Analytics Posts

SEC Performance Rankings through Week 2

Using a straightforward approach of comparing points-per-play (PPP)on offense and defense, then calculating the difference, we can see how teams are doing overall. The difference was then standardized to give a sense of “distance” between teams beyond just that of ordinal ranking.

TeamPPP Differential
Alabama0.932
Florida0.668
Georgia0.540
Tennessee0.391
LSU0.317
Auburn0.069
MSU0.018
Arkansas-0.110
Ole Miss-0.168
Kentucky-0.415
Texas AM-0.504
USCe-0.540
Missouri-0.578
Vandy-0.621
Standardized PPP Differential

Below is how each team ranks in PPP overall, offensively, and defensively:

TeamRank overallOffensive RankDefensive Rank
Alabama114
Florida228
Georgia361
Tennessee453
LSU546
Auburn6112
MSU777
Arkansas8125
Ole Miss9314
Kentucky10812
Texas AM111011
USCe12913
Missouri131310
Vandy14149

Alabama is looking like a clear number one. Florida’s offense is humming, but need to shore up that defense. Georgia is very close to Florida overall and can easily pass them as the weeks go on. Tennessee is performing pretty good through 2 weeks as well.

Some looks at College Football Spending on Recruiting and Various Outcomes

Recently, an analysis of recruiting spending was put out (https://watchstadium.com/this-is-how-much-it-costs-to-land-one-of-college-footballs-top-recruiting-classes-07-24-2019/), and I thought it was excellent. It got me curious, so I used their data and ran a couple of correlations and plotted them. (I normalized all of the variables, correlation is below each graph).

4-star
r = .761

5-star
r = .766

2018 Class Ranking
(Recruiting ranking for 2018 – forgot to invert the scale) r = .641

2019 Class Rankings
Recruiting Rankings for 2019, r = .621

2019 win percentage
2019 win percentage, r = .448

Blue Chips
Blue Chips (4 and 5 star recruits), r = .798

The SEC Championship Game History: A Breakdown.

As a Florida Gator fan, I hear a lot of fans groan about UF’s current recruiting. I understand it. From a composite ranking point of view, UF is doing quite well nationally, but is middle of the pack in the SEC (7th ranked in terms of team talent on roster in 2019, https://247sports.com/Season/2019-Football/CollegeTeamTalentComposite/?Conference=SEC). A lot of fans are also very happy with the improvement of the team since Dan Mullen took over, and expect the team to continue to improve and make a championship run. I’m an optimist, so I’ll go with the latter. But, to see where UF currently sits relative to previous SECCG participants, I took a look at the 4-year moving average (MA) of recruit ratings for each SEC team going back to 2003. And for fun, I looked at the SECCG results and found some interesting superlatives, which I put at the bottom of this post.

The density chart below shows the 4-yr MA for each team that has participated in the SECCG since 2003:

dens

The vertical lines are the averages for each group. You can see the distribution is left-skewed. The outliers, as determined using Median Absolute Deviation (MAD) were all on the low end. I kept them in, as I don’t need a Gaussian distribution for this effort.

UF’s current 2020 MA (blue vertical line) is between the average winner MA (green vertical line) to the right and average loser MA (red vertical line) to the left. The scores on the x-axis are standardized by year to control for fluctuation in national ratings. Here is the data table:

data table 1

Florida’s current 2020 MA for average recruit rating is 90.41, which is 1.692 standard deviations above the national average. Looking at that same data table with heat mapping on the scores, you can see how teams have ebbed and flowed over the years:

data table 2

A couple things are obvious from the table above. The first is that the SEC got much better overall around 2009-2010, Another is that UF clearly faded a bit from 2015-2018 in terms of 4-year MA. Back to the point: How does UF’s current average recruit rating (RR) rank relative to previous SECCG participants:

SECCG barchart

This bar chart shows the 4-year MA for RR of winners (orange) and losers (blue) of the SECCG since 2003. UF 2020 is the green bar.

Here is the same data by year, which shows the difference between the opponents:

SECCG bar chart by year Clearly there have been some talent disparities in the SECCG. The biggest disparity occurred when Missouri (4-yr MA of 85.15, Std_Score of 0.764) played Alabama (4-yr MA of 93.12, Std_Score of 2.421) in 2014.

Where Florida Stands Now vs History:

Florida’s current 4-yr MA is 90.41, or 1.7 standard deviations above the national average. This would put the current UF roster in the 65th percentile of historical (2003-present) SECCG participants. UF would be in the 54th percentile of SECCG winners and in the 76th percentile of SECCG losers.

Some Superlatives:

data table 3

Way to go, dawgs.

Various data sets:

data1

big table

credit: https://en.wikipedia.org/wiki/SEC_Championship_Game

 

2019 1000 Yard Rushers in CFB

You can scroll across the table using the bar at the bottom.

Player School Conf G Att Yds Avg TD Rec Yds Avg TD TD Att/G z_att/g
Chuba Hubbard Oklahoma State Big 12 13 328 2094 6.4 21 23 198 8.6 0 21 25.2 2.310
Malcolm Perry Navy American 13 295 2017 6.8 21 0 0 0 21 22.7 1.534
J.K. Dobbins Ohio State Big Ten 14 301 2003 6.7 21 23 247 10.7 2 23 21.5 1.169
Jonathan Taylor Wisconsin Big Ten 14 320 2003 6.3 21 26 252 9.7 5 26 22.9 1.584
Jaret Patterson Buffalo MAC 13 312 1799 5.8 19 13 209 16.1 1 20 24.0 1.934
AJ Dillon Boston College ACC 12 318 1685 5.3 14 13 195 15 1 15 26.5 2.698
Travis Etienne Clemson ACC 15 207 1614 7.8 19 37 432 11.7 4 23 13.8 -1.186
Javian Hawkins Louisville ACC 13 264 1525 5.8 9 4 58 14.5 0 9 20.3 0.805
Darrynton Evans Appalachian State Sun Belt 14 255 1480 5.8 18 21 198 9.4 5 23 18.2 0.164
LeVante Bellamy Western Michigan MAC 13 266 1472 5.5 23 15 55 3.7 0 23 20.5 0.852
Lynn Bowden Jr. Kentucky SEC 13 185 1468 7.9 13 30 348 11.6 1 14 14.2 -1.054
Kenny Gainwell Memphis American 14 231 1459 6.3 13 51 610 12 3 16 16.5 -0.360
Tra Barnett Georgia State Sun Belt 13 248 1453 5.9 12 16 69 4.3 0 12 19.1 0.428
Zack Moss Utah Pac-12 13 235 1416 6 15 28 388 13.9 2 17 18.1 0.122
Clyde Edwards-Helaire LSU SEC 15 215 1414 6.6 16 55 453 8.2 1 17 14.3 -1.023
Brenden Knox Marshall CUSA 13 270 1387 5.1 11 14 129 9.2 0 11 20.8 0.946
Kylin Hill Mississippi State SEC 13 242 1350 5.6 10 18 180 10 1 11 18.6 0.287
Josh Johnson Louisiana-Monroe Sun Belt 12 201 1298 6.5 11 13 122 9.4 0 11 16.8 -0.284
Jalen Hurts Oklahoma Big 12 14 233 1298 5.6 20 2 25 12.5 1 21 16.6 -0.316
Xavier Jones SMU American 13 244 1276 5.2 23 20 90 4.5 2 25 18.8 0.334
Caleb Huntley Ball State MAC 12 248 1275 5.1 12 7 25 3.6 0 12 20.7 0.914
Xazavian Valladay Wyoming MWC 12 247 1265 5.1 6 11 211 19.2 2 8 20.6 0.889
Michael Warren II Cincinnati American 14 261 1265 4.8 14 21 153 7.3 2 16 18.6 0.295
Charles Williams UNLV MWC 12 212 1257 5.9 11 7 54 7.7 0 11 17.7 -0.003
Najee Harris Alabama SEC 13 209 1224 5.9 13 27 304 11.3 7 20 16.1 -0.489
CJ Verdell Oregon Pac-12 14 197 1220 6.2 8 14 125 8.9 0 8 14.1 -1.103
D’Andre Swift Georgia SEC 14 196 1218 6.2 7 24 216 9 1 8 14.0 -1.125
Gaej Walker Western Kentucky CUSA 13 241 1208 5 8 24 140 5.8 0 8 18.5 0.263
Bryant Koback Toledo MAC 12 195 1187 6.1 12 8 69 8.6 2 14 16.3 -0.436
Rodney Smith Minnesota Big Ten 13 228 1163 5.1 8 7 70 10 0 8 17.5 -0.042
Elijah Mitchell Louisiana Sun Belt 14 198 1147 5.8 16 10 70 7 1 17 14.1 -1.081
Cam Akers Florida State ACC 11 231 1144 5 14 30 225 7.5 4 18 21.0 1.016
Rakeem Boyd Arkansas SEC 12 184 1133 6.2 8 19 160 8.4 0 8 15.3 -0.717
Jonathan Ward Central Michigan MAC 12 183 1108 6.1 15 34 329 9.7 1 16 15.3 -0.742
Jason Huntley New Mexico State Ind 12 154 1090 7.1 9 40 192 4.8 2 11 12.8 -1.481
Torrance Marable Coastal Carolina Sun Belt 12 204 1085 5.3 11 38 295 7.8 3 14 17.0 -0.207
Eno Benjamin Arizona State Pac-12 12 253 1083 4.3 10 42 347 8.3 2 12 21.1 1.042
Ben LeMay Charlotte CUSA 11 193 1082 5.6 9 19 242 12.7 4 13 17.5 -0.040
Kobe Lewis Central Michigan MAC 14 182 1074 5.9 12 23 164 7.1 0 12 13.0 -1.430
Justin Henderson Louisiana Tech CUSA 13 188 1062 5.6 15 24 200 8.3 1 16 14.5 -0.983
Pooka Williams Kansas Big 12 11 203 1061 5.2 3 27 214 7.9 2 5 18.5 0.238
Joshua Kelley UCLA Pac-12 11 229 1060 4.6 12 11 71 6.5 1 13 20.8 0.961
Asher O’Hara Middle Tennessee State CUSA 12 199 1058 5.3 9 0 -5 0 9 16.6 -0.334
Tra Minter South Alabama Sun Belt 12 193 1057 5.5 5 32 209 6.5 0 5 16.1 -0.487
Kadin Remsberg Air Force MWC 13 181 1050 5.8 8 3 24 8 0 8 13.9 -1.148
Shamari Brooks Tulsa American 12 227 1046 4.6 6 9 49 5.4 1 7 18.9 0.379
Frankie Hickson Liberty Ind 13 187 1041 5.6 12 11 96 8.7 1 13 14.4 -1.007
Kevin Marks Buffalo MAC 13 227 1035 4.6 8 12 41 3.4 0 8 17.5 -0.066
Ke’Shawn Vaughn Vanderbilt SEC 12 198 1028 5.2 9 28 270 9.6 1 10 16.5 -0.360
John Rhys Plumlee Ole Miss SEC 9 154 1023 6.6 12 0 0 0 12 17.1 -0.173
Tre Harbison Northern Illinois MAC 11 230 1021 4.4 8 6 47 7.8 0 8 20.9 0.988
Salvon Ahmed Washington Pac-12 12 188 1020 5.4 11 16 84 5.3 0 11 15.7 -0.615
George Holani Boise State MWC 14 192 1014 5.3 7 26 206 7.9 3 10 13.7 -1.212
Kevin Mensah Connecticut American 12 226 1013 4.5 9 8 91 11.4 0 9 18.8 0.354
Kennedy Brooks Oklahoma Big 12 13 155 1011 6.5 6 10 79 7.9 0 6 11.9 -1.760
Michael Carter North Carolina ACC 13 177 1003 5.7 3 21 154 7.3 2 5 13.6 -1.242

Which states received the best recruits? 2019 Blue Chip Migration Rates by State

There are a lot of articles about how it is important to successfully recruit talent in a school’s own state. Here is how each state did in terms of Blue-Chip (composite 4 and 5-stars) acquisition.

MigMap19

The percentages in the map and waterfall chart represent the net gain/loss for the number of blue-chips for that state. The totals are outlined in the table below.

chartmig19

 

tablemig19

Ultimately, 252 (66%) of 382 Blue-Chips left their home state. Texas, Florida, Georgia, and California had the highest levels of loss. To control for population differences, I then standardized the number of Blue-Chips produced and received and subtracted the produced from received to see how the states did proportionally.

stdmapmig19

stdtablemig19

When controlling for number of players produced, Florida moves up some and Texas moves up a bunch.

Impact of Talent Ratings on 2019 Win % in the SEC by Position Groups

We can all infer that not all position groups impact the outcome of a season equally. This could be due to numbers (5 OL vs 1 or 2 TE) or nature of importance (QB vs long snapper). I’ve already looked at the correlation between talent level and outcome by whole roster and by offense vs defense. So this time, I just broke the same data down a little further by position groups.

Each scatter plot below has a box that shows the correlation and effect size. The correlation is how closely the two variables (2019 win % and average composite talent rating for that team’s position group) are related. The effect size (technically, a coefficient of determination, but that can be interpreted as effect size in regression) essentially shows how reliable that correlation is. Every relationship was statistically significant at 0.06 or lower.

radarSECposgroup2019

Offensive Line:

OLsec2019

perfOL

Running Back:

RBsec2019

perf.RB

Wide Receiver:

WRsec2019

perfWR

Tight End:

TEsec2019

perfTE

Defensive Line:

DLsec2019

perfDL

Linebacker:

LBsec2019

perfLB

Defensive Back:

DBsec2019

perfDB

Quarterback:

QB was done differently due to sample sizes. When I used the average rating of all QBs on each roster, the outcome was random. In situations like Florida, where the highest rated QB was not the starter, the highest rated was used (So, Feleipe Franks for UF). Of course, this doesn’t make sense on the surface, but I surmised that using the highest ranking individual is a general indicator of how good the QB situation is for a team. If that QB gets beat out by a lower rated player, then the lower rated player may have been undervalued (or the higher rated player was a bust…). Either way though, using the highest rated player at the QB position had some value, so I used it. If you have a better explanation (and I hope you do), feel free to share.

QBsec2019

perfQB

Position Group Effect Table

Here is a table summarizing effect size by position

posEffect

Surprising to me, is that the LB position group had the strongest correlation to win % for SEC teams in 2019. Keep in mind, this isn’t how important a position group is to winning – this is how well the composite rating of that position correlates to winning. Most of us would agree that effective QB play is key to winning. Here is that output visually in a radar graph:

radarSECposgroup2019

Expectations:

Expectations scores around zero, whether positive or negative should be considered as meeting the expectations. Here is the table with each team’s total:

talenttable

Looking at each team’s performance relative to expectations by position group is a neat way to peer into how well the on-field coaching is. It is important to note that teams with higher talent ratings will have harder times surpassing expectations and it will be easier to fail expectations. Not surprisingly, LSU had the best on-field coaching performance. Winning all your games and a Natty should cement that. Relative to their on-roster talent levels, Florida was second only to LSU. Imagine what Dan Mullen could do with elite talent… he’s already not far off with the talent he has.

Takeaway:

Dan Mullen is probably the best on-field coach in the SEC. I am skeptical LSU will have anywhere near the success they had last year when they caught lightning in a bottle. I expect Saban and Alabama to return to the top of the West. Georgia can recruit like crazy, but just doesn’t perform up to that talent level. They are a good team, that should be great.

I would focus hand-wringing over recruiting on QB, WR, and LB. The need for an elite QB is a given (though nothing I found here would indicate that, likely because of the small number of rated QBs on a given roster). But if 2020 holds form with 2019, LB roster talent and WR roster talent will more strongly predict win percentage. RB and TE position groups are the weakest predictors. We will see how this holds up in 2020.

 

Number of 3-stars on roster negatively correlates to wins: SEC 2019

Just playing with some numbers, I found that on-roster 3-star count was pretty strongly correlated with regular-season wins for SEC teams in 2019. I’ve done this in the past, and found that it is a pretty consistent relationship.

3-stars and wins sec 2019

The correlation is 84%, R-squared is 70%, p-value is <.001, and SE is 1.76. Definitely something to this. The relationship between 3-stars and wins is strong than Blue-Chip count and wins, though it is close. This might just be a slightly different way of looking at the same thing, but I found it interesting.

Crystal Ball/Commit Accuracy for the class of 2020 by Month

The graph below shows how accurate Composite Crystal Ball (CB) predictions were relative to a recruit’s ultimate landing spot by month for the 2020 class. This is only for 4-star and 5-star players (composite).

cb

 

I scraped web data every month and recorded it last year. Then, I just compared the CB % and commits to where a recruit ultimately signed. I’m doing the same for the 2021 class to see if it follows any trend.

Players that commit to a team generally stay committed. This is the same time frame of players who were committed to a team and ended up with that team.

commits

Even as far back as February, 71% of Blue Chip recruits did not (ultimately) flip. We will see if this year is any different.

Tracking College Football Recruiting for 2021: Examining Recruiting’s Volatility

This will be a look at how much recruiting changes over the course of a year. I constructed an algorithm that tracks how each program is doing in terms of recruiting for 2021. I will reveal the full methodology after NSD, but the purpose here is to see how chaotic recruiting in college football is. Each school is assigned a random code. The star-points is each school’s current score based upon the algorithm. The column on the right is that school’s rank for that month. Schools with fewer than 4 points are excluded.

I coded each school (I also did for players and positions, but that is irrelevant here) so I don’t know how each school is doing (yes, I could look it up, but I’m not going to do that until after NSD).

Here is the first list:

5/1/2020

school sum_star_points rank_points
m22b8 345 1
d7t22 280 2
u9z14 211 3
r21v4 181 4
f7z2 172 5
v26x18 161 6
r20z2 159 7
c1d19 148 8
e19i12 144 9
l14z25 141 10
d16o5 133 11
k26m12 112 12.5
t25m16 112 12.5
k20i15 97 14.5
d11u16 97 14.5
z22v17 83 16
h4u3 73 17
h1z22 69 18
i24d10 68 19
k18a7 64 20
d17g13 57 21
c20u4 48 23
p16n8 48 23
r5f22 48 23
y5l1 44 25
h15z5 40 26
n12m12 36 27
a19c19 32 28
f18q6 28 31
m10y9 28 31
p24s14 28 31
u10j20 28 31
c26t19 28 31
r5m14 24 34.5
h24h7 24 34.5
l19d7 16 37
n22p23 16 37
x10h10 16 37
g24t21 12 41
k8f6 12 41
l2k21 12 41
o12i11 12 41
s22p11 12 41
q25f10 8 45.5
n1v26 8 45.5
d4t10 8 45.5
o14n6 8 45.5
o17q4 4 48