Mike McCarthy's Analytics Fraud

Ranching

Well-Known Member
Messages
42,802
Reaction score
107,040
CowboysZone ULTIMATE Fan
it really does not matter whether a stats 101 level regression calculation is appropriate or not for run efficiency relevancy to passing efficiency.

there is a 14 person team working on analytics. there are a few coaches in the group, but it is highly doubtful they are the ones doing the regression or god forbid, developing the model. there must be some quants in the group that is doing the work.

it is highly doubtful that one year of analytics study will teach a few jogs that much about statistics. that does not make them a fraud. all you can hope for from even one year of intense serious study in analytics is an appreciation of importance of analytics. does anyone really expect someone (likely) with ZERO background in stats and math to truly get more than that. you have got to be kidding.

and having the coach understand the proper role of analytics is really all you need and want. this is not baseball where analytics play a bigger role, and the conditions are much more relevant for a more scientific approach for improving pitching and hitting. there are far more variables that cannot be accounted for in football.

so there is no need to argue whether run efficiency matters to pass efficiency. to interpret a simple regression model is stats 101. but really understanding the model and how flimsy the model is is actually something you would hire a phd to do.
Wow!!! I didn't know you were smart! I knew you were a smart*** but not smart!
 

percyhoward

Research Tool
Messages
17,062
Reaction score
21,861
would you run a play action if the defender is penetrating the backfield in 2 seconds. i dont think so. so you effectively have eliminated data points in which the defense is too dominant. gee, i wonder what happens to the pass efficiency when you eliminate some data points when the defense is too dominant...
There's no reason to assume that this would only affect passes with play action, but not affect passes without. If the QB expects a run blitz, for example, he might eschew the play fake and target the vacated area of the defense, thus improving the success rate on non-play-action passes.
 

percyhoward

Research Tool
Messages
17,062
Reaction score
21,861
Bottom line: Jerry wasn't going to hire anybody who didn't placate him about how vital Elliott is and how happy they are that he's locked up. And because of McCarthy's history of not prioritizing running or RBs at all, he probably had to make an even bigger deal about how wildly happy he is to have Elliott, just to overcompensate.
This is what I'd like to think, that it's either lip service or "coachspeak" designed not to give anything away. Maybe both. Even if he does plan to make the team less run-happy, he certainly has nothing to gain by admitting it now.
 

waldoputty

Well-Known Member
Messages
23,375
Reaction score
21,163
There's no reason to assume that this would only affect passes with play action, but not affect passes without. If the QB expects a run blitz, for example, he might eschew the play fake and target the vacated area of the defense, thus improving the success rate on non-play-action passes.

of course, a dominant pass rush would affect everything. it is not a matter of absolutes but a matter of biasing your data set.

play action requires the qb to take his eyes away for the precious fractions of a second or more. that time becomes even more precious if defenders routinely get in the backfield in for example 2 seconds. it takes time for the qb eyes to adjust and refocus to a different depth. it would be simply riskier to run play action. not saying never, but just far less likely. so you are biasing the data set by removing a lot of the data points where the defense is dominant. by removing some of the data points, that just made play action seem more effective than it really is.

that is only one example of simple self-selection bias. i remember my professor did all type of fancy (graduate-level) math to bracket the bias value for a single econometrics case. and that case was far less complex than football. and this is just one example and only 1 type of self selection. doing a real good statistical analysis of this issue is not simple even if it was possible.
 

garyo1954

Well-Known Member
Messages
3,704
Reaction score
4,470
does it in any way take account of the fact that teams would not run a play action under certain circumstances. for example, would you run a play action if the defender is penetrating the backfield in 2 seconds.

No in that case, it's much more fun to run a trap draw with the tight end and concuss the guy to the bench until he learns to leave my QB alone!
 

percyhoward

Research Tool
Messages
17,062
Reaction score
21,861
that is only one example of simple self-selection bias.
You could come up with multiple scenarios that make the success of play action more or less likely, and you could do the same for non-play action. That doesn't mean you simply throw up your hands though.
 

McKDaddy

Well-Known Member
Messages
8,224
Reaction score
8,425
Like I said.. McCarthy is like all of these big ego coaches that speak about all the changes they want to do from what they did before but the reality is that they are just doing whatever it takes to get another million dollar job.

McCarthy sold everyone some BS and a lot of people bought it.
Yeah, I feel that some of his "growth" is marketing talk. However, doesn't mean he won't be better than what we had.
 

DarthVander

Active Member
Messages
68
Reaction score
147
Look. I'm an engineer by training and now work at a big bank in investments. So yeah I like math and "analytics." I think they are instructive especially when careful study shows where what you believe to be true is in fact false. People who understand these counter-intuitive truisms claim a consistent edge.

There was much talk about how much Mike has learned in his year hiatus and how analytics was a big part of the thing he leaned on and has incorporated. And good lord he is not off to a great start. In fact it seems he is a fraud.

See below:



This is an awful quote. What he says here is the exact opposite finding from the analytics community.

In fact this is something I looked into when studying Zeke's pedestrian efficiency numbers last year (As a reminder we greatly overpaid Zeke after greatly over drafting him). But I digress....I posted this in August of last year...



What this shows is that play action passing is the most effective passes a QB can throw. And it hardly matters if you run the ball a lot. Or if you run the ball a little bit. And it does not matter if you run the ball well. Or if you run the ball poorly.

The point? If McCarthy is getting this basic thing wrong what confidence do we have that he actually learned anything in the last year, analytics or otherwise? And just like the Garrett era we can expect to give away the small edges that smart coaches understand and benefit from.

And that's a shame.


Hey Toruk,

Just an interesting random point I wanted to bring up. Did you know that the western banking system is pretty much out of gold? It’s simply running off paper.

There is an Eastern family that runs China, and they have over 16 Olympic swimming size pools worth of gold.

Certain insides believe there will soon be a switch and China will be running, what they claim to be, a more fair banking system worldwide.
 

waldoputty

Well-Known Member
Messages
23,375
Reaction score
21,163
You could come up with multiple scenarios that make the success of play action more or less likely, and you could do the same for non-play action. That doesn't mean you simply throw up your hands though.

these are not hypothetical scenarios and there is an obviously biased data set. i certainly would not be making sweeping anti-intuitive general statements based on it.

again, i am not privy to these models. though it sure seems like a simple regression with play action or no play action. i would expect much more detailed sophisticated analysis to be done rather than a simple regression with play action or no play action.

what that analysis is, i have no idea. but if they really have a 14 man analytics team, then they could look at such a problem because i would suspect there are at least a few quants there.

now if i were the cowboys, would i be throwing resources at such a question. i dont think so. because what would such an analysis do for you in games? would it really affect your playcalling? probably not much.

i would probably spend resources in much narrower questions myself. such as tendencies of each team/coach that we will be playing. again football is a very complex situation for statistical analysis due to the complex interactions of 22 players. intuitively, the narrower you can focus your analytics, the more likely you will get something useful. the more you can reduce it to a situation like baseball pitching and hitting, the more likely you can apply whatever you figure out.
 

garyo1954

Well-Known Member
Messages
3,704
Reaction score
4,470
such as tendencies of each team/coach that we will be playing. again football is a very complex situation for statistical analysis due to the complex interactions of 22 players. intuitively, the narrower you can focus your analytics, the more likely you will get something useful.

Agreed.
 

jterrell

Penguinite
Messages
33,428
Reaction score
15,605
Look. I'm an engineer by training and now work at a big bank in investments. So yeah I like math and "analytics." I think they are instructive especially when careful study shows where what you believe to be true is in fact false. People who understand these counter-intuitive truisms claim a consistent edge.

There was much talk about how much Mike has learned in his year hiatus and how analytics was a big part of the thing he leaned on and has incorporated. And good lord he is not off to a great start. In fact it seems he is a fraud.

See below:



This is an awful quote. What he says here is the exact opposite finding from the analytics community.

In fact this is something I looked into when studying Zeke's pedestrian efficiency numbers last year (As a reminder we greatly overpaid Zeke after greatly over drafting him). But I digress....I posted this in August of last year...



What this shows is that play action passing is the most effective passes a QB can throw. And it hardly matters if you run the ball a lot. Or if you run the ball a little bit. And it does not matter if you run the ball well. Or if you run the ball poorly.

The point? If McCarthy is getting this basic thing wrong what confidence do we have that he actually learned anything in the last year, analytics or otherwise? And just like the Garrett era we can expect to give away the small edges that smart coaches understand and benefit from.

And that's a shame.

I am also an engineer that evolved to process engineer and project manager before just manager.
Analytics nerds for the NFL need to take many deep breaths and go back to work.

They have not yet managed to provide much more than raw data and some assumptions.

The top 6 teams in the NFL in time of possession were in the playoffs. Top rushing teams in the playoffs. Top passing teams, largely at home.
Reality is goofballs need to stop.
If you can make it to the AFC CG with Ryan Tannehill then passing isn't so important as we have been chirped to by the Analytics community.

In the NFL there has always been a cause and effect of what the offense does versus what the defense does.
There is also a very real time component.
EPA (analytics main football brainchild) at this point has missed the boat largely (as the master metric) because it does not sufficiently address time.

We should have analytics nerds challenging each other harshly instead of this weird nerd herd mentality vociferously defending what is largely nonsense.

PER TEAM analytics are VERY useless and helpful.
Knowing a defense likes to play a set defense in a set situation is huge.
Same in reverse.
But saying well there's a 65% chance of success with the pass here is dumb. The Defense has that same number and knows to play pass defense.

Throwing it on 1st down all game because EPA says so is just stupid and you should be beaten and dumped into a toilet like the nerds of yore.

Every project I have hated was led by some goofball thinking they understand data. Then they made changes that were worse than the original while wasting time and money. That money usually equates to jobs and people lost. That usually equates to the people still around working twice as hard with more duties.

Did Derrick Henry complete the TD pass because the play action was effective? Yes. Did they fully expect DH to run? Yes.
Would play action work if you didn't run it AT ALL? NO, lol.
Would it work if you didn't run it great? Yes. Because you are still going to have been taught certain techniques and have muscle memory.
But you are much more likely to have an extra defender in the box if you are running well the defense has to adjust before you hit them with play action.
 

percyhoward

Research Tool
Messages
17,062
Reaction score
21,861
again, i am not privy to these models. though it sure seems like a simple regression with play action or no play action. i would expect much more detailed sophisticated analysis to be done rather than a simple regression with play action or no play action....i would probably spend resources in much narrower questions myself. such as tendencies of each team/coach that we will be playing.
I'm sure that will be case as it presumably is for every team, but in a study of whether run frequency/success affects play action success then obviously you have to look at the whole league, and preferably over several seasons -- which was exactly what Baldwin did in those three studies posted above.
 
Top