Dr. Bob -- sharpshooter fallacy?

[Rococo]

I am pretty dubious about Dr. Bob's "technical" analysis. I haven't looked at it closely, but I suspect that I would find a sharpshooter fallacy problem if I did. Has anyone looked at this issue?

[SunOfBeach]

I think it's virtually a given that the're some sort of data mining/snooping bias in his models. There almost has to be at least a subtle mining bias, assuming that he began by considering things which he already had heard/read may be usable.

But, to his defense, he seems to understand multicollinearity issues and whatnot, using factors to mitigate the problem (such as looking at rushing and ToP together). So perhaps he's taken some time to avoid the mining bias as much as possible.

Also, his track record has been pretty darned sweet, over a reasonably large sample... so that gives at least some evidence to support his models being robust.

[MicroBob]

If I'm understanding this correctly:
His models being robust overall doesn't mean that some of this stuff might not be occuring with some of his observations though.

Some of the stuff seems to be reaching a bit such as:

"how home favorites perform after winning straight up as a double road underdog."

I assume he means road underdog of 10 points or more.
I think that looking at such a stat in a vacuum without context can be dangerous although I agree that if one wants to try to use the argument that a team is likely to have a letdown after a big upset win the numbers evidently indicate that may be unfounded.
And I guess if the betting public unjustly believes in a let-down scenario then there is value in the line.

So such a stat could be justified but man would I look at this cautiously and try to find other reasons like scheduling or whatever to explain why some of that stuff happened.

Obviously some of this stuff is just going to come down to variance as well and I'm sure Dr. Bob understands that but some of the numbers he throws out make me wonder.

I was with a baseball team one season that had gone 2-10 on Tuesdays but was 11-1 on Wednesdays and my hunch is that it had nothing to do with anything like the whole team partying on Monday nights or that they would 'bounce-back' after their loss or anything like that. It was just a weird variance quirk.

But it feels like some of the trend-freaks of the sports-betting world might have detected the pattern and started to fade that team on Tuesdays and bet the house on them on Wednesdays.

[SunOfBeach]

Just read these Dr Bob gems in another thread:

[ QUOTE ]
I looked up how home favorites perform after winning straight up as a double road underdog. As it turns out teams in that situation are good bets and Auburn applies to a 36-6 ATS subset of a situation based on the premise.

[/ QUOTE ]

[ QUOTE ]
Vanderbilt, meanwhile, is playing their first road game after 4 home games and the Commodores apply to a negative 23-57-3 ATS situation that plays against road teams after playing 4 or more consecutive home games.

[/ QUOTE ]

[ QUOTE ]
The only negative is Vanderbilt’s 11-2 ATS mark as a road underdog the previous 3 seasons

[/ QUOTE ]

Ouch...

[NajdorfDefense]

[ QUOTE ]
The only negative is Vanderbilt’s 11-2 ATS mark as a road underdog the previous 3 seasons

[/ QUOTE ]

Ouch...

[/ QUOTE ]

I take that to mean that otherwise Vandy would be getting more points in Vegas in his eyes.

[King Yao]

I don't suscribe to Dr. Bob's plays, so I have little experience reading his write-ups or seeing his reasonings. But of the few I have read, he seems to cross-reference his data-mining with his power rating results, and uses the two together. In that sense, it isn't nearly as bad as other people that just data-mine without respect for the line (i.e., treating all favorites the same, regardless if they are -1 or -15...or if their power ratings show the fave should be a bigger fave or lower fave).

I wouldn't be thrilled to book Dr. Bob's plays.

[Post-Oak]

What I don't get is, how can somehow who purely relies on a mathematical approach be betting on games starting with Week 1?

I can't see how anyone can bet all of these early games if they are just using "data". The season just started.

A handicapper says things like 'Rutgers is better this year than last" in Week 3-4, but someone who relies on crunching data to create power rankings can't say such a thing.

If he really does have effective, computer generated power rankings, then maybe his picks will get better as the year goes on. Maybe it just wasn't feasible to tell potential customers "Wait until Week 8 and we'll have some great picks!".

[Rococo]

[ QUOTE ]
I think it's virtually a given that the're some sort of data mining/snooping bias in his models.

[/ QUOTE ]

I think that you are being charitable. I suspect that it is 95% datamining. Now sometimes the data collects around the bullseye, but not always, especially when the sample size is small, and the cutoffs are arbitrary.

I particularly love it when he says something like, "there is a strong angle for Week 3 bounceback home favorites laying 3 pts or less, but be careful if the line moves to 4, because the angle for bounceback home favorites laying 4 points or less is not nearly as strong."

[SunOfBeach]

I wonder how much of this "Team X covered at home in 12 of their last 14 following road losses of 7+" is marketing lingo for his squareish clients, who may not understand some of the more technical aspects of his model. If he's using some sort of regression model, I don't see why he'd reveal what's really driving his picks. It's reasonable that this just could be marketing blahblah, to give him something to say other than "my model, the contents of which I can't disclose, says team x should be favored by 6 while the line is only 2.5..." yada yada yada.

[SunOfBeach]

Yup... missed that.