# Strange Lines?

Submitted by RayIsaac91 on October 23rd, 2009 at 1:39 PM
I was looking at Jeff Sagarin's BCS model here:

http://www.usatoday.com/sports/sagarin/fbt08.htm

I have limited amounts of time and knowledge of statistical models, so I am going to trust Jeff that his model is complete and accurate within reason (call me lazy, I prefer efficient with a desire to remain gainfully employed). Sagarin includes a predictor rating, which rates the teams based on all data. Since it is a slow week at work, I took the time to combine the betting lines with the predictor.

The expectation is that the predictor ratings, adjusted for the home field advantage of 3.5 pts, plus or minus the spread should equal zero. I am sure there is a noise factor in there (a threshold for statistical significance) but that is beyond me. I selected the games with the highest absolute difference from zero.

Idaho (predictor = 70.69) at Nevada (net predictor of 69.78) -15, difference = 16.41pts
WSU (59.20) at Cal (net 82.76) -35, difference = 11.44
N Illinois (75.46) - 11at Miami not that Miami (net (55.10). diff = 9.36
Maryland (59.05) at Duke (net 72.28) -4.5. diff = 8.73
Louisville (63.45) at Cinci (net 88.73) -17. diff = 8.28
Vandy (59.55) at S Car (net 80.95) -13.5. diff = 7.90

Games of local interest:
PSU (82.77) -4.5 at UM (net 81.65) diff = 3.38 (significant?)
Iowa (86.00) at MSU (net 81.10) pick 'em. difference of 4.90

There are 17 games with an absolute difference above 5.00. I assume this will get tighter as the games are played.

Notes:
• I do not have any knowledge of the Sagarin model or its corrections, adjustments etc.
• Home field advantage is a plug number, each team should have its own number.
• I pulled the lines from Yahoo sports, which may or may not be the latest.
• As a CPA, I can add and subtract, all other functions are considered immaterial or buried in a footnote in small print.
• This is for entertainment purposes only and all numbers should be verified.

Results this week:
Tulsa (72.55) -7 at UTEP (net 61.67).The play? Tulsa. The result? Loss.
FSU (78.49) at UNC (net 74.43) -2.5. The play? FSU. The result? Win.

Edit: I have not researched any external factors such as injuries or swine flu.

Go Blue!

My buddy likes to bet those spreads. Sometimes the big differnetials have to do with factors the computer doesn't calculate like injuries.

Cincy would be much higher if Pike wasn't hurt.
Maryland and Duke is because Duke is historically terrible no one would bet them laying that big of a number.
I'm guessing same for Idaho, it takes a while for people to become "believers"

I forgot to give the standard mutual fund warning that past results are not a guarantee of future performance or whatever that is.

I also agree with the Duke angle, I would have a difficult time laying points with a historically inconsistent team.

Idaho is interesting. It could be one of the teams that go 8-4 ATS which is great in the betting world. Play 'em before everyone catches on!

In the end, this method will probably be right 50% of the time, same as the monkey.

I use Jeff's ratings for my college pick up thing on Yahoo. I am winning the league as of right now. It's a good starting point. unfortunately he doesn't factor in things like "Texas Tech is starting their 3rd string QB" in his rankings (although he does like A&M with the points)

Shit. And I wanted to use this to bet. Illegally.

What are Sagarin's numbers? They can't possibly be score predictions because they're way too high. If they're not score predictions, then assuming the difference is Sagarin's predicted point spread doesn't make much sense.

There are still some points you can make there--Sagarin doesn't think Idaho is much better than Nevada, gamblers think it is, so there's a potential difference you could exploit if you believed Sagarin much more.

keeps his stuff black boxed, PECOTA style. And he's been at it for a while and publishing publicly...if all you had to do to beat Vegas was bet certain heavy Sagarin favorites (relative to the spread), Vegas would be broke.