Stringer's Power Rankings and STL @ MIA Preview

I'm not about to promote gambling on a public forum, lol.

 

Then put them in one of the other forums

 

I think that guy killed my father. If I find him, he should prepare to die.

 

Has it tested successfully in Vegas?

 

And not dumb enough to publicly post the map to El Dorado.

If this formula worked, Vegas would counter it as soon as 3904 dudes all started betting and winning the same way week-after-week for big money.

One guy, however... he could go on for a long while.

 

Stringer, your a meticulous one, and I trust your intelligence and research, so, I believe it because we beat two teams in the field of play, missed some field goals and got stuck with the losses, football is a 60 minute game, and if your beating those teams for the most part of those 60 minutes, that has to account for something.

If we make those field goals, that top 8 ranking doesn't look so weird does it?

 

Then when you run a partial correlation among winning percentage, the total model, and whatever part of the model specifically measures "randomness," the correlation between the total model and winning percentage should increase, with the variance in winning percentage associated with randomness being controlled for.

There's a way to get at the part I bolded in your post statistically. What you could end up with is a valid measure of underperformance due to negative, random events, commonly known as "bad luck."

Fineas would be interested in that. He's long been a proponent of the impact of bad luck on games and teams.

 

You're a madman Stringer...A madman I say.

 

The thing I like about this is it's not formulaic or systematic to where it can be "figured out" so to speak. But the down side to that is, if it's not formulaic or systematic then people tend not to believe it'll keep working, or at the very least there's an element of unpredictability and risk involved in it.

But hell, my real job is in asset management. I'm pretty used to the idea of people placing millions and billions of dollars of finances in the hands of individual people that they believe (hope) will make good decisions with predictable success rates, based on things like track record, process, intelligence, etc.

I'm not trying to claim I've found alchemy, here. What's worked for the first five weeks can easily go heavily south the last eleven weeks. It's no state secret, and I don't quite think it's a map to El Dorado. I just happen to be privy to the spread picks and weightings of a whole lot of amateur spread pickers with varying skill levels and three years' worth of track records. I've analyzed the data and I've come up with something that seems pretty predictive (so far) based on the behaviors of those people.

 

Sanchez is the Devil!

 

How do you establish the level of randomness in order to determine the weight?

 

Then why do you make sense of the fact that the correlation between the model and winning percentage is so much lower than the one between QB rating differential and winning percentage, for example, by attributing that difference to randomness? Isn't your model accounting for more randomness than QB rating differential? If you're "soaking up" more randomness in your model, why aren't you getting closer to a perfect correlation?

 

I guess what I'm getting at is there has to be a way to "check" on whether the "weakness" (i.e., comparatively speaking) of the correlation between winning percentage and your model is accounted for by randomness. For example, is the Denver Broncos' record relatively poor despite their high ranking based on your model because they're experiencing bad luck? I think you need a way of verifying that for your model to have the validity you want it to.

In other words, there needs to be an overall measure of "luck" (good or bad) IMO, and then the "weakness" of the correlation between your model (i.e., ability) and winning percentage should be explained by it.

 

Very fascinated by this. Usually don't give a shout about rankings, but this one intrigues me because there's substance behind it. It has it's kinks but I think it looks pretty good. Looking forward to see it develop.

 

From what I've noticed after yrs of watching football seasons is generally 2 games are won or lost by special teams.

INOW, every game carries a 12.5% chance that is the game ST wins or loses.

We've lost 2 games so far directly attributable to ST performance so in that sense we are below league average and it is likely to happen again as there has been no changes made on the ST unit.

In that sense, we simply must have a greater then 3 pt lead in the final 2 minutes or the Rams superior ST unit will likely win the game for them.

If we are down 3 or less in that situation we are in serious trouble depending on whether or not there is a kick off return.

 

What i love is that it's based on a wide view of many people's assessments and track records. That sort of triangulation (well, way more than just THREE points of view, actually) produce unusual accuracy if you can really understand how, when and why those people's successful picks overlap.

A human-based model is far more superior, imo, than a statistics-only model... but only if you choose the right people and the right interpretation of their behavior.

I don't remember much at all of stat analysis class, but I did a similar cross-referencing of draft pundits picks and mocks one year and won my draft pool handily. $50. It bought pizza at least.

 

Hah, yeah I mean the idea of tracking humans rather than statistics does appeal to me more. Humans process way more than a formula ever could. And because of that, they can create order out of complete chaos. The entire concept of candle stick trading is based on the predictability of a mass of humans. There are thousands of other valid concepts for stock trading that are based on mass behaviors, but candle stick trading is the oldest that I know as it's been documented since the 18th century in Japan.

I just thought about applying some of those concepts to football and how that would happen. The mistake would be trying to do that based on where the money is flowing. That's just too broad a stat. As a stock guy I look at those and I'm like...uhhh...wanna give me a little more than that please? Even a candle stick chart requires open, close, high and low figures.

So this year I get invited into an office pool that has like 350 people in it across the country, most of which have been playing in this pool for years, and I decided to try something very basic and see if it worked. I got their track records from previous years (no easy feat, btw) and used it to identify groups of people that I was interested in, and then I play picks off the interaction between those two groups.

But that's neither here nor there as far as the original topic. On the original topic, I'm actually going to give Stringer's system a little whirl (not for money or anything) and see if he's really got anything here. I'm interested in seeing if he does.

 

For comparison's sake, here's advancednflstats' efficiency rankings. It's got Miami ranked 5th overall, 15th on offense and 12th on defense. So based on that and Stringer's numbers, we are pretty efficient. We are just not scoring despite moving the ball well and are getting a bit unlucky.

 

The numbers are in...and they say to take Tennessee tonight with the points...with a 14 weighting. For those curious. Who knows if it'll be right or not. The top four picks for the week are Detroit, Seattle, Tennessee and Indianapolis.

 

Looks like the Tennessee pick worked out. Should be noted that a model derived off Stringer's ratings also picked the Titans pretty big. The model I put together off it had the Titans with a 12 weighting (3 to 16).

 

So, taking these rankings and a teams record you could predict the NFLs most and least lucky teams. lol

 

So far, this week is 88 of 133 with 4 points still up in the air...somewhere between 66% and 69% for the week.

On the downside, the simple system I put together real quick using Stringer's ratings...didn't do so well. It's got 32 of a potential 133 points, with 13 points still up for grabs tomorrow night. Somewhere between 24% and 34%. Ouch.

 

To me, the dolphins game did come down to that 13% randomness when ST tips the scales one way or another.

3 missed Fg's, and one fourth down conversion on a fake punt pretty much decided the game.

 

Thanks for the update. Will be interesting to track the progress of these rankings over the season....

 

So we beat the Rams and they jump up to 7th from 13, and we fall from 7th to 12th??

what formula is this, the BCS?

 

Take out the C

 

Bradford's passer rating was 91.2. Tannehill's passer rating was 111.9.

The Dolphins passed and defended the pass better than the Rams.

 

Does this mean you're using yards per pass attempt as opposed to QBR? Or just giving it more weight. I forget which correlated higher with winning when I crunched the numbers about a year ago.

 

Do you think much of the issue is that you're treating turnovers as largely random when they're so strongly correlated with winning?

 

I guess what is defined as an issue is determined by your overall goal of the model. What is it?

Let me ask you this as well: have you considered that there may be variables that account for the relatively weak correlations between some variables and themselves? In other words, you may find that there are variables that mediate a variable's correlation with itself in the future.

For example, you may find that turnovers, while random overall because they correlate weakly with themselves, become significantly more systematic and correlate with themselves much more strongly when you include mediating variables.