Big 10 Holds Drawn Per Play vs Sack rating. Michigan is an Extreme Outlier.

Submitted by TTT on October 2nd, 2018 at 12:17 PM

Taken from Reddit:



Lanky Kong

October 2nd, 2018 at 1:27 PM ^

But wait, there's more! 

Using the data from , Michigan is #1 (or last, depending on how you look at it) in the FBS in opponent first downs from penalties with 20.

The FBS average is 8.92, with a 3.59 standard deviation, meaning Michigan is 3.08 standard deviations above the mean (a statistical outlier with a .001 probability assuming a normal distribution) 

Despite this, Michigan is still in the upper half of the FBS in opponent first downs. This means other teams out there with worse defenses are potentially out on the field longer, leading to more chances for penalty first downs to happen. This led me to also look at Percentage of Opponent First Downs from Penalties. No surprise, but Michigan is #1 again with a ridiculous 23% of our opponent's first downs coming from penalties.

The FBS average is 9.67%, with a standard deviation of 3.76%, meaning Michigan is 3.55 standard deviations above the mean (0.0002 probability in a normal distribution). 


fyi I'm not a stats wizard so be nice to me if I'm making bad assumptions. Thought this would maybe be worth making a thread for, but I have 0 points because I'm new and voting doesn't work anymore.


Lanky Kong

October 2nd, 2018 at 7:44 PM ^

Yeah it's obviously not perfectly normal, I said that smart ass. I didn't say it was literally a .0002 probability. It's clear though that those two teams are far, far away from the other 127 teams for whatever reason, which is all I've been trying to show. 

What're the odds that Michigan draws the fewest holding calls in the big ten (and it's not even close), while also having their opponents get the most first downs given to them by refs (and outside of one other team in FBS, isn't even close)? 

It's low.


October 3rd, 2018 at 7:48 AM ^

Like I said, what do I know. Maybe I picked something up in grad school or 30 years of working in statistics?

But, yeah... I’ll take your eyes’ word for it. 

Would one use an OLS regression model to predict this?  Likely not, as it would allow for negative predicted values. Poisson regression, however, would be appropriate for predicting a discrete/ count distribution. 

But, yeah, your eyes are probably right. 


October 2nd, 2018 at 12:23 PM ^

Now if a statistician wants to take this a step further they can evaluate whether the bias is more pronounced by in specific officiating crews. I.e. If the O'Neill crew is a lot more biased than the others Michigan should refuse to play in a game officiated by them.


October 2nd, 2018 at 12:24 PM ^

That Michigan is not getting the holding calls that they should is expected, but Jeebus, seeing this chart is shocking. This is complete injustice. 


October 2nd, 2018 at 12:27 PM ^

Does anyone have a statistic for holding calls that are declined?

If the defense gets to the quarterback and sacks him we would decline the holding.

Just want to see if that stat does anything to equal out the whole story.

With all that said I feel we're getting screwed.


October 2nd, 2018 at 1:12 PM ^

On average, teams defend about 70 plays per game (for 2017 offensive plays per game for 2/3 of D1 was between 65 and 75).


Indiana gets just under one hold per game called against opponents.

Michigan gets just over one half of one hold per game called against opponents.


Sample size is a SERIOUS issue with any attempt to analyze this. You really need a lot bigger data pool, both in terms of number of teams and number of years under consideration.