and other factors which we may not even be aware of. How do you quanitfy all that?

I think FO has done a wonderful job by starting to ask serious questions about all of this. Most importantly, they have approached the whole matter in a systematic way, unlike ESPN and other writers who are more interested in a story than a system. But let us not lose sight of the fact that it is a hideously complicated task, one which is continually changing as offensive and defensive philosophies change and as players grow stronger and faster.

Further complicating the matter is the extreme secrecy and deceptiveness that the NFL teams routienly engage in. They lie about injuries, about depth charts, about plays called, about ticket revenue, about draft decisions, and what they don't lie about they don't talk about. Without any transparancy, it will continue to be practically impossible to discern what's actually going on in a game enough to predict with any accurracy what will happen in the future.

Ah well, keep fighting the good fight, Aaron. You're doing a better job that anyone else right now.

26
by Pat (not verified) :: Tue, 08/30/2005 - 12:06pm

Ray:

It's points. Positive means more points. This is good for offense, bad for defense. I prefer it that way.

I think the "adjusted points per game" stat is much more accessible, actually. For one thing, it aligns better to DPAR.

So when you have "offense adjusted points scored per game" of 35 ppg that's good, and "defense adjusted points allowed per game" of 35, that's bad.

27
by Parker (not verified) :: Tue, 08/30/2005 - 12:12pm

RE: #17

Glad to hear it will be Tom Brady that will lead the charge to bring football into the next stage of statistical analysis. But when will he have time to heal the infirm and teach all the children to read?

Uuuugh.

28
by Mike (not verified) :: Tue, 08/30/2005 - 12:19pm

Personally, I like the "negative is good for defense, positive is good for offense". But I am not everyone. And I think the people that read FO regularly do NOT represent the average football fan (which is one reason why I like the message boards here so much more than any other football message board on the web!). I actually think Joe Sixpack wants one number where the better it is, the better his team is.

As to the issue of how hard DVOA is to explain--Pat, I second what Ray said. It's not too bad for people with a background in statistics, or even with a background in math or science, but there are more people out there that don't then there are that do. I have a friend who once commented "Oh, I was never any good at Math. I always got stuck on the Word Problems..." I don't think DVOR is very easy to explain to a layman, let alone DPAR.

Still, I don't think that's a condemnation of the stat. I doubt 95 % of football fans could recite the formula for QB rating, and it did catch on. They just know "completions=good, TD's = better, INT's = bad". Similarly, when DVOA catches on, people won't need to be able to calculate it themselves, they just will know "1st Downs = good (for offense), TD's = better, Punts = bad, Turnovers = worse".

29
by carl s (not verified) :: Tue, 08/30/2005 - 12:33pm

I think the whole "LT2 is better than Chester, therefore stats suck" argument is way overstated. A season itself is a pretty small sample size, and flukes happen. Joe Harris (who?) had a better season than Babe Ruth in 1925. It's just a fact, and it says nothing about the overall skill of either player.

Stats are not meant to replace your personal judgements, but they are almost always a strong evidence, and you really need reasons that they should be ignored. You happen to think LT was the best RB last year? You'd better be ready to explain why only 45% of his runs were succesful. I'm not saying you can't do it, maybe his offensive line was poor, maybe he just got unlucky, but you can't rationally just say "he is the best, so these stats must be wrong."

30
by Ray (not verified) :: Tue, 08/30/2005 - 12:39pm

RE #24, 26

I know what it means, and I know why Aaron has it set up that way. The thing is, DVOA is measured in a percentage, not in points. I know it's supposed to translate over to points, but how do you make that direct corellation?

DVOA is not a concrete number that an average person can see and understand intuitively (such as high points gained being better for an offence and high points allowed being worse for a defence). So already people have to conceptualize what this number means. Adding in complications like positive is good for offence but bad for defence just makes an already up-hill battle for acceptance more difficult.

Is it super-hard to understand? No. Does it take a little personal effort? Yes. Do people like putting effort into understanding stuff? No. It's fine for the intellectuals we have around here, but if you really want widespread acceptance, simplification is key.

Personally, I think that if you want to take an even better step towards making DVOA more accessable, drop the percentage and just make it a numeric value. People are used to arbitrary numbers that they don't understand (see QB Rating). They know that they'll get used to getting a feel for what's 'good' and what's 'bad'.

But a percentage is confusing. People like percentages to be between 0 and 100. You got a perfect on your test? 100%! You got every question wrong? 0%! Easy, understandable worst and best values. When you break out of that range, it makes people uncomfortable.

All that can change when it's just a number. The whole pre-concieved 0 to 100 range is gone and people can see 30 and learn to realize that hey, that's pretty good.

Of course, you'd definitely have to take the AbsVal of the defencive DVOA then. But I still think the average person will be more comfortable thinking positive = good, negative = bad and not having to change that thought process when scanning a line.

Unfortunately, I'm not sure it isn't too late to change, especially considering there's a print book out there. But I still think it would make DVOA easier for more people to understand, which can't be a bad thing.

31
by jimmo (not verified) :: Tue, 08/30/2005 - 1:09pm

"I think the whole “LT2 is better than Chester, therefore stats suck� argument is way overstated"

Me too, that's why I'm glad nobody made that argument.

32
by Jeff (not verified) :: Tue, 08/30/2005 - 1:24pm

"You happen to think LT was the best RB last year? You’d better be ready to explain why only 45% of his runs were succesful. I’m not saying you can’t do it, maybe his offensive line was poor...."

This hits on the exact problem with DVOA and DPAR. It does not account for the quality of teammates, nor can it. To me, this is a huge (and probably unsolvable) problem with trying to evaluate individual players with stats.

33
by Miles (not verified) :: Tue, 08/30/2005 - 1:52pm

Re: #24. You just add the two numbers, if they’re positive you’d expect the offense to have a better game than normal...

Not so compelling for me because we could use subtraction instead of addition. I think makes more sense because it gives you intuition about what a positive/negative number means, and probably follows a mental model more closely -- offense trying to gain positive yard, defense trying to stop them. Adding the two forces doesn't give this intuition and thus forces one to remember that better defenses are negative.

34
by B (not verified) :: Tue, 08/30/2005 - 5:23pm

"Aaron, that’s part of the/(my) problem; the stats are supposed to replace judgement, or at least a large part of it, and stand as an objective measure of a guy’s value."
This sounds like a rehash of the old andi-moneyball argument, to which I counter:
DVOA/DPAR arn't supposed to replace judgement with stats, they replace bad statistics with good statistics. Or, if you prefer, flawed statistics with less-flawed statistics.

35
by Ray (not verified) :: Tue, 08/30/2005 - 6:08pm

I guess what it really comes down to is Aaron's goal with this site.

Does he want to more people to know about his stats so that he doesn't have to get annoyed when guys like John Clayton publicly wish that there are stats like DVOA when there is in fact DVOA? Then he'll need to make the stats more accesable and easier to understand.

Does he want to appeal to his current audience who likes complicated stats and don't care if the stats are over the head of your average football fan? Then he should keep things the way they are to keep us intellectuals happy, but then he shouldn't get annoyed when guys like John Clayton don't know about DVOA. ;^)

Really it doesn't matter much to me which path Aaron decides to take. I'll be here either way. :^)

36
by B (not verified) :: Tue, 08/30/2005 - 6:15pm

I think Aaron's goal here is to get the site popular enough to pay for his daughter's tuition at Brown.

37
by Miles (not verified) :: Tue, 08/30/2005 - 9:24pm

I've long beleived that the reason John Madden used to be the best NFL commentator was that he actually explained exactly why plays worked or didn't -- what the effect of this guy doing x was on this other guy's role etc. I thought he made fans understand that there was more to football than knocking the crud out of someone. This directly carried over to the EA games -- they are by far and away the most complicated football games on the market. Of course, Madden seems to have attributed his sucess to the Maddenisms, and started forcing them, making him possibly the worst commentator. Long story short, I think the average fan wants to understand football a lot more than they do, but isn't given the opportunity to learn (I've always wanted a tutorial in Madden Football -- what *should* I call if I think the offense is going to do x??). I hope this site continues to help fill that gap. For me, the Football 101 articles were awesome, and the Football Prospectus is the first football book I've ever bought and read.

38
by jimmo (not verified) :: Tue, 08/30/2005 - 11:16pm

B, re:34 & 20, that's actually a pro-Moneyball statement. Beane doesn't want the old-boy scout's network sitting around over beers extolling the virtues of a prospect's tools. Way too subjective and of course often ineffective.
Given data that he values, Beane is able to remove at least partly the human element, the intuition, the guesswork and replace it with objective analysis. At least that's my take on Moneyball, and I'm all for it.

39
by Dean (not verified) :: Wed, 08/31/2005 - 12:02am

Just a lazy question (ie I cant be bothered looking for the answer), but are the FO formulas protected in any way from being copied?

I'd hate to be watching Superbowl ?? and hearing about DVOA and DPAR and some non-FO bozo taking all of the credit.

40
by MCS (not verified) :: Wed, 08/31/2005 - 12:55pm

Disclaimer: Just an opinion, not meaning to offend the mathematical intellects out there. I love what the FO team and loyal readers are trying to do here.

I tend to think of myself as an educated man and I have had some basic statistical training, but I have a hard time following some of the discussions on this site. Note that I want to understand the statistics, not just take them at face value. I have gone as far as to conduct other research on to try and understand the math better. (I never liked people that said they didn't like Dennis Miller because he spoke above their heads. My opinion is read more to try and understand. Let's not reduce the entire country to the lowest common denominator).

If the FO team wants to make this site understandable by all, then maybe they need to create a "Statistics for Dummies" page. Not just to explain the statistics used here, but to explain statistics in general.

For example, on another thread, FO poster Pat states that he/she (sorry I don't know, no offense intended) had to "do a regression". What does that mean?

How do we know certain statistics tend to "Regress to the mean". Some may not know what that means.

What is R^2 and what is the effect?

What is a P-Value and how does it apply?

With research, all of this information is readily available, but in my experience, most people do not want to do that extra work.

The question that the FO team needs to ask themselves (as someone posted above) is: What is the target demographic for this site?

41
by B (not verified) :: Wed, 08/31/2005 - 1:00pm

I don't understand all the math either, I just ignore anything I don't get and look at the conclusions. I found these links on wikipedia, though, I hope it helps:
http://en.wikipedia.org/wiki/Regression_analysis

http://en.wikipedia.org/wiki/Statistics

42
by Pat (not verified) :: Thu, 09/01/2005 - 2:22am

I know what it means, and I know why Aaron has it set up that way. The thing is, DVOA is measured in a percentage, not in points. I know it’s supposed to translate over to points, but how do you make that direct corellation?

It's direct. If you have a 10% offensive DVOA, you score 10% more points than average. If you have a -10% offensive DVOA, you score 10% fewer points than average.

What is R^2 and what is the effect?

That stuff was stat geeks numbers. I stated what the actual meaning was in that same thread.

R^2 is "goodness" of a fit. It can be interpreted as the amount of the spread that can be attributed to your function. It goes from 0->1. Zero means that your fit is crap. 1 means your fit is perfect.

Simple example: I plotted point differential in first game versus a division rival against point differential in second game. So if Philly beats NYG by 10 in the first game, and loses by 3 in the second, I put a point at (10,-3). Do that for all teams, and then fit a straight line to it. The R^2 is about 0.08: that means roughly that 8% of the spread in that plot is due to a correlation between the first game and the second game. (The rest, presumedly, is from game-to-game variations).

What is a P-Value and how does it apply?

P-value (which I hate) tells you how likely it is that the fit you've got is from random fluctuations. So if you've got a P-value of 0.05, there's a 5% chance that a random data set would have the same correlation.

P-value of less than 0.05 is typically used in statistics to mean "this is real". Higher than 0.05 means "this is random."

(Note that none of this is really strictly true - but it is true in a handwavy sort of way.)

43
by MCS (not verified) :: Thu, 09/01/2005 - 8:46am

Pat and B,

Thanks for the clarification and the help. I am sorry I wasn't more clear. Those were just example questions that someone may ask. Personally, I have a basic understanding of the concepts discussed here. Not the in-depth knowledge that you have, but enough to be dangerous.

I meant to comment that "the masses" probably don't have the statistical know how to follow a lot of the discussions nor do they have the initiative to conduct their own research in order to understand the concepts better.

Just look at how Dennis Miller was received. Most people didn't understand him and rather than read and improve their intellect, they complained and now we have Madden who just ums, ers, and yells. Madden has become a caricature of himself. But I digress, that's for a different discussion.

The point I was trying to make to the FO team is that, if they want to revolutionize the NFL stats world, they may need to educate some people better.

44
by Pat (not verified) :: Thu, 09/01/2005 - 5:53pm

I meant to comment that “the masses� probably don’t have the statistical know how to follow a lot of the discussions nor do they have the initiative to conduct their own research in order to understand the concepts better.

Yah, but I think they know that. Aaron avoids using stat-geeks terms, except by prefixing it with "For you stat geeks". It's easy enough to explain qualitatively what the quantitative R^2 and P-value mean such that people don't need to know.

High R^2: "significant effect"
Low P-value: "statistically significant"

I don't think people need to be educated. Just need to be careful about making sure there's always a simple explanation as well as the stat geek stuff.