Defensive Statistics and Player Attributes - Cont.

[Molarpistols]

Yesterday, I did some regressions examining the correlation between player attributes and their effect on the players total number of tackles. During the gathering of the data for that one, I also gathered data for most every other defensive statistical category.

Today, I'll examine Sacks and in a separate article, Interceptions/Pass Defense. For the sacks, I excluded the Defensive Backs, because most DBs didn't record many sacks, and including them would further cloud the already cloudy outputs from my regressions. Conversely, for Interceptions and Passes Defensed I excluded the Defensive Linemen, because it wouldn't add much to the analysis. I did include linebackers in both examinations because LBs spend a fair amount of time rushing the passer and in pass defense, leading to more statistical data points for my use.


Sacks




Sacks are somewhat rare, so we have a lack of data points sufficient to make this regression significant statistically. An F-Stat P-value of .121977 is actually better than I was expecting, this regression is almost significant if we use the 90% confidence interval that is low-end acceptable by some statisticians.

The regression above shows defensive lineman only. None of the coefficients are significant, but we can still glean some information from them. Speed has a high coefficient (yes, .249801 is high in this case. Basically it's saying that for every 5 points in speed you will earn one sack). Speed's coefficient is also nearly significant in terms of p-value, so the grain of salt we're taking with this can be a smaller grain. Strength, agility, and intelligence also help contribute to sack totals, though none are statistically significant.

The negative coefficients in front of the Tackle attribute and endurance seem pretty out of whack, they aren't significant so we know they can't be fully trusted, but the p-value is still lower than I would have expected. My logic is telling me that none of the attributes should have negative coefficients, because none of them should actually hurt your chances of making a sack.



Above is the linebacker output for sacks. Like just about every regression I will show today, the output is not statistically significant. Sacks being fairly rare makes it difficult to determine exactly which attribute is lending aid to the successful completion of the sack.

For linebackers, it appears that agility is pretty important to obtaining sacks. The coefficient of .56094 is quite high, and the p-value is low enough to be considered significant if we use the 90% confidence interval. I still generally use the 95% confidence interval, so I don't deem it significant, but it's still a pretty good indication that agility is important.

Other important attributes are speed and tackle, weirdly enough hands pops up here as well. None of the p-values are significant so we're definitely going to be over our daily sodium intake goals by a nutritional standpoint.



This regression is defensive lineman AND linebackers added together for sacks. The most interesting thing I noticed right away was the F-Stat p-value. This regression is statistically significant! Most of our coefficients are not, but the relationship as a whole is.

This output shows strength as being the most important for sacks, and is actually significant in terms of p-value. Agility isn't as important as it was for linebackers when taking the DL into account as well. The p-value isn't significant either. The speed coefficient has dropped into insignificance as well, but is still relatively high when including DL and LB.

Tackle is shown as negative and significant, which I find odd. I've included playing time in all of these, similar to last time, and we're still getting little correlation between playing time and statistics. Having only 16 linebackers in the data I think leads to a lot of this. In this simulator and in football in general, LBs eat up a lot of the statistics by cannibalizing off the work of their DL and DBs. The DL/DBs have lower statistics but the same time-played as the linebackers, making the relationship seem worse.


Overall, I'm not a statistician, I just enjoy analyzing what I can using my knowledge from my economics classes. I didn't do anything with stat weighting or anything, which I think would likely give us better results. There is also a pretty severe lack of data, 16 linebackers and 20 defensive linemen were included in the results, with only a single season of work between them. A few seasons down the line we might get a better glimpse of the player attribute effect on sacks.
GRADED

[timeconsumer]

Cool stuff. Your DL sack R-squared of .60 is pretty nice. It looks like you put a lot of work into the data collection on this.

I think if you want to improve your R-squared and ANOVA F-value you need to remove the attributes that we know are not significant and are watering down the overall regression. Change sacks to Str, Agi, Tackle, Speed. See how that changes your f-value and r-squared.

As far as the negative coefficient on stats like tackling that's pointing to a extraneous variable. Because you are examining NSFL stats we have to consider that more TPE spent in one area is less TPE in another. So a player who has spent more TPE on tackling is one who has spent less TPE in speed, strength, etc. It does not mean that tackling gives you fewer sacks, only that at our current levels of TPE the opportunity cost of picking tackling over a different stat is too great and resulting in lower production. However as the cost to get speed and strength up (say once guys are at 90) increases significantly, the opportunity cost of spending TPE on tackling could suddenly swing the results in the opposite direction.

[kckolbe]

Cool stuff. Your DL sack R-squared of .60 is pretty nice. I think if you want to improve your R-squared and ANOVA F-value you need to remove the attributes that we know are not significant and are watering down the overall regression. Change sacks to Str, Agi, Tackle, Speed. See how that changes your f-value and r-squared.

Do we know that int has no value on sacks? I mean, I expect it to be small, but since int seems important to o-line, it seems that an interior rusher might need it.

[timeconsumer]

Do we know that int has no value on sacks?  I mean, I expect it to be small, but since int seems important to o-line, it seems that an interior rusher might need it.

Where did you hear about it having an importance for o-line?

[kckolbe]

Where did you hear about it having an importance for o-line?

I said seems to be since blockers have higher int than a lot of positions. I don't know that it does anything, but it's there.

[timeconsumer]

I said seems to be since blockers have higher int than a lot of positions.  I don't know that it does anything, but it's there.

I hesitate to draw conclusions on relationships without tests. A blocker's average int being higher can be the result of many things that do not translate to production.

[Molarpistols]

@timeconsumer

Oh shit
I edited your post instead of quoting it. Always read the button you click boys, sorry about that.

Think I fixed it back to what it was. This is what I meant to say...


Intelligence is listed as "Medium" importance on the post linked in the rookie guide for OL.

As for removing the other variables that aren't important, I thought about doing it, but didn't for two reasons. The first being laziness. Second being, I was interested at getting a basic look at it, and honestly was hoping Int was a little more important for a lot of things.

With all the film study NFL'ers do, I was hoping the sim took that into account a little bit. Like a smarter player would be able to diagnose the play and get around the blocking scheme for the sack easier and whatnot. Wishful thinking, I know, but I did want a basic look at it all.

If I need more cash again, I might do it all over again and take out the extraneous ones just for a more accurate look. I still have all the hard data saved so it wouldn't really be that difficult.

[timeconsumer]

The "Medium" for OL int in the rookie guide does not mean what actually is important for that player. That is how the stat scales a player's Overall rating in the sim.

Unfortunately this really means nothing at all, because Overall is not important. I play this game and never look at a player's overall rating when drafting or signing players. I look at the stats I know to be important for that position. I have found a ton of success with an OT that has 35 int because his other stats are through the roof.

[kckolbe]

Okay, maybe it's a distraction, then. For what it's worth, though, O line has the highest average wonderlic score in the NFL.