Mark Farley Thinks His 5-Loss Team Should Finish Top-5

[AZGrizFan]

Top 25...not just your home conference.

If you can't get at least 80% of your own conference you're a moran.

Also ran a conference pick em this year in the mvfc and average was 85%.

Much easier to pick what you're familiar with.

A computer can't take into account injuries, field conditions, team momentum, match up issues, etc...

The fact is that for a computer to have the same, or better %, as humans tells me quite a bit about human picking. Humans should have an advantage that can't evev come close to being matched by a computer.


5% over the course of thousands of games makes a HUGE difference

Nice work, Clenz. If UM spent any money on academics, Z might just get this.

Why do you care? You don't post in the football threads here anyway...

[JohnStOnge]

[quote="Mvemjsunpx"]I'm not sure how this thread got on this subject, but I've picked every game involving a DI team (including games against non-DIs) straight-up since 2004. Here are the figures for those that want some sort of idea how these numbers look:

Disclaimer - Somewhere in there (2008 on, I believe), I started counting postseason games (incl. conference championship games) as worth double.[quote]

I think you're good but I also don't see why you have so little confidence in good power rating systems. While it doesn't provide a direct comparison you can go to that site I linked and see that the Sagarin system, for instance, has achieved results very similar to your own. The percentages I added just after yours are the Sagarin system's percent correct for games for which lines were established. I don't know if I could find out for sure but that probably means games between I-A/FBS teams. My guess is that your percentages would drop a little if you look only at games between I-A/FBS teams because there won't have been as many total mismatches one would expect to get when one picks all games involving any Division I teams. When you pick all games involving Division I teams you're picking games like Savannah State at Oklahoma State and Wisconsin Oshkosh. For example you could just pick the I-A team without knowing anything at all about any teams in I-A/FBS vs. I-AA/FCS matchups and be right somewhere around 90% of the time. That's going to pad your winning percentage.

Here are your percentages followed by Sagarin's I-A/FBS vs. I-A/FBS percentages for each year:

2004 - 1000-396 (71.6%) 75.6%
2005 - 990-409 (70.8%) 69.3%
2006 - 1072-412 (72.2%) 76.1%
2007 - 1068-384 (73.6%) 71.5%
2008 - 1178-412 (74.1%) 73.7%
2009 - 1133-416 (73.1%) 72.7%
2010 - 1142-420 (73.1%) 72.8%
2011 - 1164-408 (74.0%) 73.4%
2012 - 1158-403 (74.2%) 76.2%
2013 - 1247-412 (75.2%) 76.7%
2014 - 1258-396 (76.1%) 74.3%

If we were to test you against the system for fun we would also look at bias and absolute error. And we'd have to test the differences between what you did and what the system did for statistical "significance" to see if we could say that there is sufficient evidence to say either you are the system is better at predicting. My guess is that if your percentages picking just I-A games are close to what the ones you listed for picking all games involving D-I teams are there's a good chance we wouldn't see a "significant" difference between you and Sagarin's system in terms of percent correct. Where I'd expect you to have more trouble is bias and absolute error. That's if you don't "cheat" by doing things like looking at the line or any power rating systems to inform yourself as to what the predicted spread to be. If you want an honest comparison you can't do that. You have to develop your predicted margins of victory on a completely independent basis.

[JohnStOnge]

The fact is that for a computer to have the same, or better %, as humans tells me quite a bit about human picking. Humans should have an advantage that can't evev come close to being matched by a computer.

Yes that's what amazed me when I first started looking at this sort of thing during the 1990s by tracking all the game outcomes myself, looking at the lines in the local paper on the mornings of game day, and looking at the Sagarin and Dunkel Index predictions. I looked at the line as a measure of the opinions of human beings who at least think they know football. We all know how that works so I think you can see that's a reasonable interpretation.

So we have lines being "adjusted" as people take into account things like weather, the types of offenses, matchups, injuries, and so on vs. mathematical models that just look at scores and don't "know" about any of that stuff yet the models did about as well as the line did. The line did do very slightly better but not to a statistically "significant" extent. It was that way with respect to percent correct as well as bias and mean squared error (or you can look at absolute error). That, to me, says a LOT for how good those models are.

But the models to have one advantage over human beings. Once they get to a certain point of the season they can instantaneously, simultaneously, and comprehensively take into account every result of every game played to date in a systematic way. No human brain can do that. Well, I probably shouldn't say "no" because I've seen some pretty amazing stuff on TV; especially involving the "idiot savant" phenomenon and mathematics. But if there are ANY human brains that can do that there aren't many.

[JohnStOnge]

BTW the reason I did that during the 1990s is I got into a debate with a guy who hosted a pick 'em radio show. He studied vegas line stuff all the time and was/is viewed locally as an expert football prognisticator. And of course he's a big SEC homer. I was telling him the Big 10 was a tougher conference at the time based on Sagarin Conference strength ratings. He scoffed. Made fun of the idea.

So I issued the challenge. I told him I would randomly select a set of I-A vs. I-A games over the upcoming season. Then each week I would call him, give him the games for that week, and get him to give me his pick to win and by how much. He agreed.

Both Dunkel and Sagarin beat him in percent correct and REALLY creamed him in terms of getting closer to the actual margins of victory on average and bias.

Of course it made no difference in the end. Even though he'd just been beaten by both systems he still scoffed at them. Such is human nature.

[clenz]

42 of 43 ESPN analysts picked Alabama to beat Ohio State http://bit.ly/1EUvC0S" onclick="window.open(this.href);return false;

[clenz]

On the topic of the subject of the thread...

Another computer ranking, done by a guy who lives in Iowa that I've really enjoyed for the better part of a decade.

http://ncaa.bcmoorerankings.com/fb/2014 ... BRank.html" onclick="window.open(this.href);return false;

R Team Conf ( W- L) Mod Str Sched Rank Off Ave Def Ave
1 North Dakota St Miss (14- 1) 149.23 5W 128.47 ( 7) 33.20 ( 25) 13.20 ( 2)
2 Illinois St Miss (13- 1) 144.53 6W 125.74 ( 11) 39.07 ( 9) 20.93 ( 19)
3 Northern Iowa Miss ( 9- 5) 142.65 1L 131.70 ( 5) 32.29 ( 32) 19.93 ( 15)
4 South Dakota St Miss ( 9- 5) 138.51 1L 132.19 ( 4) 31.79 ( 39) 25.36 ( 50)
5 Chattanooga Sout (10- 4) 137.92 1L 122.35 ( 18) 34.71 ( 21) 19.36 ( 11)
6 New Hampshire Colo (12- 2) 134.81 1L 119.48 ( 32) 36.00 ( 17) 21.21 ( 23)
7 Villanova Colo (11- 3) 134.51 1L 117.45 ( 45) 37.71 ( 13) 22.14 ( 29)
8 Coastal Carolina Big (12- 2) 134.03 1L 117.19 ( 49) 37.29 ( 14) 19.57 ( 13)
9 Indiana St Miss ( 8- 6) 133.89 1L 133.33 ( 2) 25.29 ( 75) 24.21 ( 44)
10 Jacksonville St Ohio (10- 2) 133.89 1L 121.03 ( 26) 35.17 ( 20) 21.67 ( 26)


He does FBS, FCS, D2, D3, NAIA rankings for teams and conferences.

Also does Iowa high school boys and girls basketball, football, baseball, and wrestling
http://bcmoorerankings.com/" onclick="window.open(this.href);return false;

He's accurate enough that there is a push from a decent number of people in Iowa to go away from the way they seed the state tournaments to using his rankings to seed teams

[D1B]

Go ahead Clenz, you can tell people it's me.

[Mvemjsunpx]

I think you're good but I also don't see why you have so little confidence in good power rating systems. While it doesn't provide a direct comparison you can go to that site I linked and see that the Sagarin system, for instance, has achieved results very similar to your own. The percentages I added just after yours are the Sagarin system's percent correct for games for which lines were established. I don't know if I could find out for sure but that probably means games between I-A/FBS teams. My guess is that your percentages would drop a little if you look only at games between I-A/FBS teams because there won't have been as many total mismatches one would expect to get when one picks all games involving any Division I teams. When you pick all games involving Division I teams you're picking games like Savannah State at Oklahoma State and Wisconsin Oshkosh. For example you could just pick the I-A team without knowing anything at all about any teams in I-A/FBS vs. I-AA/FCS matchups and be right somewhere around 90% of the time. That's going to pad your winning percentage.

Here are your percentages followed by Sagarin's I-A/FBS vs. I-A/FBS percentages for each year:

2004 - 1000-396 (71.6%) 75.6%
2005 - 990-409 (70.8%) 69.3%
2006 - 1072-412 (72.2%) 76.1%
2007 - 1068-384 (73.6%) 71.5%
2008 - 1178-412 (74.1%) 73.7%
2009 - 1133-416 (73.1%) 72.7%
2010 - 1142-420 (73.1%) 72.8%
2011 - 1164-408 (74.0%) 73.4%
2012 - 1158-403 (74.2%) 76.2%
2013 - 1247-412 (75.2%) 76.7%
2014 - 1258-396 (76.1%) 74.3%

If we were to test you against the system for fun we would also look at bias and absolute error. And we'd have to test the differences between what you did and what the system did for statistical "significance" to see if we could say that there is sufficient evidence to say either you are the system is better at predicting. My guess is that if your percentages picking just I-A games are close to what the ones you listed for picking all games involving D-I teams are there's a good chance we wouldn't see a "significant" difference between you and Sagarin's system in terms of percent correct. Where I'd expect you to have more trouble is bias and absolute error. That's if you don't "cheat" by doing things like looking at the line or any power rating systems to inform yourself as to what the predicted spread to be. If you want an honest comparison you can't do that. You have to develop your predicted margins of victory on a completely independent basis.

I wasn't necessarily posting this to compare myself to computers, just to give others an idea of what a solid human % might look like, but…


The Sagarins rank all DI teams, both FBS & FCS, so I assume he picks both (& versus each other). He also has a generic placeholder for non-DIs (for SOS purposes) I've never checked out his picks or how he does them so I can't comment on that.

My issue with computers for football is more the rankings than the picks—there just aren't enough games to make it as accurate as basketball or whatever. Most head-to-head picks don't involve teams that are really close to each other, so that's a bit simpler than a top-25. What the numbers you posted show me is that the Sagarins aren't really any better than me, and I'm pretty sure there are some people that have a higher percentage than I do.

[SDHornet]

My issue with computers for football is more the rankings than the picks—there just aren't enough games to make it as accurate as basketball or whatever. Most head-to-head picks don't involve teams that are really close to each other, so that's a bit simpler than a top-25. What the numbers you posted show me is that the Sagarins aren't really any better than me, and I'm pretty sure there are some people that have a higher percentage than I do.

Don't lie, no one has a higher % than you.

Clenz I'll drag you along the next time I go to Vegas, the sports book winnings should cover the trip.

[Mvemjsunpx]

My issue with computers for football is more the rankings than the picks—there just aren't enough games to make it as accurate as basketball or whatever. Most head-to-head picks don't involve teams that are really close to each other, so that's a bit simpler than a top-25. What the numbers you posted show me is that the Sagarins aren't really any better than me, and I'm pretty sure there are some people that have a higher percentage than I do.

Don't lie, no one has a higher % than you.

Clenz I'll drag you along the next time I go to Vegas, the sports book winnings should cover the trip.

Zzuh?

I mean, clenz does kinda come off as an evil version of me, but still…

[SuperHornet]

clenzy can be kinda rough with me, but he's not the REAL evil one here. He actually has cogent arguments. The evil pair just resort to ad hominem attacks. clenz at least brings a good argument whether he gets verbally violent or not. Plus, clenzy is our resident SME for all things officiating. It's great to have a guy like that around here....

[Cap'n Cat]

clenzy can be kinda rough with me, but he's not the REAL evil one here. He actually has cogent arguments. The evil pair just resort to ad hominem attacks. clenz at least brings a good argument whether he gets verbally violent or not. Plus, clenzy is our resident SME for all things officiating. It's great to have a guy like that around here....

You're a cunt and I'm banging your sister.

[JohnStOnge]

Most head-to-head picks don't involve teams that are really close to each other, so that's a bit simpler than a top-25. What the numbers you posted show me is that the Sagarins aren't really any better than me, and I'm pretty sure there are some people that have a higher percentage than I do.

My guess is that if you were to look at your rates for just the games considered on that site I linked your percentage would decline some. What you see on that site is just games for which vegas lines were established. So mostly if not entirely FBS vs. FBS. That means the FCS vs. FBS games are not, for the most part, considered in the summary. Also the FCS vs. D-II games are not included in the summary. And even though I've said many times that top D-II teams can compete with even top FCS teams, overall FCS teams win the overwhelming majority of such matchups and most of them are easy picks.

A measure you can look at that is independent of how close teams are to each other is the mean squared error. That's also in the report. A related measure is the absolute error. That's the absolute value of how far off the predicted margin of victory is from the actual margin of victory.

It seems reasonable to me to say that if you rank teams the test of how good you are at doing that is how frequently you are correct about what's going to happen when two teams play. The thing that got me into looking at this was contemplation of what the difference would be between rankings in a top 25 poll and rankings produced on the basis of who would be favored to win if teams played each other.

So I started looking to see how close various power rating systems are to the vegas line in order to identify proxies for ranking according to the vegas line. And certain systems compare pretty well with the vegas line. Sagarin is one of them. Overall the line is, I think, a little better. That's especially true when you look at mean squared error or absolute error. But the difference is not great.

The "small sample size" concept applies no matter what approach one takes to ranking college football teams. It applies whether it's human opinion or power ratings.

[Mvemjsunpx]

Most head-to-head picks don't involve teams that are really close to each other, so that's a bit simpler than a top-25. What the numbers you posted show me is that the Sagarins aren't really any better than me, and I'm pretty sure there are some people that have a higher percentage than I do.

My guess is that if you were to look at your rates for just the games considered on that site I linked your percentage would decline some. What you see on that site is just games for which vegas lines were established. So mostly if not entirely FBS vs. FBS. That means the FCS vs. FBS games are not, for the most part, considered in the summary. Also the FCS vs. D-II games are not included in the summary. And even though I've said many times that top D-II teams can compete with even top FCS teams, overall FCS teams win the overwhelming majority of such matchups and most of them are easy picks.

A measure you can look at that is independent of how close teams are to each other is the mean squared error. That's also in the report. A related measure is the absolute error. That's the absolute value of how far off the predicted margin of victory is from the actual margin of victory.

It seems reasonable to me to say that if you rank teams the test of how good you are at doing that is how frequently you are correct about what's going to happen when two teams play. The thing that got me into looking at this was contemplation of what the difference would be between rankings in a top 25 poll and rankings produced on the basis of who would be favored to win if teams played each other.

So I started looking to see how close various power rating systems are to the vegas line in order to identify proxies for ranking according to the vegas line. And certain systems compare pretty well with the vegas line. Sagarin is one of them. Overall the line is, I think, a little better. That's especially true when you look at mean squared error or absolute error. But the difference is not great.

The "small sample size" concept applies no matter what approach one takes to ranking college football teams. It applies whether it's human opinion or power ratings.

I don't know about lines in Vegas specifically, but 5dimes puts spreads on virtually every game involving a DI team (including against non-DIs).