Regression silliness

[Justin7]

I'm running regressions on interceptions as a function of rushing attempts and passing attempts. I hypothesize that more pass attempts means more interceptions, and a lower ratio of pa to ra will reduce ints...

Crank out 5 years of data. What is the best fitting curve for home teams?
INTs=1. Don't bother looking at any statistics. Just go with 1 pick.

I find all kinds of things I find amusing, but this really made me laugh.

[G's pks]

Justin you are not allowed to use your points right? Send me some...

[fiveteamer]

LOX

[CarpeDime]

I'll guess .7

[mathdotcom]

What do you expect when you have only 2 explanatory variables?

In this case the obvious missing explanatory variable is "tendency to make risky throws".

[flyingillini]

LOX

and Bagels.

[MonkeyF0cker]

What do you expect when you have only 2 explanatory variables?

In this case the obvious missing explanatory variable is "tendency to make risky throws".

Yeah. That's quantitative. Nobody will ever accuse you of being a genius.

[mathdotcom]

Yeah. That's quantitative. Nobody will ever accuse you of being a genius.

Ah so because it can't be measured, it's alright to ignore the effects that it has?

You really need to take a few stats classes pal.

[MonkeyF0cker]

You have to be the biggest fukkin joke on this forum. Go waste hours upon hours upon hours of your time trying to find correlation in some ridiculous subjective variable by rating the "riskiness" of every pass thrown by every single quarterback. There are far better predictors. That's probably one of the dumbest things anyone who pretends to have a clue has ever said in this forum. Do you and Nicky take pills to be as fukkin dumb as you are?

[themajormt]

[SlickFazzer]

[Justin7]

What do you expect when you have only 2 explanatory variables?

In this case the obvious missing explanatory variable is "tendency to make risky throws".

That's a very interesting idea. 3d and 20 - do you run a draw, or throw a risk pass into dime coverage? Are most of your passes short to backs and tight ends, or are you throwing at wide-outs into double coverage?

thanks for the idea.

[MonkeyF0cker]

Are you building a sim, Justin? That's much more of a coaching tendency than a quarterback...

[mathdotcom]

You have to be the biggest fukkin joke on this forum. Go waste hours upon hours upon hours of your time trying to find correlation in some ridiculous subjective variable by rating the "riskiness" of every pass thrown by every single quarterback. There are far better predictors. That's probably one of the dumbest things anyone who pretends to have a clue has ever said in this forum. Do you and Nicky take pills to be as fukkin dumb as you are?

Wow your reading skills are truly horrible.

Monkey, if you have the audacity to even reply to this message, show me where I said that Justin should try to rate the riskyness of each pass thrown by each quarterback. I never said any such thing. Where? Where?

He ran y = f(x1,x2). I say the true model is actually y = f(x1,x2,x3) where x3 is highly relevant. Please show me where I said he should try rating the riskyness of each pass. I will give you all of my SBR points if you do. All I said was that given y = f(x1,x2,x3), running y = f(x1,x2) is necessarily going to give you "funny" results.

Monkey if you like I can recommend some great introductory undergraduate-level books that explain omitted variables bias and the problem it poses for OLS regression.

I am glad my 5000th post was illustrating what a hack you are.

[G's pks]

now everyone in this thread send me points... silliness...

[MonkeyF0cker]

LOL. How else are you going to quantify it? The problem is that your "model" is probably worse. Not only is it not quantifiable unless you subjectively rate every single pass but "risky" passes are often not the cause of interceptions.

[Justin7]

LOL. How else are you going to quantify it? The problem is that your "model" is probably worse. Not only is it not quantifiable unless you subjectively rate every single pass but "risky" passes are often not the cause of interceptions.

Quantifying it is the key. There are objective tests that can be evaluated using a play-by-play log.

[MonkeyF0cker]

Justin, you should be looking at objective statistics like play selection, score, hurries, pressures, sacks, etc. Riskiness is a horrible idea. The best way to incorporate those variables is via an ** sim in my opinion.

[Justin7]

Justin, you should be looking at objective statistics like play selection, score, hurries, pressures, sacks, etc. Riskiness is a horrible idea. The best way to incorporate those variables is via an ** sim in my opinion.

Is Riskiness any worse than a karma rating? My objective measures of past karma, and ability to predict reversion of the mean for this has worked well in the past.

A QB risk index is a brilliant idea - it might be able to differentiate lucky and unlucky quarterbacks that appears the same from a box-score stat based analysis. I'll back the unlucky low-risk QB in most matchups (or at least make a positive adjustment for his team).

[MonkeyF0cker]

How would you account for a tipped pass at the line of scrimmage? Is the pass being tipped luck? Or is the ball being deflected to another defender considered luck? And how do you quantify that?

[mathdotcom]

Wow your reading skills are truly horrible.

Monkey, if you have the audacity to even reply to this message, show me where I said that Justin should try to rate the riskyness of each pass thrown by each quarterback. I never said any such thing. Where? Where?

He ran y = f(x1,x2). I say the true model is actually y = f(x1,x2,x3) where x3 is highly relevant. Please show me where I said he should try rating the riskyness of each pass. I will give you all of my SBR points if you do. All I said was that given y = f(x1,x2,x3), running y = f(x1,x2) is necessarily going to give you "funny" results.

Monkey if you like I can recommend some great introductory undergraduate-level books that explain omitted variables bias and the problem it poses for OLS regression.

I am glad my 5000th post was illustrating what a hack you are.

Monkey, still waiting for you to reply with where I said he should try to include a measure of riskiness. Where did I say that? I am still waiting for you to back up your claims. Either show me where I said he should try to rate the riskiness of passes thrown, or admit you're full of shit as usual and made it up. Do you not want 300 SBR points in exchange for just highlighting where I said he should try to rate the riskiness of passes thrown? Let me know if you're busy and I can bump the thread again tomorrow.

[mathdotcom]

Just because I say he has an omitted variable does not mean I think he should try to use some necessarily bad/subjective measure of it.

Monkey I am aching to send you 300 sbr points.

[MonkeyF0cker]

Let's argue semantics now. That sounds delightful. The point is that it's not a quantitative variable, it's hardly encompassing of interception rates, and you yourself couldn't even prove that such a variable is correlated. End of story.

P.S. I don't care about fukkin points, son.

[mathdotcom]

Son,

You are confusing the regression model with the theoretical model. The theoretical model looks something like this:

# of interceptions = f(quality of other team's defense, quarterback's talent, how many passes are thrown, etc.)

The only explanatory variable in that equation that is observable is how many passes are thrown. The quarterback's talent is not quantitative, nor is the quality of the other team's defense. But they are explanatory variables that we can all agree affect the number of interceptions. You can try to throw in a bunch of observable variables to account for those variables, and can argue indefinitely about which are appropriate. But if you cannot find any variable to account for the quarterback's innate tendency to throw interceptions, or to make wild throws, or to take chances, then your model is going to suck. I'm not suggesting you use the quarterback's height as a proxy since a taller quarterback may have a better view of the field, or some b.s. like that. Just stating the fact that if you don't even attempt to account for a (highly) relevant explanatory variable, your model is necessarily going to give you "funny" output.

This is not semantics. Here's a simpler example for your horseshit brain:

Suppose you run: income = f(age, education) and try to say anything about the effect of education on income. You will get laughed out of the room. Why? Because you've omitted a relevant variable: intelligence. Nobody in that room will claim there is an accurate measure of intelligence, but the fact that intelligence is so highly correlated with one's level of education is going to hugely bias your results. If no one can come up with a decent measure of intelligence, then you're just SOL. That is all I said about Justin's model.

[MonkeyF0cker]

So you actually think that sports regression models include every explanatory variable? LOL. Wow. As I stated in a previous post, there are objective variables that are predictive for interceptions. And yes, there are ways to measure a quarterback's accuracy and efficiency (or talent as you put it) via objective statistics. Whether you were simply restating the obvious, that Justin's model was flawed either theoretically or functionally, doesn't really matter. And now I'm bored.

[mathdotcom]

So you actually think that sports regression models include every explanatory variable? LOL. Wow. As I stated in a previous post, there are objective variables that are predictive for interceptions. And yes, there are ways to measure a quarterback's accuracy and efficiency (or talent as you put it) via objective statistics. Whether you were simply restating the obvious, that Justin's model was flawed either theoretically or functionally, doesn't really matter. And now I'm bored.

Where are you getting this from??

Is there somewhere online where I can challenge you to a reading comprehension test?

Every one of your posts starts with a claim you can't back up (see above for another example if you've forgotten). Try googling "true regression model" or "theoretical regression model". That is what I'm talking about. You're talking about the line you type into stata.

[MonkeyF0cker]

But if you cannot find any variable to account for the quarterback's innate tendency to throw interceptions, or to make wild throws, or to take chances, then your model is going to suck.

This is the sentiment I was alluding to. You're certainly not going to get 100% true probability for every statistic. However, the closer you are to it, the greater your edge is over the market.