Statistical analysis of football

[RLflutie7]

Total yards per game is not as indicative as the total yards per play because yards per play directly correlates to the number of plays run and the yards gained/given up per each. Total yards in a game are often skewed, especially by losing teams because they often have HUGE passing numbers against prevent defenses. That's why RUSHING yards per attempt for/against is more valuable. Show me a team that have astronomical passing yards every game and I'll show you a losing team more often that not. Show me a team that runs the ball more, for a higher average while holding opponents to lower averages and I'll show you a winning team way more often than not.

 

GO BILLS!!!

 

 

Yes and that's why I don't put too much value in passing stats. Also a receiver who has over 100 yards receiving in a game is a worthless stat, because it doesn't indicate a winning team.

[K-9]

This statistic is not causal. This is a great misnomer in the NFL. Assume the following - in the first half Team A (a power running team - say PIT) and Team B (a wide open offense - say NE). Lets say PIT runs 25 plays of which 15 are runs and NE runs 25 plays 15 of which are passes. They go into the half with NE leading by 14. In the 3rd quarter PIT and NE hold their gameplans and the lead staty the same. PIT running 13 plays (8 runs) and NE running 13 plays (5 runs). In the 4th quarter, PIT trying to claw back runs 12 plays (11 passes) and NE runs 12 plays (11 runs). In the end PIT has run the ball 24 times (48% runs) vs NE's 26 times (52% runs). Now NE has run the ball more and has more attempts, but this stat is a result of their winning not the cause of it.

 

I am not stating that this is how it always plays out, but if you examine those statistics and take out the run only/pass only downs at the end of games, you will get a much different outlook on run/pass ideal ratio.

 

See: Yards per play AGAINST/FOR.

 

This takes into consideration run/pass ideal ratio. By ideal ratio I assume you mean balance and, yes, that is ideal. Anytime a defense can make an offense one dimensional they're going to dictate. Of course their are exceptions. Teams with QBs that call their own plays like the Colts and Patriots with both QBs calling the play at the LOS are exceptions to the rule. At least as far as OFFENSIVE stats go. But what was the FIRST stat I listed: yards per rushing play AGAINST. Look at NE and Pitt over the last several seasons and they've been consistently dominant in this category. NE tailed off last year, but their offense put opposing teams in so many deficit situations that they were able to dictate defensively more often than not.

 

I hear what you're saying and I agree, but over the long haul the stats that coaches deem most indicitive are the ones I mention. All the others are offshoots. And of course, when it comes to breaking down game tape for those tendencies previously mentioned, they DO pay attention to what teams like NE and PITT did against each other and how and in what situations.

 

GO BILLS!!!

[Coach55]

See: Yards per play AGAINST/FOR.

 

This takes into consideration run/pass ideal ratio. By ideal ratio I assume you mean balance and, yes, that is ideal. Anytime a defense can make an offense one dimensional they're going to dictate. Of course their are exceptions. Teams with QBs that call their own plays like the Colts and Patriots with both QBs calling the play at the LOS are exceptions to the rule. At least as far as OFFENSIVE stats go. But what was the FIRST stat I listed: yards per rushing play AGAINST. Look at NE and Pitt over the last several seasons and they've been consistently dominant in this category. NE tailed off last year, but their offense put opposing teams in so many deficit situations that they were able to dictate defensively more often than not.

 

I hear what you're saying and I agree, but over the long haul the stats that coaches deem most indicitive are the ones I mention. All the others are offshoots. And of course, when it comes to breaking down game tape for those tendencies previously mentioned, they DO pay attention to what teams like NE and PITT did against each other and how and in what situations.

 

GO BILLS!!!

 

We can agree to disagree on this one.

[RLflutie7]

This statistic is not causal. This is a great misnomer in the NFL. Assume the following - in the first half Team A (a power running team - say PIT) and Team B (a wide open offense - say NE). Lets say PIT runs 25 plays of which 15 are runs and NE runs 25 plays 15 of which are passes. They go into the half with NE leading by 14. In the 3rd quarter PIT and NE hold their gameplans and the lead stays the same. PIT running 13 plays (8 runs) and NE running 13 plays (5 runs). In the 4th quarter, PIT trying to claw back runs 12 plays (11 passes) and NE runs 12 plays (11 runs). In the end PIT has run the ball 24 times (48% runs) vs NE's 26 times (52% runs). Now NE has run the ball more and has more attempts, but this stat is a result of their winning not the cause of it.

 

I am not stating that this is how it always plays out, but if you examine those statistics and take out the run only/pass only downs at the end of games, you will get a much different outlook on run/pass ideal ratio.

 

The picture you paint is why coaches like Marty Schottenheimer always lose in the playoffs. Marty wants to run first all the time and he runs too much in the redzone, never taking chances, even with good field position outside the redzone.

 

In 2004 he changed his tune and the team had a great offense. He liked Antonio Gates, so he was more willing to throw the ball down field.

 

Marty and Drew Brees had a fight in 2003 on the sidelines when Marty benched Brees for no reason (in Pittsburgh after a goofy turnover that wasn't Brees' fault). I love Doug Flutie and everything, but that benching was uncalled for. It was Marty's conservative game plan that was causing the problems. He had no confidence in Drew Brees.

 

In 2004, Marty said "Martyball" is dead and they became a great offense.

 

Bottom line: You have to have confidence in the passing game as well as running the ball, because all good teams stop the run and if you can't pass you have no chance to win. The best defenses in the league stuff the run and control field position and the only way to counter that is to have a good passing game.

[OCinBuffalo]

The final arbiter of all stats models is: can they be used to accurately predict future results based on REAL correlations or causation? Kneeling on the ball doesn't cause you to win the game, it's the effect of winning the game.

 

Before you can hope to suggest any kind of predictive model, in fact before you can even start any real comparison, you have to make sure that your raw data is pure, and not likely to be blatantly skewed by consistent outside influences = playing at least 9 of your games inside each year = makes people think Dante Culpepper is a SB winning QB, and then some idiot signs him to play in the AFC East(and against the AFC North that year). Sorry dude. Still laughing at Miami...

 

However, you can make consistent adjustments to raw data, just like the Football Outsiders do. Proper statistical analysis of football can be found here. I have nothing to do with these guys, I just recognize sound statistical work when I see it.

 

Even more important: if you are going to create a relationship between sets of data, then that relationship needs to be consistent, otherwise, all conclusions based on that relationship are a waste of time.

 

For example(and this is just one of many examples I can give, for that certain idiot out there), KC Joyner's latest foolishness attempts to relate O line to RB success. The problems he has already built into his "model" are the relationship between Point of Attack blocks missed and Non-Point of Attacks blocks missed, and then trying to relate that data to yards per carry, with no adjustments for quality of defenses faced, and no adjustment for quality of the running backs in question.

 

The first problem is: he is treating POA blocks the same as NPOA blocks. I think we can all agree that we have seen plenty of running plays off the Left Tackle that are completely unaffected by the success/failure of the Right Tackle's block and vice versa. KC treats all missed blocks the same in his "analysis", and QED that's a major flaw.

 

One way to adjust for that would be to assign an arbitrary multiplier to weight the effect of missed POA blocks heavier, but I don't like arbitrary multipliers(see: why half the country always has a stats quibble with the BCS) A better way to do it would be, once again, do what the Football Outsiders do and let that multiplier be determined by the actual data itself, similar to the "defenses faced" adjustment below.

 

The next problem is no adjustment for "defenses faced": The Outsiders use "expected" averages and %s in the beginning of the season based on team defenses from the season before, and then gradually replace those #s with actual #s, decreasing the "expected" #s weight each week. By week 7 they are using all real data. By then, the actual data is now statistically relevant, because, similar to batting average, defenses have played enough games for their #s to be reliable performance indicators.

 

The last problem, although I am certain there will be more when KC "finishes" his "analysis", is that you can't compare 2 moving targets, O line and RBs without some way to hold one of them constant, or build a consistent relationship between them, and use that relationship as a constant: otherwise, it's chaos and means nothing. IF we are attempting to study O line quality, then there has to be an adjustment for quality of the RBs that can hold them as a relative constant. I am aware that you can't get this perfect, but you are also wasting everyone's time by treating Ladanian Tomlinson the same as Ricky Williams.

 

Consider: "With Denver's O line, anybody can run behind them and be an all-pro".

IF we believe that, then the O line is the constant, and we would measure RBs success against that O line.

Consider the Converse: "Walter Payton/Barry Sanders could run behind old ladies and still get 1000 yds".

IF we believe that, then the RB is constant and we would measure the O lines success against the constant performance of the RBs.

 

The only thing you can do is compare the same RBs and their O lines year to year. But...don't we already know if the line is better this year vs. last year? There already are stats for that: individual blocks missed, sacks allowed, total offensive yards, Scoreboard, and 3rd down conversions. So again, what's the point? KC might as well be telling us that dome/good weather teams throw the ball more often and that good weather makes for better passing....oh...wait...he already did.

 

Based on the "predict future results" test, KC's "math" simply doesn't pass...or did Tavaris Jackson not suck miserably the last 2 years?

[K-9]

We can agree to disagree on this one.

 

Isn't that what makes America beautiful!

 

GO BILL!!!

[billsfan89]

The picture you paint is why coaches like Marty Schottenheimer always lose in the playoffs. Marty wants to run first all the time and he runs too much in the redzone, never taking chances, even with good field position outside the redzone.

 

In 2004 he changed his tune and the team had a great offense. He liked Antonio Gates, so he was more willing to throw the ball down field.

 

Marty and Drew Brees had a fight in 2003 on the sidelines when Marty benched Brees for no reason (in Pittsburgh after a goofy turnover that wasn't Brees' fault). I love Doug Flutie and everything, but that benching was uncalled for. It was Marty's conservative game plan that was causing the problems. He had no confidence in Drew Brees.

 

In 2004, Marty said "Martyball" is dead and they became a great offense.

 

Bottom line: You have to have confidence in the passing game as well as running the ball, because all good teams stop the run and if you can't pass you have no chance to win. The best defenses in the league stuff the run and control field position and the only way to counter that is to have a good passing game.

 

Martyball is playing offense to move the ball steadily up the field through running and smart short passes. It works during the regular season but fails in the playoffs. The Dolphins did this last year and they got a division title but got smoked in the playoffs. When you face better defenses they can stuff the box and make you ineffective on offense. You can play not to loose in the regular season but you play to win in the playoffs or else you are just lucky to win even one game

[DonInBuffalo]

I see people misusing statistics all the time, especially when it comes to sports statistics. Many people misuse statistics simply because they don't properly understand the most basic concepts.

 

Probability theory is the branch of mathematics that deals with computing the likelihood of specific events occurring, given certain assumptions. One assumption is the selection method is Random, which means each event is equally likely to be selected. To apply probability theory to a real world situation, you need to build a mathematical model of that situation, and confirm that it properly applies. For example, if a die isn't "true", if some numbers are more likely to be rolled than others, then probability theory doesn't apply, because the selection method isn't random.

 

Statistics studies historical information, and draws conclusions. That information can then be used to make future predictions.

 

One common use of statistics is to determine if something is truly random. For example, a casino owner could use statistics to determine if a die was "loaded", by calculating the likelihood of a certain sequence of rolls occurring. This is sometimes referred to as a Confidence Interval. I wouldn't be surprised if casinos monitored results statistically, and use this technique to determine which tables to watch more closely, because of the likelihood of cheating.

 

Another common use is to predict election results. As long as you are selecting a truly random sample of voters, you can make fairly accurate predictions with a sample sizes as small as 500 voters. Those would typically have an error margin of something like 3-4%.

 

When building statistical models for sports, it's important to keep all of the basics in mind. Statistics works best for baseball, because you have large sample sizes and parts of the model come close to being random. There are 162 games in a season; players get as many as 500 plate appearances, see over 1000 pitches. Defenses are close enough to being equal that assuming they are equal doesn't throw your results significantly off. In baseball, there's one major battle going on (pitcher vs. batter) with a few variables that effect that. (score, inning, men on base...)

 

Football is much more difficult because your sample sizes are so much smaller. On top of that, there are very few things that can be approximated to be random. BuffaloBill phrased that well in his post above. Because there are just so many factors that come into consideration on every play, it's very difficult to draw conclusions. People need to be careful to draw the distinction between what a statistic is actually saying, and what (appropriate) conclusions can be drawn from that statistic.