Who’s the “Best” College Football Coach in the Past Decade? We think he works for the Philadelphia Eagles.

As the title implies, we are about to go down a road that will inspire debate and we expect considerable hate.  As we suspected, and have since confirmed in our 3 months of publishing, there is no sport with more passionate fans than college football.  We know that as soon as we provide our ranking of college coaches that we will immediately be told that we are wrong (and in rare cases that we are right).

For our coaching analysis, the starting point is the idea that we should rank coaches’ performance relative to the resources that are at their disposal.  In other words, we can’t compare the coaches at the University of Illinois and the University of Florida simply based on win-loss records. For the analysis, we gathered data on results (winning percentage, major bowl participation), football expenditures, historical performance (won-lost records, major bowls, national championships, etc…), attendance and other factors.

We use this data to create models that predict team success based on financial resources, historical performance and market potential.  These factors can all contribute to on-field success.  For example, the logic of including historical performance is that a more storied program may be more attractive to both bowls and to potential recruits.  We use the models to predict the performance of each school for each of the last 10 years.  We then assess the contribution of the coach by comparing actual to predicted performance.

We analyzed coaches using the past 10 years of data in terms of two criteria: winning percentage and selection to play in a major bowl (Rose, Orange, Fiesta, Sugar, and National Championship).   As an aside, our performance models were both estimated using logistic regression.

In terms of incremental winning percentage effects, the top coach was Chris Petersen from Boise State. Coach Petersen has achieved a 91.3% winning percentage while at a school with only moderate football expenditures (even among non-BCS schools) and limited history.  In terms of specific numbers, we find that Petersen has achieved a winning percentage that is 37% higher than what a school of comparable resources and history achieves.  The top five also includes Urban Meyer, Brian Kelly, Bret Bielema and Bobby Petrino.  In positions six through ten we have Steve Spurrier, Bob Stoops, Gary Patterson and Frank Beamer.  Two other former coaches produced notable results on the winning percentage criterion.  Chip Kelly won 32% and Pete Carroll won 13.3% more games than expected.

Okay so who is missing?

We already anticipate complaints from Alabama fans along the lines of “The list is invalid because at Alabama we play for championships, not for winning an incremental game or two.”  Well, Coach Saban does finish 11th on the incremental winning list, and there is some merit to this argument.  It is more difficult to drive incremental wins at schools like Alabama than Boise State.

In part two of our analysis we looked at incremental participation rates in BCS bowl games (not just the BCS championship).  Our approach was similar as in the winning percentage analysis and used the same set of predictor variables.  On this metric, the top performer was Bret Bielema.  Bielema’s record includes taking Wisconsin to the last three Rose Bowls.  In terms of percentages, we find that with Bielema in charge, Wisconsin improved their rate of BCS bowl participation by 28%.  It looks like Arkansas made a great choice! In positions two through five we have Chris Petersen, Bob Stoops, Frank Beamer and Urban Meyer.  And where does Coach Saban fall?  Just behind Urban Meyer in 6th place.

If we also look at former college coaches one name really stands out.  Chip Kelly was by far the leader on the BCS bowl participation metric. Combined with the winning rate results, one can argue that Chip Kelly has been the most effective college coach over the past few years.

Returning to Coach Saban, first, it must be noted that he scores really well on our measures (11th for incremental wins and 6th for incremental BCS games).  And we definitely understand the argument that he should be number one.  In terms of winning championships, it is hard to argue that he is not the go-to coach.

Mike Lewis & Manish Tripathi, Emory University, 2013.

Introducing Our Heisman Series…

July 14, 2013

The annual competition for the Heisman trophy is one of our favorite things to watch at Emory Sports Marketing Analytics because it is the perfect synthesis of marketing and sports.  While the Heisman marketing campaigns started to attract public attention with Oregon’s notorious Joey Harrington billboard in 2001, we now have the tools to do real time tracking of the buzz that surrounds Heisman candidates.

This fall we will be monitoring the race for the Heisman using a combination of performance, marketing and social media data.  To begin our series, we collected Twitter activity for the past 90 days for all the Heisman candidates that currently have betting lines established.  We then ranked the candidates and created the pie chart below that shows the share of tweets of the top five candidates.  The “other” category is the share for candidates ranked 6 through 10.

Over the course of the season we will update the chart each week and provide data that explains each week’s changes.  This is something of an experiment for us as we don’t really know how closely social media buzz will correlate with eventual Heisman voting or even how marketing campaigns and on-field performance will drive weekly changes.

Who’s Ready for Some College Football?

Please note that the chart below colors states based on their percentile rank of share of Twitter voice for college football (e.g. Alabama has the highest percent of tweets that are college football related, thus its percentile rank is 100).

The start of the college football season is less than two months away.  Fans have been waiting for the 2013 season to kickoff ever since the Alabama-Notre Dame game ended (which was in the 1st quarter).  The chart above is a Geo-based Twitter analysis of conversation about college football by state in the last six months.  We essentially rank states by the percentage of all Twitter conversations in that state that mention college football.  The top five states are: 1) Alabama, 2) Mississippi, 3) Tennessee, 4) South Carolina, and 5) West Virginia.

Best & Worst Colleges for Converting Elite Recruits into NFL Draft Picks


As a follow-up to our previous post, we would like to provide a more complete view of the best and worst colleges based on converting elite recruits into NFL Draft picks (number of draft picks divided by the number of four and five star recruits).  This study considers the last six NFL drafts, and examines the top sixty schools in terms of relevant recruiting cycles over the past twelve years.  We exclude schools that have not produced a single draft pick in the 2007-2012 period or that have less than four elite recruits over the relevant recruiting cycles.

Since there have been a considerable number of inquiries, it should be noted that the University of Alabama finished in the middle of the pack.  While the table above is produced from a rather straight-forward statistical analysis (descriptive statistics for each school), we would like to stress that our main results are derived from multiple linear regression models.  The key result is the significant negative relationship between the number of four-star prospects and the draft conversion rate of high ranked prospect. It is also worth noting that while we focused on the ratio of draft picks to elite recruits we also conducted a number of other analyses that focused on alternative measures of draft success such as the number of Day 1 or First round picks per elite recruit.  Our key finding was replicated across the various models.

By  Dr. Mike Lewis & Dr. Manish Tripathi, Emory University 2013

Methodology for the study explained here.

Want to Get Drafted in the NFL? Go to Kentucky, Not Alabama!

For our latest posts and updates, follow @sportsmktprof

In the 2012 NFL draft, the first round selections included multiple players from Alabama, USC, LSU and Notre Dame.  The Crimson Tide had 4 of the first 25 picks!  To hardcore college football fans, these first round draft results are expected since colleges like Alabama and USC are perennial members of the lists of top recruiting classes.

A statistical analysis of draft picks versus college recruiting rankings confirms this simple story.  A regression analysis of the number of picks selected from a given school versus the number of five-star, four-star, and three-star players in the school’s recruiting class reveals a significant positive relationship between the number of players drafted and the number of four and five-star recruits.  The relationship between three-star recruits and draft picks is insignificant.  Specifically, we found that every five-star player signed by a school translates to 0.33 draft picks, and every four-star player translates to approximately 0.09 draft picks.  If we examine only players selected in the first three rounds of the draft, then each five-star recruit produces 0.23 picks and each four-star recruit results in 0.05 picks.

While the preceding results provide evidence that fans should be happy about recruiting victories, the story from the high school recruit’s perspective is far more complex.  That each five-star player only results in 0.33 picks obviously suggests that there is a great deal of error in the rankings.  However, an additional explanation is that a player’s draft outcome may be adversely impacted by joining programs with many other highly rated recruits.

To explain the situation, we performed several additional analyses that examined the conversion rate of college recruits to drafted players.  For these analyses, the measure of interest was the number of NFL draft picks from a school divided by the total number of four and five-star recruits signed by the school.  We then modeled this conversion rate by the number of five, four and three-star players signed by the school.  In this analysis, we found a significant negative relationship between the number of four-star prospects and the draft conversion rate of high ranked prospects.  When we limited the draft picks to only round one or only day one (rounds 1-3), this negative relationship persisted.

To further illustrate this point, let’s examine two schools.  Over the last six NFL drafts, the University of Florida has brought in one hundred four and five-star recruits, and had “only” twenty-nine players drafted.  The average recruiting class rank for the Gators over this period was 6.1.  In contrast, the University of Pittsburgh brought in twenty-five four and five-star prospects but had seventeen draft picks.  Pitt’s average recruiting rank was 32.

The following chart shows the draft pick conversion rate for a variety of schools.  Interestingly, over the last six drafts, the most effective school at converting prospects to picks is the University of Kentucky.  Kentucky actually produced more draft picks than the number of four and five star prospects they recruited.  At the other extreme, Florida State only produced about 0.2 draft picks per elite recruit.

(*Elite = Five and Four Star Recruits)

We also extended the analysis to include other factors that could impact a school’s ability to produce draft worthy players.  Specifically, we examined the influence of the school’s investment in its football program, the team’s winning percentage, whether or not the school played in a major bowl, and each school’s conference affiliation.  Investment may matter because greater resources could translate to improved coaching or strength training programs.  While the latter three factors relate to the publicity players receive at various schools.  The only significant factor from these additional variables was school’s investment in the football program.

The bottom line seems to be that for players with a goal of playing in the NFL, program selection should not be based on the glamour provided by the big time programs such as Ohio State, Alabama, Notre Dame, and USC.  Rather players should seek out opportunities at schools with substantial budgets but lower ranked recruiting classes.  In other words, it’s probably more important to increase your probability of getting on to the field early, rather than maximizing the number of times you play on a big national stage.

To further illustrate the preceding point, in 2009, Alabama had the top rated recruiting class with several four-star high school recruits.  A current examination of these recruits shows that several have left the team, others have red-shirted, and a few are projected as free agents or late-round picks.  One recruit is projected as an early-round pick.  These outcomes could be the result of over-recruiting and the lack of resources for top recruits.

Finally, using first round mock draft data for this year’s draft, we can determine the number of projected first round draft picks per five and four star recruits at a given school.  The chart below highlights some of the schools with projected first round picks this year.  It is interesting to note that schools such as Syracuse and Missouri do a better job with conversion of elite recruits than the football powerhouses of Alabama, Georgia, and Florida.

(*Elite = Five and Four Star Recruits)

By  Dr. Mike Lewis & Dr. Manish Tripathi, Emory University 2013

Methodology for the study explained here.