## My Football Preview, Part 1: Methodology and the Big Ten

Jim Young-USA TODAY Sports

It is a bit hard to believe, but in a little less than a month, a new season of college football will have begun. Every serious fan or family of fans has their own special set of traditions when it comes to college football. Mine has taken a bit of a nerdy path over the past two decades.

You see, over the years I have developed an increasing complex set of formulas and algorithms which I use to analyze the college football landscape. While this "math" is perhaps most useful over the course of the season, many of the tools that I have developed are quite helpful when it comes to predicting how the upcoming football season might play out. Putting together these predictions is one of my new favorite late summer traditions.

I have written about my general methodology in the past, so here is the "short" version. Over the years I have developed a very simple algorithm to generate power rankings from the final scores of every single Division 1 football game in a given season. I can use these rankings to estimate point spreads (corrected for home/away) which generally correlate very well to the Vegas line. I have also developed a very good correlation between point spreads and the probability of victory. With these two pieces of information, I can easily calculate the odds that any give team will beat any other team given the location.

Once I have the set of probabilities, I can make all sorts of other calculations and simulations. I can estimate the total number of expected wins a given team will have given their schedule. This correlates well to the over/under for that team. With some additional mathematical framework (code for up to a 68 MB Excel file for each conference) I can also calculate the odds that any team will win their division, conference, and I can estimate the odds that they will make the college football playoffs. Finally, I also like to run a simulation on the entire season where I adjust the parameters to artificially inject a historically realistic (and mathematically consistent) rate of upsets, specifically in games where higher rated teams have to play on the road against slightly weaker opponents.

As the season progresses, there is enough data for the power rankings to converge and be self-consistent. However, in the first few weeks of the season, and especially for the preseason, I have to use some input data to set the initial strength of each team. Around the first of July, I scour the preseason magazines and websites, looking for as many sources that rank all 130 teams. I then use an average of all of these rankings to develop my own preseason consensus rankings.

All of my calculations rely on the assumption that the preseason rankings are... correct. Well, based on the data that I have collected, this is not actually true. My previous studies showed that the average deviation between any given team's preseason and postseason power ranking is 20 slots. Knowing this, I perform one more set of calculations that takes this uncertainty into account and generates an estimate of the odds that any team will finish the regular season with 'x' wins. This calculation winds up being very conservative (all teams expected win total gets pushed towards 50%) but it is also quite accurate.

In my analysis of each conference, I typically use all of these pieces of information to develop a picture of how the season might play out. In years past, I have dumped all of that information into one huge post. This year, I thought I would try to divide it up into sections to make it a bit more palatable. Plus, I have even more data than usual to share. So, to kick things off, let's take a close look at the Big Ten.

Big Ten East

I am going to start this section with a disclaimer. Based on the input data that I have, Michigan is predicted to do very well. That does not mean that I personally think that the Wolverines are going to win the Big Ten (more of that later), but that is what the math says this year and so that is how I am going to report it. Now, for the details.

The table below gives a good summary of the overall set of data that I can generate for each conference based on the preseason data. The table gives the projected odds for each team to win its Division and make it to Indy, the consensus preseason ranking for each team, the odds for each team to hoist the Big Ten trophy in Indy, the odds to make the playoffs (by winning in Indy or by finishing 11-1 but not making it to Indy, i.e. the Alabama plan), the number of games each team is likely to be favored in, and the expected value for win total, using both the raw preseason rankings and my more conservative correction applying the average level of preseason ranking uncertainty. There is A LOT to take in here, but in total, I think it gives a great insight into the season to come.

So, here is the deal with Michigan (5). It really should come as no surprise that they have the best odds, based on my calculations, to win the East, win in Indy, and make the playoffs. Why? Based on the preseason rankings, they are a consensus Top 5 team, which is tops in the Big Ten. Also, they have a favorable schedule, as they draw the only other preseason Top 10 Big Ten team, Ohio State (8), at home. So, based on projected point spreads, they have a good chance to be favored in all 12 games, as the table indicates.

That said, the Wolverine's schedule is not without potential pitfalls. Specially, they have to travel to Wisconsin (21) and Penn State (15). In my simulation of the season, UofM drops both of these games yet wins their slate of tough home games vs. Iowa (18), Notre Dame (9), MSU (16), and Ohio State (8) to finish at 7-2 overall. In principle, this could leave the door open for Ohio State, MSU, or Penn State to sneak in and steal the crown. However, all of those teams have tough road games as well, and my simulation still projects UofM to come out on top, tied with Ohio State at 7-2 and with the head-to-head tie-breaker in hand. But, it is close, as UofM's odds to win the East are at 42%, while OSU is sitting not far behind at 29%.

As for the Buckeyes, similar to Michigan, they have the advantage of drawing 2 of the other top 3 East teams at home this year (MSU and Penn State), and their road schedule is clearly easier than UofM's. So, there is a pretty clear path to Indy for the Buckeyes as long as they can protect their home turf, even if they lose to Michigan. The biggest potential road challenge for the Buckeyes will take place on the last weekend of September, when OSU travels to Lincoln to play the Cornhuskers (27). My simulation predicts a very narrow (less than a point) upset loss, which in this scenario is just enough to cost the Buckeyes a trip to Indy. The other road games (at Indiana (53), at Northwestern (42), and at Rutgers (101)) are less scary, but the Buckeyes also need to take care of business when Cincinnati (38) and Wisconsin (21) both come to town.

As for the battle for 3rd place, the current data suggests that it will be a neck-and-neck race between MSU (16) and Penn State (15), with MSU having the slightest edge right now. Based on the consensus ratings, both teams are even, and their schedule appear to be comparable, as the two schools have virtually identical odds (14.3%) to win the East.

A closer look at the schedule shows why MSU has the slight edge. Both teams travel to Ohio State, so that's a wash. PSU travels to Iowa (18) while MSU travels to Wisconsin (21). PSU travels to Minnesota (35) while MSU travels to Northwestern (42). If we believe the preseason rankings, MSU has the slightly easier path in both situations. Penn State does get the advantage of playing Michigan in Happy Valley, while MSU has to travel to Ann Arbor, but neither team is expected to beat UofM anyway. But, the biggest advantage for MSU is that they host Penn State this year.

If I consider all of these factors, Penn State is only projected to be favored in 5 Big Ten games, while MSU is favored in 6. In my simulation, MSU's W/L are same, while Penn State would be projected to beat the Wolverines at home, yet lose to the Gophers on the road. In either scenario, I project PSU to lose at OSU, Iowa, and at MSU.

Speaking of MSU, let's now take a deeper dive into the details of the Spartan's schedule. If the preseason rankings are correct, MSU is going to be facing a lot of bad teams, mostly at home. I project MSU to have over a 95% chance to win each of their 5 easiest games (Tulsa (98), at Rutgers (101), Illinois (91), WMU (72), Maryland (69)) and over a 90% chance to win their 6th hardest game (home vs. Indiana (53)). Only one of those games ins on the road and that is against Rutgers, who projects to be the worst team MSU will face all year. Based on this data, MSU has a 83% chance of winning all 6 of those game.

In any event, the next 2 hardest games on the schedule (vs. Arizona State (43) and at Northwestern (42)) pose a bigger challenge, obviously. But, MSU is currently projected to be a 14- and 7-point favorite. The math suggests a 60% chance the MSU wins both games. Combined with the previous results and MSU has a 50-50 shot to win all 8 of their easiest 8 games with an expected win total of 7.4. In other words, as long as MSU is as good as the experts think, wining all 8 of those games is fairly likely and 7 seems like a gimme.

All this does not even consider the toughest 4 games on the docket. MSU is currently projected to have anywhere from a 30% to 60% chance to win those toughest four contests: vs. Penn State (15), at Wisconsin (21), at Ohio State (8), and at Michigan (5). If these odds hold up, the math suggests that MSU would be expected to win between 1 and 2 of those games (1.6 is the expected value). More specifically, MSU would have an 89% chance of winning at least 1 of those games, a 54% of winning at least 2, and an 18% chance of winning 3 or more. The odds of going 0-fer on the four toughest games is only 11%. Take all of that together and MSU's expected win total of 9 seems spot on.

Essentially, MSU is likely to be favored in 9 games and is expected to lose the other 3 (the tough road games at Wisconsin, OSU, and Michigan). Based on the way the schedule sets up (easy games at home and tough games on the road), my simulation of the season does not change the results of any of MSU's games. For this reason, the potential variance in MSU's expected win total is lower than most teams. The ceiling appears lower, but the floor seems higher than last year.

It is probably obvious, but just to complete the picture in the Big Ten East, there is a huge gap between the Top 4 in the Division and the Bottom 3. None of those teams (Indiana, Maryland, and Rutgers) is projected to have more than a 0.1% shot to win the Division. Actually, I don't project any of those teams to even make a bowl.

Big Ten West

The battle for the Big Ten West crown looks to be surprisingly wide open, with the top 4 teams (Iowa (18), Nebraska (27), Wisconsin (21), and Minnesota (35)) all with between a 15% and 32% chance to advance to Indy. As the table above shows, the Top 3 teams are all expected to be favored in 9 games total, while Minnesota is projected to by favored in 8.

While the Huskers are the trendy pick, Iowa is actually the highest ranked West team in the preseason and my analysis suggests that Iowa has the slight edge (32% to 28%). Based on the projected spreads, Iowa should be favored in 7 conference games and 2 of their 3 non-conference games (the game at Iowa State (26) projects the Hawkeyes to be a slight underdog.)

That said, looking at the schedules in detail provides some hints as to why the Huskers, despite being the 3rd highest ranked team in the Division, still might be the favorite. In fact, my simulation picks Nebraska to finish 8-1 and take the Division by 2 games. The Huskers road slate is very forgiving, as they play at Illinois (91), at Purdue (51), and at Maryland (69). Their toughest road game is at Minnesota (35). Somehow, they draw Iowa (18), Wisconsin (21), and Northwestern (42) all at home. Sure, they also host Ohio State (8), but even if they drop that game and the one at Minnesota, they simply have the manageable task of winning the rest of their home games and they can likely book their hotels in Indy.

As for Iowa, they seem to have the opposite schedule as Nebraska. As mentioned above, they travel to Lincoln, but also to Madison, Evanston, and Ann Arbor. Oh, and they get to host Penn State. Yikes!

As for Wisconsin, their Big Ten West schedule is not as bad as Iowa's, but it's close. The Badgers have to travel to Nebraska (27) and Minnesota (35) which is not great, but the cherry on top is that their three East cross-overs are Michigan (5), Michigan State (16), and at Ohio State (8). Good luck with that.

While it is a bit harder to imagine the Gophers (35) making their first trip to Indy, they might just be strong enough, and their schedule might be just manageable enough to allow them to squeak through. While they do have to travel to Iowa (18), their only other road games are at Purdue (51), at Rutgers (101), and at Northwestern (42). They also got a break with the East cross-over games. In addition to the road game at Rutgers, they play host to Maryland (69) and Penn State (15). While my simulation pegs them at 6-3 in the conference play, one well-timed upset could put them right in the thick of the race.

As for the rest of the West, Northwestern (42), Purdue (51), and Illinois (91) are all projected to be outside of the Top 40 and have less than a 4% chance to win the West. The Wildcats and the Boilers are projected to struggle to even make a Bowl Game.

Other Details

As a supplement to the summary table above, I can expand upon the simple expected win totals shown in the final two columns. My calculations can also project the probabilities that each team will win any give number of games. I summarize this information in a "win matrix" and at the beginning of the season I have two of them: one based on the raw preseason rankings, and my more conservative corrected version that takes into account the uncertainty of those rankings. Below, I show the corrected matrix for the Big Ten.

Note that the individual probabilities are not overly useful. For example, the matrix says MSU has a 21.3% chance to win 9 games. What does that even mean? I find it more useful to look at blocks of data instead. For example, if I sum the cells, MSU has a ~62% chance to win between 7 and 9 games. That seems reasonable. Also, MSU has a 20% chance to win 10 games or more in the regular season. While that might seem low, note that Michigan's odds for 10 wins or more is only 28%.

This brings up another point, which is that even though these numbers seem low, my retrospective analysis of the past 2-3 season shows that they are pretty close to correct. What I mean by that is if a team has a 20% chance to do something, at the end of the year roughly 20% of all Division 1 teams (26 teams) will hit that mark. Also, the standard deviations shown in the final column are also dead on. Basically, ~68% of all teams will have win totals that fall within one standard deviation of the list mean and 95% of all teams will have totals within 2 deviations.

As you can see from the table, the typical standard deviation is about 2 wins for this analysis, and that is actually a huge number. The bottom line is that college football is far less predictable than we think. The hidden message in the data is that 8 of the 14 Big Ten teams have expected win total that is between roughly 7.5 and 8.5. The Big Ten race is actually much more open than it may appear.

As a final set of data, I wanted to show a quick comparison of my calculated expected wins compared to the Vegas over/under numbers that I pulled off of the CBS Sports website. In this case, I am using the uncorrected expected win values. As you can see, my projections agree pretty well with the Vegas lines.

If I ever wanted to attempt to use my analysis to inform a betting strategy (which is a TERRIBLE idea, BTW) it looks like the "under" bet for Purdue is the best bet, and Maryland and Ohio State also both under the curve. As for teams where the "over" might make sense, MSU, Iowa, and Minnesota are the most likely, based on my analysis.

When making my "official" predictions in the preseason, it is my tradition to use the results of my simulation as my final picks. In this case, that mean a pick of Michigan (10-2) to win the East and Nebraska (11-1) to win the West. In this scenario, Michigan would be favored to beat the Huskers and take their first Big Ten Title since 2004, albeit with 2 losses, which most likely would mean simply a Rose Bowl trip and not a Playoff appearance. I think I just threw up in my mouth a little bit.

But, before any Wolverine fans out there start to look for hotels in Indianapolis or Pasadena, let me rain on the parade a bit. First of all, this entire analysis above is simply a projection, and while I think my methods are sound, as I stated above, college football predictions are in general not very reliable. Case in point: in this very space I predicted MSU to make the playoffs in both 2016 and 2018. Both predictions were completely defensible based on the data. Whoops.

So, let's think about about the reality of the situation. For Michigan to win the Big Ten, several things need to happen. First, they have to actually be good. The preseason rankings all have them in the Top 5, but so was Florida State two years ago, and they wound up finishing 6-6. The Wolverines have no running back and a lost several starters on the defense. But, hey, "Leaders and Best," and "winged helmets!" They must be good again sometime... right? After all, MSU had a down year last year, players seem to be bailing from James Franklin's program, and Urban Meyer left OSU. It seems like the prognosticators are mostly picking Michigan by default more than anything else. Personally, I will believe it when I see it.

Second, even if the Wolverines are pretty good, they still have to actually win big games. By my count, six of their twelve opponents are ranked in the Top 21 in the preseason. Granted, four of those six games are at home, but still... Michigan's struggles against ranked teams over the past decade (and longer) are well documented. Why, again, do some assume that this problem will suddenly get fixed? As for the home games, two of those are against MSU and Ohio State. Michigan has demonstrated time and time again for a very long time that they cannot beat a decent MSU or OSU team no matter where that game is played. So, once again, why is that going to change this year? Again, I will believe it when I see it. The Wolverines lost the benefit of the doubt a very long time ago.

Finally, I wanted to point out again the clustering of adjusted expected wins in my table above. As stated above, this implies that there actually is a fair amount of parity in the Big Ten this year. So, what happens if Michigan gets the injury bug or simply is not quite as good as projected? Right now, if they truly are a Top 5 team, they are projected to be favored to win all 12 games. But, what if they are in reality only a Top 25 team? If I rerun the numbers with Michigan as merely the 25th best team, they suddenly are only likely favored in seven games, and in one additional game (vs. Iowa) they project as less than a 1-point favorite. AND, that doesn't even consider the potentially tricky game against Army (60) in Week 2, which many experts are looking at as a potential upset.

Yes, the numbers look good for Michigan right now, but there is very much a nightmare scenario out there for them, and it is not that far-fetched. My math suggests only a 12% chance Michigan finishes at 6-6 or worse, which is only slightly less likely than either MSU or Penn State winning the East. Sleep well Wolverine fans...

That will do it for my breakdown of the Big Ten. Next up I will dive into the rest of the Power 5, with not quite the detail of this post. I will also touch on the Group of Five and make my final NY6 / Playoff Bowl projections. Stay tuned, enjoy, and Go Green!

This is a FanPost, written by a member of the TOC community. It does not represent the official positions of The Only Colors, Inc.--largely because we have no official positions.