Quick Analysis of the VT Rushing Attack Over the Years

Well, it's the offseason. A strange combination of me being slightly bored along with a desire to quantify how much fun I was having while watching VT play this past year caused me to go out and obtain some good old fashioned data. Specifically a play-by-play college football database going back to 2002. Armed with this database and a copy of my beloved Matlab, I've spent some time over the last day or two looking through and analyzing VT's rushing attack from 2004 to 2017.


This is probably going to be a really long post, but I would like to start by acknowledging BlueSCar over on reddit for posting the database. If anyone else is interested in digging through it, I got it off of the post below. Note that while the title only lists the database going up until 2016, BlueSCar apparently keeps the link in the OP updated as it also contains the data from 2017.


Analysis Limitations

This database only provides round numbers for play statistics. IE, it only reports yards gained and position markers on plays as whole numbers. This should be good enough for large sample sizes though, especially for what I will be looking at. I also haven't had a chance to confirm it yet, but I believe rushing plays that end in fumbles are not listed as rushing plays. This shouldn't affect too much, but it's something to be aware of.

While today I will only be looking at individual and team rushing statistics, I've already had some friends mention looking at advanced passing statistics like catch rate, times targeted and whatnot. This database does not track any passing/targeting information, throwaways or anything like that. That said, if someone knows of where to get this information, please let me know and I might do something more in-depth on our passing game later on. I enjoy coding and looking at data, but am unwilling to sit through watching hours of film to get that info myself lol!

I have access to most FEI total, offensive and defensive rankings, but have not incorporated them into this analysis yet. It is something that I hope to eventually do, but for now I am being lazy and just setting a 65 attempt/year threshold for looking at any individual players. This ensures that every player should have played against multiple teams each year and the results should be averaged out against multiple teams. I set 65 as I assumed that if a player were getting no more than 20 Att/game, they would have to play against more than 3 teams to be looked at. In reality, all the players we will be looking at were rotated in regularly throughout the year.

Finally, I'd like to point out that I am not a statistician and don't have any deep, inside knowledge of the sport or anything. I do spend a lot of time looking at data at work, but if you are looking for some sort of advanced statistics on this data, I am not the guy to do it. Hopefully, my alternate view on looking at this data might provide some useful insights though.


VT's rushing attack over the years has been a very hot debate that I too had a lot of questions about. The advanced statistics, what Joel Smith has shown in his columns, and in general just our eyeballs have shown that we have lacked a powerful rushing attack over the last few years, with a distinct lack of explosive plays occuring most years. But I wanted to know, what kind of yardage could we expect to get on a rushing play each year? Yds/Att is normally pointed to, but averages are heavily skewed by outliers. Since most running plays only see short losses or a wide range of gains, Yds/Att is almost always skewed high. Besides, trying to condense data down to a single number almost always sees the loss of a lot of information in terms of nuance.

The second question I wanted to answer is one that has been a hot one ever since Shane was our RB coach and continued once Fuente took over using a RB by committee approach. Does using a running by committee approach make it hard for RBs to 'get in rhythm'? While I obviously can't answer what the players do or don't feel, I will be investigating how having a player in for more or less time in a series affects their yardage.

Rushing Profile Results

Let's start by looking at our rushing attack in 2016. Carry percentages are shown in Figure 1 for individual players both for all attempts on the left, and for attempts on 'running downs' on the left. Here and in all future references, I am considering a 'running down' as one of the following situations: 3rd or 4th down and less than 3 to go; any rushing attempt within 3 yards of the endzone. From this figure we see what I think anyone who watched us last year knew - we relied very heavily on Jerod Evans running the ball when we really needed it.

Figure 1 - Pie charts showing running attempt shares by player in 2016 for all situations and running downs.

A histogram of VT's rushing results and those of our players who eclipsed 65 attempts for the year is shown at the top of Figure 2. The lower graph in this plot shows the same data, but as a percentage to help directly compare individuals with differing rushing attempts. Note the triangles on edges of figures in this post represent summed counts or percentages outside of displayed range, circles on top represent players' season averages, dashed lines include all rushing attempts, and solid lines exclude touchdown runs.

Figure 2 - Histograms of rushing yardage from 2016 for team and players with >65 attempts.

From this we can see that Sam Rogers did better than everyone else at not losing yardage, but only gained 2 yards 21.5% of his attempts and was generally worse than everyone else at picking up large gains above 10 yds. This is fairly consistent with what you might expect from a fullback though, and we all know he did most of his damage as a receiver. Travon McMillian is another interesting case here. He seemed to have had most of his carries go for pickups of around 4 yards, while he had virtually no carries go for 12-20 yd pickups. He did, however have runs go for over 20 yds on 5% of his attempts. This generally confirms what I had thought, in that once he got to the second level he was typically fast enough to pick up some large chunks.

Having seen this data, let's move on to trying to show what kind of yardage you might expect a running back to get on an attempt. I am going to define a term I will be calling 'Minimum Achieved Yards', which is (1 - (number of attempts gaining at least X yds)/(rushing attempts)). This essentially will tell us what percentage of attempts gained at LEAST X yards. Another way to look at it is that the median of yards gained will be where 'Minimum Achieved Yards' crosses 50%.

Figure 3 - Minimum achieved yards from 2016 for team and players with >65 attempts for various situations.

Figure 3 shows the 'minimum achieved yards' for 2016 for all attempts (top), 1st and 2nd down only (bottom left) and for running downs (bottom right). From this we can see that while Sam Rogers was the most likely not to lose yardage on an attempt, Jerod Evans was the most likely to pick up at least 2-14 yards on an attempt. This is fairly consistent with what you would expect, as QBs typically have more blockers and better looks when they scramble. For picking up yardage greater than ~8 yds, McMillian and Evans were more likely to do so than Rogers, though big plays of 10 yds or more only occurred for McMillian and the team as a whole on about 10% of attempts.

With VT's 2016 rushing attack as a baseline, let's move on to 2017, where things really start to get interesting. Without such a dominant scrambler at QB, VT had to start relying much more on it's RBs. With a slew of young talent, none of who really became bellcow RBs, the VT coaches relied on more of a RB by committee approach. The result was having 5 players with more than 65 attempts on the year, broken down as shown in Figure 4. Here we see that once again Fuente relied heavily on his QB in the running game, though Jackson was featured far less outside of running downs than Evans was. Most interestingly, Deshawn McClease had 0 carries in what was considered a running down.

Figure 4 - Pie charts showing running attempt shares by player in 2017 for all situations and running downs.

Figure 5 - Histograms of rushing yardage from 2017 for team and players with >65 attempts.

Figure 5 shows the histograms of rushing attempts for 2017 as before. It is immediately apparent that fewer big plays of over 20 yds were seen by the team on the ground, going from 2.56% of plays of over 20 yds in 2016, to just 1.27% over 20 yds in 2017. This is due to both the loss of Jerod Evans, and McMillian getting much fewer big plays over 20 yds on the ground (4.9% of attempts in 2016 to just 0.96% in 2017).

Figure 6 - Minimum achieved yards from 2017 for team and players with >65 attempts for various situations.

Looking at the 'minimum achieved yards' for each player in 2017 shows McMillian and Peoples were very similar for the expected yardage gain for each of their attempts. This is somewhat surprising to me as Peoples was generally considered a larger power back. However, these two players have strikingly similar results profiles for every situation except for maybe 1st and 2nd down. Here we can see that peoples had no rushes larger than 13 yds while McMillian, having fewer big plays than in 2016, still had enough to be make a significant difference in avgerage yds/att when compared to Peoples.

I was a little bit surprised to see Jalen Holston having the fewest expected yards for both total attempts and on 1st and 2nd down. He was, however, at the top of the RB group in running situations, which seems to fit more closely with what I remembered of the year and him coming through on several very tough run situations. The fact that almost all of his attempts came at the end of the year against consistently competent opponents may also be skewing his results lower.

Finally, it is time to look at Deshawn McClease. I will admit he was certainly my favorite RB at the end of the year, and the data supports his case for performing very well. While his total attempts may be skewed by not having had any attempts during running situations, his performance compares very well to others even when only considering 1st and 2nd down. This is illustrated more clearly in Figure 7, which is the same data as shown in Figure 6 with the team-as-a-whole's results being used as a reference and subtracted from all the individual players 'minimum achieved yards'. The result is an easy way to see how each player did relative to the team's overall performance. From this we can see that McClease was more likely than any other RB to gain a given yardage for a range of 0-15 yards. In other words, McClease was more consistent in picking up yardage in this range than any other VT RB on 1st or 2nd down. Referring back to Figure 5 though, we see that while he was a consistent runner, he was not an explosive one, picking up gains of 20 yds or more on only 0.93% of his attempts.

Figure 7 - Difference of minimum achieved yards from team value during for various situations.

Having looked at the data from the rushing attack under Fuente, let us now look back at some select prior years for comparisons. I will start by progressing backwards and looking at 2015, 2011, 2010 and 2009 in isolation, before showing a comparison between each year's teams and key contributors.

Figure 8 - Pie charts showing running attempt shares by player in 2015 for all situations and running downs.

Figure 9 - Histograms of rushing yardage from 2015 for team and players with >65 attempts.

Figure 10 - Minimum achieved yards from 2015 for team and players with >65 attempts for various situations.

In 2015, Figures 8-10 show how Beamer & Loeffler in their last year relied much more on a single RB, McMillian, than we have seen in the two years under Fuente. Having McMillian on all 3 teams from 2015-2017 makes an interesting case study. While each of these teams were able to gain at least 5 yds on the ground ~30-31% of the time, McMillian went from gaining at least 5 yds 33.7% of the time in 2015 to gaining at least 5 yds on only 22.1% and 23.2% of his attempts in 2016 and 2017 respectively. In fact, he was able to do this while being less explosive in 2015 than he was in 2016, getting gains of over 20 yds on 3.5% of attempts in 2015 compared to 4.9% in 2016. This data just reaffirms assertions many have made that he seemed better suited and more consistent as a RB under Loeffler's system than he was under Fuente's.
At least to me, the degree to which this had been occuring had not been as pronounced when just looking at his drop in his yds/att under Fuente.

Figure 11 - Pie charts showing running attempt shares by player in 2011 for all situations and running downs.

Figure 12 - Histograms of rushing yardage from 2011 for team and players with >65 attempts.

Figure 13 - Minimum achieved yards from 2011 for team and players with >65 attempts for various situations.

Figures 11-13 show the rushing profiles for the top VT rushers in 2011. From these we see that Logan Thomas was heavily relied upon for rushing downs similar to our offense under Fuente, though on every other rushing down David Wilson was clearly the lead back. I am sure it comes as no surprise either that from these plots we can clearly see that Logan Thomas was rarely taken down for a loss when given the ball, though he was not able to pick up yardage of 5 yds or greater quite as well as Jerod Evans or Josh Jackson under Fuente's system.

Looking at these figures, we also see David Wilson's explosiveness really stand out. While he was not as good on rushing downs of picking up at least 2 yards as most other players, we see that he was getting gains of 20+ yds on 6.9% of his attempts, and gaining at least 10 yds on a whopping 18.2% of his attempts. Since 2011, the only player comparable to this has been QB Jerod Evans, who gained at least 10 yds on 19.4% of his attempts, but only had gains of 20+ yds on 3.0% of his attempts.

Figure 14 - Pie charts showing running attempt shares by player in 2010 for all situations and running downs.

Figure 15 - Histograms of rushing yardage from 2010 for team and players with >65 attempts.

Figure 16 - Minimum achieved yards from 2010 for team and players with >65 attempts for various situations.

On a 2010 team filled with running back talent, Figures 14-16 helped remind me what a truly dominant force on the ground Tyrod Taylor was that year. Here we can see that he was able to pick up at least 5 yds on nearly half (47.3%) of his attempts, 10 yds on an outstanding 27.6% of his attempts, and had gains of 20+ yds on 7.3% of his scrambles.

From the running back group, it is likely of little surprise that David Wilson again lead the group for picking up gains of 20+ yds on 5.4% of his attempts. However, when looking at it from a consistency standpoint, Darren Evans really was the most reliable back that year in all situations. While he unsurprisingly dominated the other backs in running situations, he also was more consistent in picking up gains in the 0-12 yd area when compared to the other running backs. Ryan Williams, while dominant the year before, spent significant time coming back from his injury in 2010, and it really shows in this data.

Figure 17 - Pie charts showing running attempt shares by player in 2009 for all situations and running downs.

Figure 18 - Histograms of rushing yardage from 2009 for team and players with >65 attempts.

Figure 19 - Minimum achieved yards from 2009 for team and players with >65 attempts for various situations.

The rushing profiles for the top rushers in 2009 are shown in Figures 17-19. In his healthy year, Ryan Williams easily led the team in attempts in every situation, had plays go for over 20 yards 6.2% of the time, and was able to pick up at least 5 yds on 28.1% of his attempts. It is interesting to note that this rushing profile looks remarkably similar to that of Travon McMillian in his freshman year in 2015.

Figure 20 - Histograms and minimum achieved yards from various years for VT as team and select players.

Figure 21 - Difference of minimum achieved yards from various years for VT as team and select players.

A direct comparison between select teams and individuals from 2008-2017 can be seen in Figure 20. To help differentiate between all these years, Figure 21 is provided showing this same data but with the 2017 VT Team used as a reference and subtracted from other years' teams and individuals. Looking at the data this way provides what I would say is some very interesting results. Noticeably, 2017's iteration of the team was better than every other year when in terms of both not having a loss of yards and picking up a minimum of 1 yd. They were in the mix as one of the better teams at not falling behind the chains and picking up at least 2-6 yds, though were worse than every other team at picking up any larger yardage on the ground. The 2008, 2011 and 2012 teams were some of the worst VT teams in consistently moving the ball for at least 4 yds on attempts, though the 2008-2011 teams had the most gains of at least 10 yds. The 2010 team was easily the most consistent at picking up yardage in the 4-20 yard range.

When looking at the the select players profiles, we can see that Deshawn McClease did very well at routinely picking up at least some yards on the ground, being near or at the top of this group for rushes of at least (-5)-8 yards. After this, however, he quickly drops to the bottom of the grouping. David Wilson was much the opposite, having been near the bottom at consistently picking up a few yards, while shooting to the top in his ability to consistently have gains of at least 5 - 20+ yds. His only real competitor for consistently picking up at least 8 - 20+yds was Darren Evans in 2010. This is actually fairly encouraging, as this trend shows that even larger 'power backs' like Evans are capable of routinely gaining large pickups on the ground, even after a fairly unexplosive freshman year in 2008.

RB Rotation Results

This section is going to be a lot shorter, I promise. But here I wanted to look at how a running back being rotated in or out could affect their rhythm and if there are any signs that it adversely affects their performance.

To analyze this, I gathered a list of all players going back to 2004 that have had more than 100 rushing attempts in their career at VT. I broke down their yardage gained per offensive possession they were involved in as well as the number of rushing attempts they had in that possession. Note that this means if a running back is used at the beginning of a drive, taken out, and put back in before the drive ends, this would still count as just one possession. Next, I removed any plays that occured within 5 yards from the endzone, as I did not want to skew the results due to a back running against a goalline defense.

Figure 22 - Plot showing relation between number of attempts per possession vs average yards per attempt.

Plotting an individual runner's yards/attempt for a single offensive possession versus the number of attempts they had in that possession produces a plot like those seen in Figure 22 for Ryan Williams. Note that the top plot shows all of the collected data, while I have removed any possessions where the runner had only 1 attempt in the bottom plot of this figure. From this data, I calculated the correlation coefficient for testing if these variables are related (R) and the null hypothesis (P).

The results of this calculation are shown in Table 1 for all the players I used. What is important here is this: a perfect correlation, meaning the variables are directly related will yield an R value of 1, while if the variables are perfectly inversely related, we will get an R of -1. More importantly, P is the Null Hypothesis and tests the same variables but assumes they are completely unrelated. The takeaway should thus be that a runner's yards/attempt for an offensive possession are only statistically related to their number of carries on that possesion if R > P (if you are a statistics major, feel free to correct me in the comments below). I have taken the liberty of highlighting in green the only cases where this holds true.

From Table 1 one can see that a runners yards/attempt and their attempts/possession are almost never related. There are a few cases where they appear to show significant correlation, though even in these cases, the correlation is very weak. It is also of note that Branden Ore, Kenny Lewis and Marshawn Williams all spent a significant part of their career as backups or change of pace runners, so feel free to interpret this as you wish.

Table 1 - Table showing correlation coefficients for relationship between attempts per possession and average yards per attempt.

In conclusion, it was found that the rushing attack under Fuente, especially in 2017, has been very good at consistently picking up short and medium gains on the ground. This holds true even when compared to VT's offenses from 2009-2011 when our rushing attack was headlined by a series of elite running back talents. The big difference it seems is VT's lack of explosiveness and inability to more consistently pick up large chunks of yards on the ground.

Deshawn McClease and Jalen Holston largely took over as the primary backs at the end of 2017 and the data shows that they, have been able to consistently pick up yardage when called upon, with McClease even having a higher chance of gaining short and medium yardage than even some of our past more celebrated running backs. It is concerning, however, that McMillian was our only running back who has shown a history of being able to pick up large chunk yardages but who is leaving for Colorado. As McClease and Holston are both still somewhat young, there is still hope that they can still become much more explosive, much as Darren Evans was able to when going from 2008 to 2010.

Finally, there does not appear to be strong case against against having a running back by committee as it affects a runners yards per carry. While the analysis showed a select few runners who seemed to get better with increased carries on a possession, all of these players have long graduated or retired from football.

DISCLAIMER: Forum topics may not have been written or edited by The Key Play staff.


Will comment more when I'm finished reading this in 2019 but for now, great job and amazing depths of research.

I just sit on my couch and b*tch. - HokieChemE2016


Haha, yea that was the initial thought. There was one point where I thought I'd show figures from every year since being in the ACC. Guess that's the danger of scripting everything - takes as long to plot everything as it does to plot just one thing. I dunno, when I was done I left the title because I thought it was funny

Great analysis. Putting Matlab to good use!

Very interesting that we were using Tyrod as a runner less than we are using our current QBs. Also great correlation results at the end of the post. Really puts a nail in the coffin on the RB by Committee debate for me.

From my understanding, the big difference with Tyrod and our current QBs is that Fuente will fairly often call designed QB runs, especially for picking up tough yards. Beamer's system I don't think had nearly as many called QB run plays, instead relying on the QB to scramble when things broke down. As such I think we likely left quite a few yards on the field so to speak, but it also meant that when he scrambled, Tyrod was usually doing so against better looks and took much fewer hard hits as he rarely ran up the center of a defense expecting the run.

Yeah I totally agree that Tyrod didn't get the number of designed runs.It just challenged my perception of those years versus my perception of the past couple.

Great job putting together the data to challenge so many assumptions, that is the power of analysis!

Tyrod took his fair share of hits. I just think we executed better down hill type running when you have backs like Darren Evans. We really haven't had that until Marshawn...man that hurt when he had to retire.

Amazing post. I really appreciate the effort and throughly enjoyed reading through all of that.

Warning: this post occasionally contains strong language (which may be unsuitable for children), unusual humor (which may be unsuitable for adults), and advanced mathematics (which may be unsuitable for liberal-arts majors)..

I applaud the amount of time and effort that went into this research. Thank you for putting it together and sharing.

Finally, there does not appear to be strong case against against having a running back by committee as it affects a runners yards per carry. While the analysis showed a select few runners who seemed to get better with increased carries on a possession, all of these players have long graduated or retired from football.

Warning: this post occasionally contains strong language (which may be unsuitable for children), unusual humor (which may be unsuitable for adults), and advanced mathematics (which may be unsuitable for liberal-arts majors)..

Can I make a request? Could you put together the Figure 2 and Figure 3 charts by RB for their entire career? I get that Y2Y changes on offense will be a factor as will the switch from Stinespring to O'Cain to Loeffler to Cornelson, but I would like to see how all the RBs compare. Most here hold Ryan Williams and David Wilson on a (deservedly) high pedestal and it would be nice to see if the fans expectations have changed or if RB performance has changed. I think from what you have show, it is the latter, but there may be many schematic reasons as much as talent reasons.

Also did you find the Pareto principle (that 80% of the yardage came from 20% of the ball carriers) applicable? If so you could reduce your scope to the RB/QB that fall into the 20% rather than a 65 carry criteria.

Come to Blacksburg and see what the Hokie Pokie is really all about