One new piece of analysis I'm doing this year is looking at some of key matchups and what Compu-Picks thinks the results of these matchups will say about the rest of the season.
With three weeks of results in the books, we've learned a fair amount about a number of teams. Given what we've learned, how should we adjust our projections? This analysis attempts to do that, adjusting the original preseason projections given win/loss information (as in previous articles, ONLY wins and losses are considered; in-game information such as margin or other statistics are not considered).

First up is Clemson at Florida St. I start with the preseason spread (as can be seen here) and current spread for comparison purposes, then look at the original Compu-Picks preseason projection of Florida St's probability of winning this game. I then show the adjusted probability given each successive piece of information. As you can see, Clemson's win over Auburn made an impact, though a number of other data points (most notably Auburn's subsequent loss to Miss St) dropped the impact a bit.

Overall, this analysis gives Clemson a bit better than a 1 in 4 shot at this game, which is more like what you'd expect from a 10 point line rather than a 14 point line. Essentially, the Compu-Picks estimate would have originally leaned towards Florida St (though not by enough to hit even a level 1 confidence point), but while the Vegas line has shifted towards FSU, the Compu-Picks analysis has moved a bit towards Clemson, overall changing the lean. That said, it's not a particularly meaningful lean, especially given the relatively looseness of this analysis.

One thing to note here is that I've added a column to the right showing the number of simulations left after taking each additional W/L result into account (i.e. throwing out simulations where Auburn beat Clemson leaves us with 1,980; throwing out the simulations where FSU lost their AA opponent leaves us with 1,978 [i.e. that was an extremely unlikely result], etc.). As the number of simulations decreases, it's worth taking the resultant win probabilities with some degree of skepticism. In this case, the numbers stay pretty decently high, but in some later tables, this isn't the case.

Preseason Line for Clemson at Florida St. | 10.5 |

Current Line for Clemson at Florida St. | 14.0 | Simulations Left |

Original Compu-Picks Probability of Clemson win over Florida St. | 22.38% | 5001 |

Probability Given Clemson Win over Auburn | 30.15% | 1980 |

Probability Given Florida St. Win over AA | 30.08% | 1978 |

Probability Given Clemson Win over Ball St. | 30.29% | 1948 |

Probability Given Florida St. Win over AA | 30.29% | 1948 |

Probability Given Clemson Win over AA | 30.30% | 1947 |

Probability Given Florida St. Win over Wake Forest | 29.64% | 1903 |

Probability Given Wake Forest Win over North Carolina | 28.32% | 1317 |

Probability Given Mississippi St. Win over Auburn | 27.13% | 774 |

Next up is the other key game of the weekend, Michigan at Notre Dame. The same types of calculations occur here as we saw for Clemson - Florida St. Here the system largely agrees with the line movement; it makes sense, given what we've seen on the field, that it would shift towards Notre Dame. However, in this case the system made Michigan a meaningful favorite over Notre Dame preseason, and while the estimate has certainly moved, it wasn't by enough to justify the current line. Then again, this simple W/L analysis doesn't factor in the blowout loss by Michigan to Bama, the dominant win by Notre Dame over Michigan St, or Michigan's struggles against Air Force (then again, it also doesn't factor in Michigan's struggles against Purdue).

The system's lean here is overall stronger than Clemson-FSU, but it's hardly overwhelming, especially when you note the data that simple W/L necessarily overlooks.

Preseason Line for Michigan at Notre Dame | 2.5 |

Current Line for Michigan at Notre Dame | 7.0 | Simulations Left |

Original Compu-Picks Probability of Michigan win over Notre Dame | 64.33% | 5001 |

Probability Given Michigan Loss to Alabama | 60.54% | 3609 |

Probability Given Notre Dame Win over Navy | 58.31% | 3121 |

Probability Given Michigan Win over Air Force | 59.83% | 2940 |

Probability Given Notre Dame Win over Purdue | 57.49% | 2369 |

Probability Given Michigan Win over Massachusetts | 57.57% | 2364 |

Probability Given Notre Dame Win over Michigan St. | 51.61% | 1242 |

Probability Given Alabama Win over Arkansas | 52.55% | 1058 |

Probability Given Michigan St. Win over Boise St. | 53.85% | 572 |

Thursday night's game between BYU and Boise could be interesting, though the books don't seem to think so, with a 7.5 point line (and apparently rising). Have the books seen something Compu-Picks missed?

It's certainly possible, though if so it's fairly hard to see it. Boise's loss to Michigan St was close, but Michigan St's loss to Notre Dame wasn't. And meanwhile, BYU has been dominant in their wins and very close in their loss (and Utah's loss was extremely close). If anything, digging into the box scores suggests that the probability movement may be a bit overstated.

This is especially true given that there are only 33 simulations left after removing the non-compliant simulations. That's a pretty small number to think that 42% is necessarily accurate, especially since the final move (49% to 42%) was in the opposite direction of what you'd expect given Michigan St losing to the Irish (normally, you'd expect that to increase, not decrease BYU's chances).

So overall, this is a strong pick. This game looks like a tossup, but the line is over a touchdown. In other words, we're basically looking at last Thursday's Rutgers-USF game all over again (except here the line is 7.5 instead of 10), and we all remember how that ended...

Preseason Line for BYU at Boise St. | n/a |

Current Line for BYU at Boise St. | 7.5 | Simulations Left |

Original Compu-Picks Probability of BYU win over Boise St. | 57.19% | 5001 |

Probability Given BYU Win over Washington St. | 58.88% | 4662 |

Probability Given Boise St. Loss to Michigan St. | 65.26% | 2838 |

Probability Given BYU Win over AA | 65.33% | 2835 |

Probability Given Boise St. Win over Miami (Ohio) | 61.83% | 2261 |

Probability Given BYU Loss to Utah | 53.62% | 1214 |

Probability Given Utah St. Win over Utah | 49.37% | 79 |

Probability Given Notre Dame Win over Michigan St. | 42.42% | 33 |

Another key game of this weekend is Kansas St at Oklahoma. Here, Compu-Picks called it a near-lock of a Sooner win preseason, and while results so far push it a bit towards Kansas St (as does the line movement), neither one was a big move. Compu-Picks liked the Sooners on the preseason line, and hasn't seen enough data to change its mind. Factoring in KSU's dominance over Miami might have shifted it a bit, but this still looks like value on the home team (though as noted in the season preview, the system may simply be underestimating KSU, as Bill Snyder is no stranger at all to over-achieving expectations).

Preseason Line for Kansas St. at Oklahoma | 16.5 |

Current Line for Kansas St. at Oklahoma | 14.0 | Simulations Left |

Original Compu-Picks Probability of Kansas St. win over Oklahoma | 5.46% | 5001 |

Probability Given Kansas St. Win over AA | 5.51% | 4958 |

Probability Given Oklahoma Win over UTEP | 5.27% | 4875 |

Probability Given Kansas St. Win over Miami (Fla.) | 6.11% | 3766 |

Probability Given Oklahoma Win over AA | 6.11% | 3765 |

Probability Given Kansas St. Win over North Texas | 6.24% | 3575 |

Probability Given Miami (Fla.) Win over Boston College | 6.94% | 720 |

Probability Given Mississippi Win over UTEP | 7.28% | 563 |

Compu-Picks isn't surprised at all by the extreme line in Arizona-Oregon, despite the Wildcats' surprsingly strong start. The system rated Oregon as a juggernaut, and while there have been injury issues, it's hard to see the Ducks' performances to date as undermining that assessment in any meaningful way (other than that their second string isn't all that great). And given Arizona's struggles against Toledo (again, simple W/L doesn't reflect that level of detail), it's hard to really see this assessment as being wrong. This looks like a blowout waiting to happen.

Preseason Line for Arizona at Oregon | n/a |

Current Line for Arizona at Oregon | 24.0 | Simulations Left |

Original Compu-Picks Probability of Arizona win over Oregon | 0.40% | 5001 |

Probability Given Arizona Win over Toledo | 0.66% | 2428 |

Probability Given Oregon Win over Arkansas St. | 0.66% | 2411 |

Probability Given Arizona Win over Oklahoma St. | 1.80% | 388 |

Probability Given Oregon Win over Fresno St. | 1.81% | 386 |

Probability Given Arizona Win over AA | 1.86% | 376 |

Probability Given Oregon Win over AA | 1.86% | 376 |

Probability Given Toledo Win over Wyoming | 1.31% | 153 |

Probability Given Fresno St. Win over Colorado | 1.72% | 116 |

Staying in the Pac-12, Oregon St at UCLA is another game that the system assesses as being too low of a line. The opening line of 11 seemed in the ballpark of right, but it's now down to 8 points, which seems like value on UCLA, and that's before factoring in Wisconsin's near-loss to Utah St and struggles against their AA opponent, both of which take a definite bite out of Oregon St's only win to date.

Preseason Line for Oregon St. at UCLA | n/a |

Current Line for Oregon St. at UCLA | 8.0 | Simulations Left |

Original Compu-Picks Probability of Oregon St. win over UCLA | 30.09% | 5001 |

Probability Given UCLA Win over Rice | 26.96% | 4258 |

Probability Given Oregon St. Win over Wisconsin | 34.83% | 2337 |

Probability Given UCLA Win over Nebraska | 28.46% | 1279 |

Probability Given UCLA Win over Houston | 27.11% | 1195 |

Probability Given Wisconsin Win over Utah St. | 28.10% | 1007 |

Probability Given Nebraska Win over Southern Miss | 26.78% | 844 |

Another line movement that Compu-Picks disagrees with is Rutgers-Arkansas. It opened at pick, and has now moved a full touchdown towards the Hogs. Even if you think that Arkansas QB Tyler Wilson will play (and he's still questionable from what I've seen), it's hard to lay any points, much less a full touchdown, on this absolute train wreck of an Arkansas team. Especially against a Rutgers team that's been playing very well so far. This may well be the best value on the whole board this week.

Preseason Line for Rutgers at Arkansas | n/a |

Current Line for Rutgers at Arkansas | 7.0 | Simulations Left |

Original Compu-Picks Probability of Rutgers win over Arkansas | 33.55% | 5001 |

Probability Given Rutgers Win over Tulane | 35.80% | 4385 |

Probability Given Arkansas Win over AA | 35.68% | 4370 |

Probability Given Rutgers Win over AA | 35.80% | 4344 |

Probability Given Arkansas Loss to La.-Monroe | 57.47% | 308 |

Probability Given Rutgers Win over South Florida | 63.33% | 150 |

Probability Given Arkansas Loss to Alabama | 64.49% | 138 |

Probability Given South Florida Win over Nevada | 64.36% | 101 |

Probability Given Alabama Win over Michigan | 64.56% | 79 |

Well, except for ECU-UNC anyway. Here, Compu-Picks started out thinking it was a tossup, and results to date have only swung the assessment towards ECU. It's true that UNC's losses were close, but that were getting annihilated at Louisville before the late comeback. It's just as likely that the Cardinals simply took their foot off the gas pedal and had trouble turning it back on as it is that UNC is going to turn a solid second half effort into a run going forward. I would say that Compu-Picks is overstating ECU's chances here a bit, but honestly this seems like a tossup to me.

Preseason Line for East Carolina at North Carolina | n/a |

Current Line for East Carolina at North Carolina | 17.0 | Simulations Left |

Original Compu-Picks Probability of East Carolina win over North Carolina | 47.05% | 5001 |

Probability Given East Carolina Win over AA | 49.17% | 4562 |

Probability Given North Carolina Win over AA | 46.49% | 4070 |

Probability Given East Carolina Loss to South Carolina | 44.85% | 3799 |

Probability Given North Carolina Loss to Wake Forest | 50.52% | 2490 |

Probability Given East Carolina Win over Southern Miss | 62.30% | 573 |

Probability Given North Carolina Loss to Louisville | 62.88% | 528 |

Probability Given Nebraska Win over Southern Miss | 63.16% | 494 |

Probability Given Louisville Win over Kentucky | 63.48% | 408 |

This is one of those "maybe, MAYBE there's a chance" games. Compu-Picks called it preseason for Auburn and has (obviously) switched its predicted winner, but it still gives the home team a shot. This is partially because it doesn't get to realize how badly Auburn lost at Miss St or how close they came to dropping ULM, and partially because it doesn't see the dominance of LSU's wins. All that said, Auburn is still a very talented team, and this is LSU's first road game, and probably against the best team they've seen so far. There's a reasonable chance Auburn gets totally blown out, but there's also at least a minor shot at an upset here. It's probably worth noting that I'm probably going to stay pretty skeptical of LSU until and unless they get out the Swamp still undefeated. I could certainly be wrong... but for now I'm not sold yet.

Preseason Line for LSU at Auburn | 9.5 |

Current Line for LSU at Auburn | 20.5 | Simulations Left |

Original Compu-Picks Probability of LSU win over Auburn | 43.71% | 5001 |

Probability Given LSU Win over North Texas | 44.04% | 4957 |

Probability Given Auburn Loss to Clemson | 53.74% | 1963 |

Probability Given LSU Win over Washington | 56.34% | 1633 |

Probability Given Auburn Loss to Mississippi St. | 63.97% | 952 |

Probability Given LSU Win over Idaho | 64.04% | 951 |

Probability Given Auburn Win over La.-Monroe | 62.82% | 893 |

Probability Given Washington Win over San Diego St. | 64.05% | 776 |

Probability Given La.-Monroe Win over Arkansas | 60.42% | 48 |

There are a few important notes and caveats I need to make about this model:

1) Compu-Picks does not endorse implicitly or explicitly any form of illegal gambling. Compu-Picks is intended to be used for entertainment purposes only.

2) No guarantee or warranty is offered or implied by Compu-Picks for any information provided and/or predictions made.

3) This preseason model is primarily based on the main compu-picks model. Essentially, it attempts to predict how well a team will rate given its rating history, as well as a number of other data points, such as returning starters, draft talent lost, turnovers, recruiting, etc. This means, among other things, that the rankings are power rankings based on how good a team projects to be, as opposed to a more cynical (though accurate) model that attempts to project how the BCS will rank a team by making adjustments to favor those with easy schedules and punish those with tough schedules.

2012 Compu-Picks Blog

Questions, comments or suggestions? Email me at cfn_ms@hotmail.com