The Impact of Returning Starters from 2008-2011

Submitted by JohnnyV123 on June 25th, 2011 at 8:07 AM

Inspired by the general concensus that the number of returning starters in college football matter and a diary by NOLA Blue in which he discussed how Michigan would fare against opponents in 2011-12 based on returning starters, some of the comments (including my own) criticized looking at pure numbers of returning starters rather than the actual players returning.

It got me thinking if there was any predictablity to be found in pure numbers of returning starters (from now on referred to as RS) and if that translated into wins the next year by having a high amount of RS or losses by having a low amount.

Using Phil Steele's lists of RS I looked at the record for every team in a BCS conference plus Notre Dame in 2008-09, then listed how many starters they would be returning for the 2009-10 season, then added their record for the 2009-10 season, and noted the change in the amount of wins between the two seasons. I repeated that for the 2009-10 season going into the 2010-11 season.

One important note is that I had to decide what to do when teams played a different amount of games in consecutive seasons. For example a team plays 13 games in season one and goes 10-3. The next year in season two the same team plays 14 games and goes 10-4. Technically, they won the same amount of games in both years and the difference in wins is 0 but the team had an extra game to get 10 wins. I decided to handle this by using 0.5's  In this case I would give the team a win change of -0.5 for winning the same amount of games but having an extra game to do it in. This also works to the benefit of some teams.

Here is what I came up with:

Seasons 2008-09 to 2009-10

Team 2008-09 Record 2009 Returning Starters (* Denotes QB Return) 2009-10 Record Net Win Change
         
Florida State 9-4 13* 7-6 -2
Boston College 9-5 14 8-5 -0.5
Maryland 8-5 10* 2-10 -5.5
Wake Forest 8-5 14* 5-7 -2.5
Clemson 7-6 15 9-5 +1.5
NC State 6-7 14* 5-7 -0.5
Virginia Tech 10-4 16* 10-3 +0.5
Georgia Tech 9-4 20* 11-3 +1.5
North Carolina 8-5 15* 8-5 0
Miami 7-6 17 9-4 +2
Virginia 5-7 13* 3-9 -2
Duke 4-8 13* 5-7 +1
Cincinnati 11-3 10* 12-1 +1.5
Pittsburgh 9-4 15* 10-3 +1
West Virginia 9-4 12 9-4 0
Rutgers 8-5 15 9-4 +1
Connecticut 8-5 14* 8-5 0
USF 8-5 14* 8-5 0
Louisville 5-7 15 4-8 -1
Syracuse 3-9 14* 4-8 +1
Penn State 11-2 10* 11-2 0
Ohio State 10-3 12* 11-2 +1
Michigan State 9-4 17 6-7 -3
Iowa 9-4 16* 11-2 +2
Northwestern 9-4 14 8-5 -1
Minnesota 7-6 16* 6-7 -1
Wisconsin 7-6 13* 10-3 +3
Illinois 5-7 15* 3-9 -2
Purdue 4-8 14 5-7 +1
Michigan 3-9 16 5-7 +2
Indiana 3-9 16 4-8 +1
Missouri 10-4 10 8-5 -1.5
Nebraska 9-4 14 10-4 +0.5
Kansas 8-5 16* 5-7 -2.5
Colorado 5-7 14* 3-9 -2
Kansas State 5-7 14 6-6 +1
Iowa State 2-10 17* 7-6 +4.5
Texas 12-1 17* 13-1 +0.5
Oklahoma 12-2 15* 8-5 -3.5
Texas Tech 11-2 13 9-4 -2
Oklahoma State 9-4 14* 9-4 0
Baylor 4-8 18* 4-8 0
Texas A&M 4-8 17* 6-7 +1.5
USC 12-1 12 9-4 -3
Oregon 10-3 9* 10-3 0
Oregon State 9-4 12* 8-5 -1
California 9-4 17* 8-5 -1
Arizona 8-5 14 8-5 0
Arizona State 5-7 15 4-8 -1
Stanford 5-7 18* 8-5 +2.5
UCLA 4-8 17* 7-6 +2.5
Washington State 2-11 15* 1-11 -0.5
Washington 0-12 18* 5-7 +5
Florida 13-1 20* 13-1 0
Georgia 10-3 16 8-5 -2
South Carolina 7-6 12 7-6 0
Vanderbilt 7-6 19 2-10 -4.5
Tennessee 5-7 13* 7-6 +1.5
Kentucky 7-6 11* 7-6 0
Alabama 12-2 14 14-0 +2
Mississippi 9-4 17* 9-4 0
LSU 8-5 14* 9-4 +1
Arkansas 5-7 19 8-5 +2.5
Auburn 5-7 17* 8-5 +2.5
Mississippi State 4-8 10* 5-7 +1
Notre Dame 7-6 17* 6-6 -0.5

And here is a table of number of RS and how many won more games, less games, or no change

Number of RS No Change + Wins - Wins Total of Teams in
Each Group
9 0 0 0 1
9* 1 0 0
10 0 0 1 5
10* 1 2 1
11 0 0 0 1
11* 1 0 0
12 2 0 1 5
12* 0 1 1
13 0 0 1 6
13* 0 3 2
14 1 4 2 15
14* 3 2 3
15 0 2 2 9
15* 1 1 3
16 0 2 1 7
16* 0 2 2
17 0 1 1 10
17* 1 5 2
18 0 0 0 3
18* 1 2 0
19 0 1 1 2
19* 0 0 0
20 0 0 0 2
20* 1 1 0

Here is some info to take away from this.

When I refer to same, more, or less I am talking about the amount of wins between the two seasons.

Overall Win Amount: 13 same (19.69%), 29 more (43.93%), 24 less (36.36%) Total: 66 teams

RS with a QB win difference: 10 same (23.26%), 19 more (44.19%), 14 less (32.56%) Total: 43 teams

RS without a QB win difference: 3 same (13.04%), 10 more (43.48%), 10 less (43.48%) Total: 23 teams

9-13 RS: 5 same (27.78%), 6 more (33.33%), 7 less (38.89%) Total: 18 teams
14-16 RS: 5 same (16.13%), 13 more (41.94%), 13 less (41.94%) Total: 31 teams
17-20 RS: 3 same (17.65%), 10 more (58.82%), 4 less (23.53%) Total: 17 teams

I figured teams would be more successful with a returning QB and while that is supported somewhat in these years with 44.19% of teams going on to a better record the next season the teams without a returning QB were equally as likely to be more or less successful proving the lack of an experienced QB didn't significantly lessen the chances of improvement.

As the number of RS increased more teams did improve but I was surprised to see that not until a team returned 17 starters was it significantly more likely to. In the 15 or 16 RS number it still seemed close to a 50/50 to expect more or less wins.

More after the break

Seasons 2009-10 to 2010-11 

Team Name 2009-10 Record 2010 Returning Starters
(* Denotes QB Return)
2010-11 Record Net Win
Change
         
Florida State 7-6 14* 10-4 +2.5
Boston College 8-5 15* 7-6 -1
Maryland 2-10 13 9-4 +6.5
Wake Forest 5-7 15 3-9 -2
Clemson 9-5 14* 6-7 -2.5
NC State 5-7 13* 9-4 +3.5
Virginia Tech 10-3 13* 11-3 +0.5
Georgia Tech 11-3 18* 6-7 -4.5
North Carolina 8-5 19* 8-5 0
Miami 9-4 16* 7-6 -2
Virginia 3-9 16 4-8 -1
Duke 5-7 16 3-9 -2
Cincinnati 12-1 13 4-8 -7.5
Pittsburgh 10-3 12 8-5 -2
West Virginia 9-4 17 9-4 0
Rutgers 9-4 14* 4-8 -4.5
Connecticut 8-5 16* 8-5 0
USF 8-5 14* 8-5 0
Louisville 4-8 14* 7-6 +2.5
Syracuse 4-8 20 8-5 +3.5
Penn State 11-2 13 7-6 -4
Ohio State 11-2 15* 12-1 +1
Michigan State 6-7 15* 11-2 +5
Iowa 11-2 14* 8-5 -3
Northwestern 8-5 16 7-6 -1
Minnesota 6-7 12* 3-9 -2.5
Wisconsin 10-3 18* 11-2 +1
Illinois 3-9 14 7-6 +3.5
Purdue 5-7 12 4-8 -1
Michigan 5-7 15* 7-6 +1.5
Indiana 4-8 14* 5-7 +1
Missouri 8-5 19* 10-3 +2
Nebraska 10-4 18* 10-4 0
Kansas 5-7 16 3-9 -2
Colorado 3-9 16* 5-7 +2
Kansas State 6-6 15 7-6 +0.5
Iowa State 7-6 13* 5-7 -1.5
Texas 13-1 14 5-7 -7
Oklahoma 8-5 15* 12-2 +3.5
Texas Tech 9-4 15* 8-5 -1
Oklahoma State 9-4 10 11-2 +2
Baylor 4-8 14* 7-6 +2.5
Texas A&M 6-7 19* 9-4 +3
USC 9-4 12* 8-5 -1
Oregon 10-3 18* 12-1 +2
Oregon State 8-5 17 5-7 -2.5
California 8-5 16* 5-7 -2.5
Arizona 8-5 13* 7-6 -1
Arizona State 4-8 10 6-6 +2
Stanford 8-5 16* 12-1 +4
UCLA 7-6 15* 4-8 -2.5
Washington State 1-11 18* 2-10 +1
Washington 5-7 20* 7-6 +1.5
Florida 13-1 15 8-5 -4.5
Georgia 8-5 18 6-7 -2
South Carolina 7-6 17* 9-5 +1.5
Vanderbilt 2-10 11* 2-10 0
Tennessee 7-6 13 6-7 -1
Kentucky 7-6 13* 6-7 -1
Alabama 14-0 10* 10-3 -3.5
Mississippi 9-4 11 4-8 -4.5
LSU 9-4 12* 11-2 +2
Arkansas 8-5 18* 10-3 +2
Auburn 8-5 16 14-0 +5.5
Mississippi State 5-7 17 9-4 +3.5
Notre Dame 6-6 17 8-5 +1.5

 

Number of RS No Change +Wins -Wins Total Teams
 in
Each Group
10 0 2 0 3
10* 0 0 1
11 0 0 1 2
11* 1 0 0
12 0 0 2 5
12* 0 1 2
13 0 1 3 9
13* 0 2 3
14 0 1 1 10
14* 1 4 3
15 0 1 2 10
15* 0 4 3
16 0 1 4 10
16* 1 2 2
17 1 2 1 5
17* 0 1 0
18 0 0 1 7
18* 1 4 1
19 0 0 0 3
19* 1 2 0
20 0 1 0 2
20* 0 1 0

Overall Win Amount: 6 same (9.09%), 30 more (45.45%), 30 less (45.45%) Total: 66 teams

RS with a QB win difference: 5 same (12.20%), 21 more (51.22%), 15 less (36.59%) Total: 41 teams

RS without a QB win difference: 1 same (4%), 9 more (36%), 15 less (60%) Total: 25 teams

10-13 RS: 1 same (5.26%), 6 more (31.58%), 12 less (63.16%) Total: 19 teams
14-16 RS: 2 same (6.67%), 13 more (43.33%), 15 less (50%) Total: 30 teams
17-20 RS: 3 same (17.65%), 11 more (64.71%), 3 less (17.65%) Total: 17 teams

RS QBs seemed to be more important in these years 51.22% improved on wins with a returning QB and even more 60% got worse when they did not have one. Still the effect of a QB is not quite as overwhelming as might be expected.

Again with the magic number of 17 RS being the big difference where you could predict it was likely teams with 17 RS or more would improve. More teams actually got worse when they returned 16 starters and with 15 RS it was 50/50

Combined numbers for the two:

Overall: 19 same (14.39%), 59 more (44.70%), 54 less (40.91%) Total: 132 teams

RS with a QB win difference: 15 same (17.86%), 40 more (47.62%), 29 less (34.52%) Total: 84 teams

RS without a QB win difference: 4 same (8.33%), 19 more (39.58%), 25 less (52.08%) Total: 48 teams

9-13 RS: 6 same (16.22%), 12 more (32.43%), 19 less (51.35%) Total: 37 teams
14-16 RS: 7 same (11.48%), 26 more (42.62%), 28 less (45.90%) Total: 61 teams
17-20 RS: 6 same (17.65%), 21 more (61.76%), 7 less (20.59%) Total: 34 teams

Tidbits:

For teams with 17+ RS and QB in both years combined 5 same (20.83%), 16 better (66.67%), 3 worse (12.50%) Total: 24 teams

Cincinnati was the only team that played less games a season later yet had more wins so even though only 1 win better gave them +1.5
Texas is only team to play 2 less games a season later so even though they won 8 less games they get only -7 (Two 0.5's)

Analysis:

While these numbers are fun to look at and see RS alone only seems to matter for the really large number of RS or really small number, these numbers do very little to accomplish predicting the future. With only two sets of data that I have looked at so far it is hard to say anything is a trend. I may add the 2007-08 to 2008-09 season and the 2010-11 to 2011-12 eventually too to hopefully increase accuracy.

There are other problems like including a team that goes from 14-0 to 13-1 as getting worse in these numbers but hopefully that is counteracted by teams going 2-10 and getting better.

So many more things can be done with this kind of data if someone's so inclined like including the non BCS teams or comparing the BCS teams against the non BCS. Maybe do something more clever like score how many wins a team improves by into some type of ranking system

Maybe RS do not predict anything but still as a Michigan fan looking at these numbers of 2011 RS and seeing we return 20 starters with that one QB we all love I am hoping they go a long way.

Comments

ILL_Legel

June 25th, 2011 at 10:01 AM ^

Thanks for doing the analysis.  I appreciate everyone who takes the time to do the work and share the results.

I've always been inclined to just accept that returning starters, especially at QB and offensive line, makes a significant difference.  Looks like I need to rethink my beliefs a bit.

MGoShoe

June 25th, 2011 at 10:24 AM ^

...get around this issue:

There are other problems like including a team that goes from 14-0 to 13-1 as getting worse in these numbers but hopefully that is counteracted by teams going 2-10 and getting better.

May be to calculate the season x, season x+1 winning percentages and winning percentage delta instead of +/- wins.

turtleboy

June 25th, 2011 at 11:40 AM ^

bit of information that could skew the results involves factoring returning coaches/new coaches in the model. I've always looked at returning starters as less of an advantage, and more of a lack of disadvantage. 

Sextus Empiricus

June 25th, 2011 at 12:30 PM ^

much??? Thanks for this read.  Vanderbilt is an outlier...they didn't get it done clearly.

Here's a fit and a 100 spline with histograms shaded for returning QB (Blue in the chart). 

Compiling this data takes a long time...it's fun to look at...thanks for this.

 

Photobucket

rkfischer

June 25th, 2011 at 12:16 PM ^

 

Excellent work. I agree with MGoShoe about added some calculations about an uneven season. I’m not what your goal is. If the goal is to build a model for predicting success . . . there are probably some people connected to betting that might shed light on how they build quantitative models for college football. They have some big financial incentives to build and fine tune models that are better than normal models.

Factors for these models would include:

Quality of RS (# of stars by various rating agencies) and by skill positions as well as groups (defensive backfield and offensive line)

# of years of experience (especially for QB/wide receivers and offensive line combo’s) where collaboration/timing and communication are so important

Quality of competition, their previous record and #/quality of RS

Injuries by both current team and competition (example is having someone like Troy Woolfolk returning)

Finally having some quality and quantity number of RS seems like it would be favorable. However, your numbers show that just the quantity of RS is not that important. RS QB is a much bigger factor. I’m sure there are other factors in the quality of RS would also help make your model a good predictor. I like the thinking.

DanGoBlue

June 25th, 2011 at 12:25 PM ^

Really nice analysis. Like ILL_Legal, I've always accepted the RS-bit and never really questioned it. I'm definitely intrigued now and want to dig deeper. Some ways to further refine this analysis:

- Expand the data set: analyze more seasons than the two above; I understand that's a lot of additional work, but it provide better info

- Factor in strength of schedule: as a way to weight win totals and account for more or less patsies in a given year

- Win factor of some sort: there has to be a better way to account for different numbers of games from year to year; I think you are on the right track, but there is a certain arbitrariness of assigning a 1/2 win that doesn't sit well. Maybe it's based on win percentage instead?

- Positional weight: are all RS equal? You already seem to be headed in this direction by noting returning QBs, and maybe this factors into the RS number somehow. Are there other positions that should be similarly factored?

EGD

June 25th, 2011 at 1:10 PM ^

I wonder what the numbers would look like if you compared teams just based on the number of upper-classmen starters, irrespective of whether those players had started the previous season.  Intutively, it would seem that a team which can replace a graduating starter with a junior or senior is unlikely to suffer as much as if a freshman has to step in.  Also, NOLA Blue's diary focused on having returning starters on the offensive and defensive lines, where having two or three years of collegiate S&C makes the most difference.  

Ziff72

June 25th, 2011 at 1:20 PM ^

Why are people down on returning starters?   Maybe I'm not reading this correctly but the numbers are pretty clear.  The more starters you have returning the better your chances of winning more games and it's pretty clear cut.  Just browsing thru you can pretty much pinpoint what happened when teams failed to achieve.   18 or more starters returning like us.

GT - They tanked.  They lost their top 2 offensive weapons and best defensive player so while they only lost 4 starters they took a toll.  Turnover and the lack of passing game killed them.

NC - They tanked as well, but they didn't have half those starters they were all suspended.

Syracuse -+4 wins

Wisc-+1 They didn't have too far to go.

A&M - +4 Great Year

Oregon- Obviously

WSU- Hey they won 1 more.  In reality they were much more competitive than they were the year before 

Was- Despite a brutal schedule they improved and made a bowl.

Georgia-Went down....True Frosh at QB top WR suspended pretty clear what happened

Arkansas-Sugar Bowl pretty good.

10 teams only GT was a team that did not perform as you might expect given outside conditions.

 

I    like   our   chances!!!!!

 

 

Irish

June 25th, 2011 at 4:59 PM ^

Probably the system change.  Denard was a massive chunk of the offense last year and he will be in for probably the largest adjustment out of anyone.  

Volverine

June 26th, 2011 at 6:54 AM ^

It seems like we should consider the position group the returning starters come from.

For Michigan, returning 10 starters from one of the country's best offenses might be more advantageous than returning ten from D.

Also, I don't think the fact that we return two starting kickers (if they count) helps in any way. (See Spring game)

I guess I think the numbers can be affected because a team can return 15 starters, but that evenly split between O, D, and SP so it wouldn't make as much a difference as returning nearly every starter from O or D exclusively. Does any of this make sense?

Again, I'm not sure about the results, it would just seem to affect the date a bit.