Predicting Michigan’s win total in Jim Harbaugh’s first year

harbaugh_streetsHow many games will Michigan win in 2015? Can they rise to the level of Ohio State and become relevant on the national stage?

This seemed like a fantasy for Michigan until the hiring of Jim Harbaugh, one of the best coaches at both the college and pro level. For a suffering Michigan fan base, it felt like getting a new Porsche for a 17th birthday.

The conventional wisdom says that Harbaugh will bring Michigan back to national prominence. And this is a good thing. Love or hate the Michigan program, college football becomes more interesting with the increased relevaance of brand programs.

But will Michigan win many games this year? Here, we’ll use analytics to get a baseline expectation in 2015.

First, let’s take a closer look at the coach himself.

Stunning success as a coach

To understand the excitement in Ann Arbor over the hiring of Harbaugh, let’s look at what happened during his time at Stanford.

This visual shows a 30 year history of Stanford football. The rating in the bottom panel is an expected margin of victory against an average FBS team according to my computer calculations of team rankings.


Harbaugh took a team that went 1-11 the year before he arrived and turned Stanford into an elite college football team. He brought a culture of toughness and physicality to the program. It also helped that he recruited QB Andrew Luck, a future number one overall NFL draft pick.

The Harbaugh culture lives on at Stanford even years after his departure for the San Francisco 49ers. The program continues to win despite the questionable decisions made by current coach David Shaw.

Why is Harbaugh a good coach? Many articles have been written about his complete devotion to coaching football. ESPN’s Seth Wickersham wrote that “everything that isn’t coaching football is a distraction” to Harbaugh, which makes him seem rude at times.

However, there’s more to Harbaugh’s greatness. He also understands people. Consider this story about his wife Sarah from a recent Sports Illustrated profile.

People often ask Sarah what it’s like when Jim yells at her, and she tells them, “He’s never yelled at me.” She yells sometimes, but he doesn’t yell back. He just tells her the real reason she is mad, and he is usually right, and yes, that can be maddening. Sometimes it would be easier if he just yelled.

Great coaches can motivate their players. Duke coach Mike Krzyzewski doesn’t create innovative basketball strategies. Instead, he gets people and what makes them tick. Same for Phil Jackson.

Can Harbaugh make a difference in 2015? Let’s estimate the talent he has to work with.

Michigan in The Power Rank’s preseason rankings

In 2014, Michigan went 5-7, which led to the firing of coach Brady Hoke. While the firing seemed justified, I don’t know how many coaches survive a season with a -16 turnover margin (takeaways minus giveaways) and a senior QB Devin Gardner whose performance regressed.

While my preseason model doesn’t account for Gardner, it does consider the turnover margin. There’s a lot of randomness in turnovers, which implies that teams with large negative (or positive) values in this statistic are unlikely to have such extreme values the next season.

There is no guarantee in this regression to the mean. In the three years that Rich Rodriguez coached Michigan, the team had a turnover margin of -10 or worse. But by the numbers, Michigan is unlikely to perform worse than -16 in turnover margin this season no matter who coaches them.

In addition to turnovers, the preseason model also considers a program’s performance over the past four seasons and returning starters. Despite the simplicity of the model, which does not consider coaching changes, it has predict the winner in 70.4% of games since 2005. In addition, the higher ranked team has won 60.2% of bowl games from 2005 through 2014.

Michigan is 34th in my 2015 preseason rankings, almost exactly what SB Nation’s Bill Connelly expects.

They’re not in the same class as Ohio State (2nd), or even Wisconsin (14th) and Michigan State (16th) in the Big Ten. However, Michigan sits between Nebraska (32nd) and Penn State (38th).

Michigan is better than many of the other Big Ten opponents on their schedule. This will impact their expected win total.

For the full preseason rankings, click here.

Want more preseason college football analysis? Keep up to date with my content through the free email newsletter. Enter your best email and click on “Sign up now!”

The most important player in 2015

Heading into the 2015 season, Michigan fans are stressing about quarterback. Devin Garnder struggled last year, and the offense ranked 71st in yards per play adjusted for strength of schedule, The Power Rank’s primary metric for evaluating offense and defense.

With the graduation of Garnder, rising sophomore Shane Morris looked like the best quarterback in spring practice. However, he showed the same inaccuracy in the spring game (11-24) that has plagued his career. While the 135 passing yards looks good, he got most of those throwing against a running back attempting to play cornerback (Dennis Norfleet).

I agree with Brian Cook of that Morris will almost certainly lose the QB job to Jake Rudock, the graduate transfer from Iowa. Rudock started two years and completed 60.3% of his passes, almost the exact same percentage as Gardner. However, Rudock will most likely be better reducing turnovers, as he threw interceptions on 2.6% of his passes, much less than the 4.1% of Garnder.

It’s becoming cliche to say that Rudock raises the floor for Michigan in 2015. However, the most important player should be high ceiling talent that allows the team to exceed expectations. For Michigan, that’s Jabrill Peppers.

Peppers was the top ranked cornerback recruit in 2014. As the third ranked recruit overall that season, recruiting guru Sam Webb thought he was vastly underrated. He played a few games last season before getting injured and will move to safety this season.

Peppers has the potential to take a Michigan defense ranked 17th by adjusted yards per play last season up into the top 10 or 5. Throw in some lucky bounces that turn into takeaways, and the Michigan defense could carry this team above expectations.

Breaking down the schedule

To determine how games Michigan will win, let’s break the schedule into four tiers.

  • Almost certain wins: UNLV (97.3%).
  • Very likely wins: Oregon State (82.2%), BYU (67.2%), at Maryland (75.2%), Northwestern (77.7%), Rutgers (83.3%), at Indiana (74.0%).
  • Toss up games: at Utah (40.4%), Michigan State (45.5%), at Minnesota (53.9%), at Penn State (46.1%)
  • Unlikely wins: Ohio State (23.5%)

Unless an earthquake swallows the Big House, Michigan will beat UNLV. They have greater than 66% chance in another 6 games, which results in 4 or 5 more wins. If they can win two of four toss up games, that’s 7 or 8 wins for the season.

Adding the win probabilities in Michigan’s 12 games gives an expected win total of 7.7, very close to the market total of 7.5. Simulations of the season give Michigan a 56% chance to win 8 or more games.


The expected 7.7 wins by my preseason model makes no adjustments for the Harbaugh factor. If he can establish his brand of physical football this season, Michigan can exceed these preseason expectations.

Michigan should also vastly improve on special teams. Harbaugh hired special teams coach John Baxter, who had a remarkable track record at Fresno State and USC. With a timely blocked kick, Michigan steal an extra game.

Want more Michigan content from a numbers point of view? I’m part of the MGoBlog Roundtable with Sam Webb, Brian Cook and Craig Ross on WTKA. The radio starts up again in August (Thursday mornings at 9am).

How to find a late round QB with value for fantasy football

Screen Shot 2015-07-15 at 4.18.14 PMFantasy football is so random.

Your buddy Jones picks up some rookie running back late in the draft and runs off with the league championship. Who can predict such things?

Even with the difficulty of predicting rookie performance, that doesn’t stop the quants from trying. Joseph Juan over at Numberfire makes the argument for Tennessee rookie QB Marcus Mariota (assuming he signs a contract).

The quantitative argument is flimsy, as it relies on a comparison with one other player (Cam Newton). But the article is worth a read if you’re in a fantasy football league.

To read the article over on Numberfire, click here.

Is the Home Run Derby slump real?


Over at the Detroit News, I look into whether hitters who participate in the Home Run Derby go into a slump afterwards.

The idea seemed ridiculous to me. However, I was surprised by the truth.

The Home Run Derby slump is a textbook example of regression to the mean. In writing this article, I think I developed a pretty good way to think about this crucial statistical concept.

In addition, this Home Run Derby curse has an impact on players who won’t participate.

To read the article, click here.

Finally!! College football preseason rankings for 2015

American Football on the Field near the hashmarks or yard lines

Which college football teams will excel in 2015 and go over their projected win total? Which teams will fail to live up to the hype and fall short?

In college football, it’s much easier to make this projection than in the NFL. Team performance tends to persist from year to year.

For example, Alabama has consistently been in the national championship picture for the last 6 years. Even with the worst luck, Nick Saban’s team usually ends the season in the top 10 of The Power Rank’s team rankings.

In contrast, Eastern Michigan has none of the resources of a program like Alabama and perform accordingly. Escaping the bottom 25 of FBS schools in computer rankings constitutes a monumental achievement for the Eagles.

Preseason regression model for college football

My preseason rankings come from a regression model that uses The Power Rank’s team rankings from the past 4 years, turnovers and returning starters. For more details, see the bottom of this post.

The college football preseason model is simple. It doesn’t break down a team into offense and defense or distinguish a starting quarterback from a linebacker. These are areas for potential improvements.

However, the model performs very well in predicting the winners of games. The model assigns each team a rating, and the difference in the rating of two teams gives a predicted margin of victory on a neutral field. The home team gets an extra 3 points.

For example, Ohio States opens the season at Virginia Tech. The Buckeyes have a 19.7 rating, which gives a predicted margin of victory against an average team. The Hokies have a 11.0 rating.

On a neutral field, Ohio State would beat Virginia Tech by 8.7 points. The home advantage for Virginia Tech brings this advantage down to 5.7 points. The model still predicts an Ohio State victory.

In 2014, this model predicted the winners in 70.4% of games. With a large sample of 678 games, the model performed very close to the 70.5% win rate from the 2005 through 2013 seasons.

If we assume no home field in bowl games, then this model predicts the higher ranked to win. This prediction has won in 59.9% of bowl games from 2005 through 2014.

For comparison, the team favored by the opening line of the markets won 59.5% of bowl games from 2005 through 2013. The markets have the benefit of evaluating an entire regular season of games.

The top preseason college football teams of 2015

The visual shows the top 15 college football teams for 2014.

Screen Shot 2015-07-10 at 12.29.32 PM

For a full list, click here or check out the list at the bottom of this post.

Let’s look at three things that jump out at me.

Ohio State

Ohio State had a preseason rank of 13th before the 2014 season.

While that might seem low, Ohio State’s performance on the field didn’t justify a higher rank. Even with a 24-2 record over the 2012 and 2013 seasons, the Buckeyes finished 15th and 12th in my rankings.

However, Ohio State surged at the end of the 2014. They obliterated Wisconsin in the Big Ten championship before winning two more games to capture the first college football playoff championship.

This season, the Buckeyes are 2nd in my preseason rankings. If the model considered the quarterback depth with 3 returning starters, they would be above Alabama for first.

The end of SEC dominance?

Alabama, Auburn, Ole Miss, Mississippi State and LSU all lost their bowl game last season. College football fans outside the south rejoiced, claiming the end of the SEC’s rule over college football.

However, my preseason numbers disagree. Let’s look at the rank of SEC teams in my preseason rankings over the past two years.

  • 2014: 2, 5, 6, 8, 9.
  • 2015: 1, 6, 8, 9, 10.

With cross country scoring, SEC teams had a total rank of 34 for its top 5 teams in 2015. Since a lower score is better, this is only slightly worse than their 30 score from last year.

The SEC continues to have traditional powers with ample financial resources and fertile recruiting territory in their back yards. They are in no danger of becoming the ACC anytime soon.

Don’t let the small sample size of 5 bowl games losses for the SEC fool you. For more on this crucial aspect of understanding sports, check out this article.

Notre Dame

Brian Kelly started coaching at Notre Dame in 2010 and has achieved these final rankings in The Power Rank’s college football rankings.

  • 2010: 32nd.
  • 2011: 21st.
  • 2012: 11th, went undefeated before losing to Alabama in the BCS title game.
  • 2013: 23rd.
  • 2014: 39th, started 6-0 before losing to Florida State and ending 8-5.

It’s not the resume of an elite team.

Moreover, Notre Dame recruits well. They finished 7th in the 2015 rankings of college football teams by recruiting. The roster is always littered with 4 and 5 star recruits.

Could 2015 be the year that all the talent meshes into a winning team? Notre Dame returns every starter besides QB Everett Golson, who graduated and transfered to Florida State, and two offensive lineman. This propels them up to 12th in my preseason rankings.

Notre Dame should be favored in every game except their trip to Stanford. If they can’t get 9 wins with this kind of talent and experience, it’s not unreasonable to ask Brian Kelly or anyone else can consistently win at Notre Dame.

More college football previews to come

Over the next two months, I’ll be adding more college football preview content based on these analytics. This includes team write ups on win totals as well as a preview of Michigan at the dawn of the Jim Harbaugh era.

To get updates on this content, sign up for my free email newsletter. As a thank you, I’ll send you a report with the top 5 football analytics articles.

Enter your email and click on “Sign up now!”.

Full preseason college football rankings for 2015

1. Alabama, 20.91.
2. Ohio State, 19.70.
3. Oregon, 19.65.
4. Baylor, 18.02.
5. TCU, 17.22.
6. Georgia, 14.84.
7. UCLA, 14.60.
8. LSU, 13.97.
9. Texas A&M, 13.87.
10. Mississippi, 13.36.
11. Oklahoma, 13.24.
12. Notre Dame, 13.14.
13. Stanford, 13.14.
14. Wisconsin, 12.72.
15. Arkansas, 12.14.
16. Michigan State, 11.97.
17. Florida State, 11.72.
18. Georgia Tech, 11.42.
19. USC, 11.41.
20. Arizona State, 11.26.
21. West Virginia, 11.12.
22. Virginia Tech, 11.00.
23. Kansas State, 10.53.
24. Tennessee, 10.49.
25. Auburn, 9.80.
26. Boise State, 9.73.
27. Missouri, 9.52.
28. Oklahoma State, 9.14.
29. South Carolina, 8.88.
30. Clemson, 8.84.
31. Mississippi State, 8.73.
32. Nebraska, 7.94.
33. Florida, 7.66.
34. Michigan, 7.60.
35. Utah, 7.58.
36. Arizona, 6.99.
37. Pittsburgh, 5.89.
38. Penn State, 5.79.
39. North Carolina, 5.68.
40. Texas, 5.35.
41. California, 5.10.
42. Brigham Young, 5.08.
43. Miami (FL), 4.84.
44. Texas Tech, 4.83.
45. Louisville, 3.99.
46. Utah State, 3.96.
47. Minnesota, 3.38.
48. Duke, 3.37.
49. Iowa, 3.19.
50. Marshall, 2.75.
51. Washington State, 2.59.
52. Washington, 1.73.
53. Vanderbilt, 1.57.
54. Cincinnati, 1.47.
55. Northwestern, 0.99.
56. East Carolina, 0.62.
57. Temple, 0.44.
58. Western Kentucky, -0.14.
59. Louisiana Tech, -0.18.
60. Boston College, -0.30.
61. North Carolina State, -0.35.
62. Virginia, -0.37.
63. Houston, -0.74.
64. Kentucky, -0.98.
65. UCF, -1.01.
66. Oregon State, -1.19.
67. San Diego State, -1.27.
68. Colorado State, -1.41.
69. Iowa State, -1.69.
70. Rutgers, -1.80.
71. Purdue, -1.99.
72. Navy, -2.28.
73. Toledo, -2.33.
74. Colorado, -2.61.
75. Illinois, -2.65.
76. Northern Illinois, -2.66.
77. Western Michigan, -2.69.
78. Nevada, -3.08.
79. Air Force, -3.40.
80. Indiana, -3.46.
81. Arkansas State, -3.54.
82. Memphis, -3.57.
83. Maryland, -3.94.
84. Wake Forest, -4.00.
85. Fresno State, -4.16.
86. Ball State, -4.44.
87. Syracuse, -5.10.
88. San Jose State, -5.75.
89. Louisiana Lafayette, -6.36.
90. Tulsa, -6.39.
91. Louisiana Monroe, -6.98.
92. Middle Tennessee State, -7.48.
93. Central Michigan, -8.27.
94. Bowling Green, -8.36.
95. Ohio, -8.37.
96. South Florida, -8.38.
97. Connecticut, -8.86.
98. Hawaii, -8.87.
99. Kansas, -9.27.
100. Florida International, -9.39.
101. Rice, -9.96.
102. Tulane, -10.03.
103. Kent State, -10.82.
104. UTEP, -11.00.
105. New Mexico, -11.57.
106. Florida Atlantic, -11.84.
107. Akron, -11.93.
108. Wyoming, -12.44.
109. Southern Miss, -12.92.
110. SMU, -13.07.
111. Buffalo, -13.36.
112. Miami (OH), -14.45.
113. New Mexico State, -14.56.
114. Troy, -14.87.
115. North Texas, -15.09.
116. Army, -16.05.
117. Idaho, -16.74.
118. UNLV, -16.94.
119. Eastern Michigan, -21.17.

Factors in The Power Rank’s preseason model

First, the model considers a team’s year end rating from each of the past four seasons. This rating comes from The Power Rank’s algorithm that considers margin of victory in games and adjusts for strength of schedule.

It’s crucial to take a four year window. With the short season in college football, a team can get a calculated rating higher than their true skill level for one year. It’s much harder to achieve a high rating for four straight years.

The second variable in the model is turnover margin. As I’ve written about before, randomness plays a large role in this statistic. A large turnover differential one year for a team has no ability to predict the same quantity next year.

Last, the model considers the number of returning starters. Teams with a large number of returning starters tend to exceed their performance from the previous season due to experience.

Win probabilities for the 2015 Gold Cup

Screen Shot 2015-07-06 at 11.37.32 AMCan the United States win the 2015 Gold Cup?

Winning the Gold Cup means more than than just bragging rights over rival Mexico. With the win, the United States qualifies for the Confederations Cup in 2017, a key tune up tournament for the World Cup in 2018.

If another team wins the Gold Cup, they play a one game playoff against the United States, winner of the 2013 Gold Cup. The winner of this playoff represents North America at the Confederations Cup.

To determine win probabilities for the Gold Cup, I combined three different estimates into an ensemble prediction. Two came from my own calculations that rank international teams on offense and defense (see the bottom of this article), and a third came from the markets.


The Gold Cup usually comes down to the United States and Mexico, but Costa Rica has emerged as a solid third team that made the final 16 of the World Cup last summer.

Let’s look at these three top contenders. The offense and defense rankings come from The Power Rank algorithm and use international matches since January 1st, 2011.

United States

12th offense, 27th defense

The United States looked fantastic in beating Germany in a friendly last month, and they also beat the Netherlands in another friendly. Both matches took place on European soil.

They need to play well at the start of the Gold Cup as their group has Honduras and Panama, two top 50 teams in my world soccer/football rankings. Meanwhile, Mexico has the dregs of CONCACAF (Guatemala, Trinidad and Tobago, Cuba, all ranked lower than 80th) in their group.

Coach Jurgen Klinsmann left central defender Matt Besler, who started every game of last year’s World Cup, off the Gold Cup roster. John Brooks will most likely start, and let’s hope they can improve a defense that has ranked 27th in the world over the last 4 years.


11th offense, 9th defense

By the numbers, Mexico has a slight edge on the United States in my rankings. They don’t have the largest win probability though since the United States will enjoy home field advantage.

I’m not 100% certain the United States should get the full .59 goals for home field. If the United States and Mexico meet in Philadelphia for the final, there will be plenty of Mexican fans wearing green in attendance.

In my ensemble calculations, the market predictions most likely account for the semi-neutral type final in Philadelphia. They gave United States a 38% win probability with Mexico at 36%. The gap was bigger in my two calculations.

For Mexico, striker Javier “Chicharito” Hernandez broke his collar bone and will miss the Gold Cup. He scored 9 goals in 33 matches for Real Madrid last season.

Costa Rica

40th offense, 10th defense

Costa Rica had an amazing World Cup last summer. In winning their group, they sent Italy and England home before the knock out stage. Then they beat Greece to advance to the Round of 16.

Costa Rica relies on its defense, which is ranked 10th in the world over the last 4 seasons. They face a road game against Canada in Toronto in the group stage, but they should win their group before most likely playing the United States in the semi-final.

List of win probabilities for the 2015 Gold Cup

1. United States, 39.1%.
2. Mexico, 33.6%.
3. Costa Rica, 11.7%.
4. Panama, 4.7%.
5. Honduras, 4.7%.
6. Jamaica, 2.3%.
7. Guatemala, 1.8%.
8. Canada, 1.4%.
9. El Salvador, 0.9%.
10. Trinidad and Tobago, 0.7%.
11. Haiti, 0.1%.
12. Cuba, 0.1%.