Angles and Numbers

[statman]

Since this is my first post (and I first posted it as a reply), I figured I start a thread with it and see if anyone was interested....

There IS a way to use angles that gets beyond the flawed approach of just tracking the record of angles vs the line. And, that is to track the record against an unbiased (for example power ratings based) prediction. It's crucical to do so... here's why (and NBA example).

Years ago, you could print money week after week playing against teams that played 4 games in five nights. But eventually, the linesmaker caught on and adjusted the opening numbers- removing the value. Here is the problem that created:

1) Angle cappers saw their w/l % go to 50% and stopped playing the method.
2) Pure stat players started seeing overlays on these games (since the lines were now preadjusted for these "angles/situations" and were "biased" from a pure power rating based line) and started playing games that really had no edge. (And so they saw THEIR results get worse too).
3) Teams playing 4 games in 5 nights still did underperform their power ratings, but no one knew what to do with them.

So, if you can, use a computer to make an unbiased stat based prediction on every game. Then, test your angles against that prediction and not the line. If there is a consistant result because of the angle (the teams fail to cover 70% of the time, by an average of 4.5 points...), then start to add these adjustments to your numbers.

INHO you cannot use numbers alone to win consistantly, nor angles... you must combine the two. Any Thoughts?

 

[count zero]

Sounds good, but I don't think blending angles and stats is as easy as you imply.

First, you'd have to find a way to quantify the various intangibles that make up the typical angle. And you can't just ask how many points is it worth to be a home team on MNF, you must ask how much of that is due to being at home, how much to being on MNF? What if the team is a favorite, or a dog? All these factors are interrelated -- they potentiate each other and/or they inhibit each other, and you have to guess how, and what it's all worth. Just use the historical average, as you suggest? Never work, at least not in the NFL -- there simply aren't very many angles that have an average covering margin that's worth anything. Most are in the point-something area. A few variants of the one I just mentioned (home teams on MNF) are historically worth a couple of points, but none is anywhere near the 200-300 trials that would legitimize going with it unquestioningly or assuming that it really is worth 2 points. And even if one is, there aren't many others, believe me -- unless of course you're willing to relax your standards for statistical significance (which, knowing you, I'm guessing you are).

And all that is just with one angle. What happens when you have 5 or 6 angles, some pulling toward one team and some toward the other? What happens to your MNF home team if it's up against a divisional underdog? MNF divisional dogs are a powerful subgroup as well. So you basically end up having to do a complete analysis of every possible intangible factor that could occur, from the simple fact of being at home or being an underdog to complex situations like being a turf team in your third road game on grass after losing the first 2, then finding the relationships between all these things and how they affect each other, and then somehow putting a number on it. At this point, the concept of angle shooting has disappeared and you're basically back to just trying to understand the intangible aspect of the game. From this perspective, angle-shooting should be seen as a failed (because simplistic) attempt to treat intangibles quantitatively.

I used to fudge my stat output to account for situations, which is a crude way of combining stats and angles, I guess. But I found I did better by just playing the stats -- I couldn't guess when the angle was going to assert itself, and when it didn't, it was hurting me more than it helped me when it did. I still feel I need to incorporate situations/emotion etc into my 'capping in a quantitative way, but don't think I have what it takes (or even know what it takes) to do that. Will gladly accept anything you come up with, but I don't think it will be easy.

[This message has been edited by count zero (edited 11-02-2001).]

 

[statman]

Hmmm.... do we know each other? Or are you assuming from my post that I am willing to compromise statistical significance?

You bring up some good points. Especially about the difficulty of having separable situations. Let me try to address them:

1) You don't need to have hundreds of results to get statistical significance. You just have to know (using something like chi squared testing) that a method that is 18-3 is only as strong as one that is 60-40. And you can scale the observed point adjustment by that confidence.
2) I don't agree with the idea of trying to avoid overlap between angles by quantifying all the subcomponents (home, dogs, rested, off a win....). I prefer a method that categorizes angles (emotional, physical, matchup,...) and only allow a certain number of points to build up in a single category if there are multiple situations that apply.

In the end it IS a bit subjective. But, I think, better than a numbers only approach that completely ignores relatively well understood situational adjustments.

To be safe, you could insist on an overlay on both the pure numbers, and the fully adjusted numbers.

Whatdyathink?

 

[count zero]

OK, requiring both sounds good, but guessing that would have an undesirable impact on the number of plays.

Think I remember a discussion of regression in a statistics text that mentioned regression could be used to figure the degree to which each of the various elements in, say, a typical angle contributed to the result. Could that be pertinent here?

Yes, we know each other from another board.

 

[bear]

Statman....Count Zero
So whos gonna win the freakin football games

Only kidding......Appreciate the analysis and the intellectual approaches you guys take.
bear