Yakuza Rich: Stats Show You Shouldn't Worry About the D for Now

[Yakuza Rich]

Stats Show You Shouldn't Worry About the Defense For Now
by Yakuza Rich
http://yakuzarich.blogspot.com

The question was posed today about the lack of plays the Cowboys defense has made in the preseason. The team only has 3 defensive sacks with 0 interceptions and forced fumbles. The general rule of thumb in preseason is that it really doesn't matter unless the team goes winless or undefeated as history has shown teams that go winless generally do not do well in the regular season while teams that go undefeated are more likely to make the playoffs.

Either way, I wanted to see how some defensive stats in the preseason correlated to the regular season. So here's the stats I looked at:

1. Plays from scrimmage per forced fumble
2. Plays from scrimmage per fumble recovery
3. Pass attempts per interception
4. Pass attempts per sack
5. QB Rating Allowed

After calculating the team's preseason and regular season numbers for the stats above, I then ran a correlation coefficent.

As simple as I can put it, a correlation coefficient is a number that ranges from -1.0 to +1.0 and it helps determine the statistical relationship between two variables.

The closer the number is to -1.0, the stronger the NEGATIVE correlation is. For instance, if I had a restaurant and wanted to know the correlation between hot soup sold and temperature and came up with a correlation coefficient of -0.8, that would mean there's a very strong negative correlation or I could term it as "the higher the temperature, it's likely that there will be less hot soup sold."

On the flip side, if the number is closer to +1.0, the stronger the POSITIVE CORRELATION is. For instance, if I wanted to know the correlation between lemonade sold and temperature and came up with a coefficient of +0.8, that would mean there's a very strong positive correlation or I could term it as "the higher the temperature, the more likely I will sell lemonade."

Lastly, the closer the number is to Zero, that means there's no correlation. So if I'm trying to determine hamburgers sold in comparison to temperature and come up with a coefficient of +0.03, that means that temperature has really no effect on burger sales. It could be 100 degrees or 25 degrees out and I'm likely to see now difference in burger sales.

So that's what I'm looking for here and here's what I came up with for the past 2 seasons (I could only find preseason stats for the past 2 years):

PLAYS FROM SCRIMMAGE PER FORCED FUMBLE

Year...............................Correlation Coefficient

2007.........................................-0.1328
2006.........................................-0.0307


PLAYS FROM SCRIMMAGE PER FUMBLE RECOVERY

Year................................Correlation Coefficient

2007.........................................+0.0927
2006.........................................-0.0335


PASS ATTEMPTS PER INTERCEPTION

Year................................Correlation Coefficient

2007........................................-0.01749
2006........................................+0.52401


PASS ATTEMPTS PER SACK

Year...............................Correlation Coefficient

2007.......................................+0.09833
2006.......................................+0.12694



QB RATING ALLOWED

Year...............................Correlation Coefficient

2007.......................................+0.356159
2006......................................+0.424259


Unfortunately I don't have that big of a sample size, but the indication is that when it comes to the stats I analyzed, preseason play is really no indicator of how the team will do in these areas in the regular season. The strongest correlation out there was pass attempts per interception in 2006 (a coefficient of 0.52) and even that appears to be an abberation as in 2007 the coefficient was near zero. Out of all of the stats listed, the biggest correlation year to year appears to be QB rating allowed, and it's strongest correlation was only 0.42 which really isn't all that strong of a correlation.

For instance, last preseason the St. Louis Rams finished 1st in pass attempts per sack in the preseason, but were 22nd in that statistic come the regular season. And the Vikings finished 1st in pass attempts per interception in preseason in 2007, only to finish 28th in that statistic in the regular season.

So while I'd like to have seen better play from the defense and the team looks talented on the field, but not exactly sharp, statistics show that it's ridiculous to worry too much about it because it is just preseason.









YAKUZA

 

[ZeroClub]

That's interesting. It must have taken some time.

Did you use rank-order data for your statistics (as opposed to interval or ratio level data)?

I'm not surprised to see very low correlations associated with the prediction of relatively low base rate events (fumbles, interceptions, and even sacks) or ratios involving those low base rate events.

Because of their statistical properties, I'd guess that more general indices (such as yardage, points and QB rating) would stand a better chance of being predictive.

 

[CrazyCowboy]

YAKUZA that was very interesting.......your a very smart person.

Thanks

 

[Cbz40]

YR...You are the man.....I appreciate the time and effort you spent bringing the info to us.

 

[VACowboy]

Good stuff, Rich, as always.

 

[fortdick]

That makes my head hurt.

 

[VietCowboy]

so with such a small sample size, what where the alphas? What was statistically significant?

 

[Alexander]

Preseason statistics being used to somehow predict real success, with a correlation coefficient no less.

I have now seen it all.

 

[Yakuza Rich]

Did you use rank-order data for your statistics (as opposed to interval or ratio level data)?

.

No, I used ratio data. It's more accurate that way.





YAKUZA

 

[Yakuza Rich]

Preseason statistics being used to somehow predict real success, with a correlation coefficient no less.

I have now seen it all.

Well, that's where you're missing the point. I showed that you really can't predict success based on preseason stats. I wanted to see if there may be something, but when I ran the correlation coefficients, it proved there was no correlation one way or another.

And I wasn't predicting it to "real success" anyway, which would be how each stat relates to victories in the regular season. I was comparing it to how the stat in the preseason correlates to the regular season.

Now, if the correlation coefficients were say -0.6 or worse o +0.6 or better, then we may be onto something.




YAKUZA

 

[WoodysGirl]

That makes my head hurt.

My eyes glazed over...

Still props to you YR for making the effort.

 

[MaineBoy]

lies, damn lies, and statistics.

 

[Yakuza Rich]

Because of their statistical properties, I'd guess that more general indices (such as yardage, points and QB rating) would stand a better chance of being predictive.

Perhaps, although I think it would be hard with QB rating because a team could have a really great QB who only plays a series, throws a pick, and that kills his QB rating. Or the backup who isn't that great plays most of the preseason and stinks up the joint and then come the regular season, the starting QB plays the entire year and has a great season. I think that the stats I posted show that there's just too many variables as to why "preseason isn't important" and I highly doubt there's any stat that has a strong positive or negative correlation.





YAKUZA

 

[Yakuza Rich]

so with such a small sample size, what where the alphas? What was statistically significant?

I didn't get into standard deviations or creating an algorithm. The correlations were so weak it's not worth the time or effort regardless of sample size.






YAKUZA

 

[MichaelWinicki]

Nice job Rich. I appreciate the effort you put into this stuff.

 

[CanadianCowboysFan]

That makes my head hurt.

I thought it might just be me, glad it gave someone else a headache.

 

[Kevinicus]

Is this based on all preseason stats, or just the stats of the 1st teams?

 

[Da Hammer]

That makes my head hurt.

that was my reaction as well.

 

[Yakuza Rich]

Is this based on all preseason stats, or just the stats of the 1st teams?

All preseason stats. There's no real possible way to determine the stats of the first teams because not only do teams sit out their first string at different times, but teams often sit out key first stringers completely in a game (i.e. the Chargers with Tomlinson). Just another reason why there's no correlation between these stats in preseason vs. the regular season.





YAKUZA

 

[Sarge]

Is there any way you can just use pictures the next time?