Investigating NFL Combine Performance to NFL On-Field Performance

Introduction

The Combine is such a fascinating time in the NFL calendar year because it is media hyperbolic insanity. One of my favorite examples of this was in 2019 with the University of Buffalo QB Tyree Jackson. In some of the Combine workouts, he showed off his ridiculous arm strength that in fairness no one else came close to possessing. As Bleacher Report’s Matt Miller stated he’s “the draft’s most intriguing QB prospect” and labeled him as a boom or bust. Scouts compared his traits to Cam Newton, and Patrick Mahomes just to name a few (notable to add that he never produced at a high level in college like the other two). But alas, Tyree went undrafted and never stuck on an NFL roster. The last time he was taking snaps on a professional football field was in the XFL prior to Covid ruining that season.

Despite having traits similar to some of the most elite athletes we have ever seen as the scouts would have you believe, he was not an NFL-caliber player. Maybe the media just had a story and rode with it, maybe he fell in the draft farther than expected and teams just decided that he was not worth the risk, or maybe teams knew all along he was not going to be a quality NFL player. You can decide on that one, but it cannot be understated how his “athletic tools and one-of-a-kind traits” stemming from an impressive combine performance, drove up his draft stock.

Thinking about this whole media and scout fascination made me think, what if the NFL Combine is extremely overrated in evaluating prospects? Is there too much merit put into individual performances of athletic skills and frankly drills that do not really show off a true in-game football perspective of what the player is capable of. So, let’s do some investigating and look at the predictability of athletic testing drills and see if they can help determine and predict NFL performance. For this project we will try to answer the question: do athletic testing numbers for Cornerbacks matter in draft position, as well as future NFL performance? Since the cornerback position is considered to be the position in football that requires the “freakiest” of athletes, I decided to take a look at their combine numbers and see what I could find.

Data Collection

One of the most important aspects of understanding my data is familiarity with the Pro Football Focus (PFF) rating scale. PFF grades players on a scale from 0 to 100 with 60 being considered an average player. They watch every play from every game, and then at the end of the game, they give that player a game grade. And then at the end of the season, they will give out a year-end grade for the player. I would say that the PFF grades are the most objective way to rate a player’s performance that we have today. It is more useful than just interceptions or pass defenses as they fail to represent the entirety of the game.

When compiling the data, I knew I needed a fair amount of data, so I took all the cornerbacks drafted from 2015–2018 and used their combine performances and draft position to create the dataset I would use for this project. To get those numbers I combined a few datasets from the websites profootballreference.com and pff.com (pro football focus). Pro Football Reference was primarily used for the combine performances, as well as draft position. Pro Football Focus was primarily used for their grading scales as well as the snaps played by each cornerback. These websites provided me with all the necessary data I needed to work with my project idea. I molded the data to suit my liking by filtering the complete data set into smaller data sets that represented different groups I was taking a look at.

Descriptive Statistics

To start analyzing the data I had compiled, I began to calculate the means for all of the combine tests and measurables to get a read for the athletic ability for the average cornerback selected in the NFL draft from 2015–2018. The following list shows that data with V1 being the mean. We would expect an average cornerback to be just under 6 foot , 195 pounds, and run a 40-yard dash of 4.48 seconds.

Table 1

Means of Combine and Testing Numbers of Cornerbacks Drafted from 2015–2018

Looking slightly deeper into some of the combine numbers, I found some rather insane athleticism displayed by one of the cornerbacks that got drafted. From figures 1 and 2 shown below, there are two clear outliers that stick out like a sore thumb in the scatterplots. The first place point for both broad jump and vertical leap are both extremely far away from the second points. What is even crazier, is that both of those points belong to the same individual, Byron Jones of Uconn. In fact, Byron set the world record for the board jump at the combine with a jump of 147 inches.

Figure 1

Scatterplot of Broad Jumps by Inches

Figure 2

Scatterplot of Vertical Leaps by Inches

The NFL Draft is often considered to be a “crap-shoot” with no guarantees that drafted players will turn out to be good NFL players. From figure 3 below, that statement seems to make a lot of sense. Despite NFL scouting departments having a rather general idea on who they think will pan out or not, it still seems to be slightly random. There’s a slight negative correlation that can be shown by the scatterplot, but there is nothing real clear, showing that a lot of the draft is still just a giant guessing game. The best cover corner according to PFF grade was drafted as the 21st corner overall in his specific class.

Figure 3

Scatterplot of First 3 Years Coverage Grade and CB Draft Rank.

After I obtained all the descriptive statistics I would be taking a look at, I wanted to create two new datasets that represented cornerbacks that have been sufficiently better than their counterparts in the first three years of their career. So I filtered the data set to include only players that had a 70 PFF grade or higher on average throughout their first three seasons in the NFL in one dataset, and the players who had less than a 70 PFF grade in another. This way I could compare these two datasets to see if there were any significant physical differences between a player with a +70 PFF grade and a player with less than a 70 PFF grade. These new datasets will be talked about more in my inferential statistics section.

Inferential Statistics

To test out the hypothesis, I decided to create two backward selection models that include all combine measurables and tests regressed against the CB rank in the draft, and overall PFF coverage grade through their first three NFL seasons. When looking at the first model, backward selection took it from eleven original factors in the model, to only two. The only significant factors in predicting the cornerback rank in the draft are their broad jump distance and their last collegiate PFF coverage grade. The models will be shown below in tables 2 and 3.

Table 2

Full Regression Model for CB Rank in Draft

Table 3

Final Regression Model for CB Rank in Draft

For the final regression, our adjusted r-squared was 0.1779 which is rather low for a predictive model. We can interpret the broad jump and final collegiate PFF grade as slightly predictive but not a great model to base anything off of. The model makes it seem as if there are a lot of variables and error, that go into deciding when a cornerback is taken in the draft. Other than just the NFL combine.

Another way of modeling I tried for the cornerback rank in the draft was grouping the combine activities into three distinct groups. The three groups I made were jumping ( vertical and broad), measurables ( height, weight, arm length, and hand size), and speed/agility (40-time, 3-cone, and short shuttle). Using these models I tried to see if there was anything notable when grouped like this. Other than broad still being significant in the jumping model, the only predictor that became significant now was the 3 cone drill (see table 4 below). But yet again, the model had a very low adjusted r-squared value which shows us that it is rather poor at predicting the cornerback rank in the draft.

Table 4

Speed/Agility Model

Now let us take a look at the PFF Coverage Grade determined by the predictors and see if there is anything interesting to report there. Again, I used backward selection to find any significant predictors. Oddly enough, there was only one significant predictor in the model. It was bench press and it was actually negatively correlated with PFF Coverage Grade. However, it should be noted that the adjusted r-squared value is very low at 0.0503. This shows us that bench press is slightly significant but not great at modeling the data. Shown below will be both the full model and then the final backward selected model.

Table 5

Full Model Predicting PFF Coverage Grade Through First Three Seasons

Table 6

Final Backwards Selection Model Predicting PFF Coverage Grade Through First Three Seasons

Like I did for the first model with cornerback rank in the draft, I then grouped the combine activities into the same three groups I had previously. The grouped regressions turned out as expected, with no predictors being significant. This has us lean towards the idea that maybe the combine does not matter too much when it comes to predicting NFL performance. Before I concluded with that statement, I wanted to try out one more thing to see if I could find anything else that changed my view.

Back in my descriptive statistics section, I mentioned how I would be creating two new datasets that represented players with over a 70 PFF Grade and below a 70 PFF Grade. From these two new datasets, I decided to do some two-sample t-tests with each predictor, to see if there was anything significantly different between the “above average” cornerbacks and “average to below average” cornerbacks. When I did the t-tests, I actually found some very interesting information regarding 40 yard dash times.

The t-test for the 40-yard dash showed that there is a significant difference between the above-average cornerbacks and average to below-average cornerbacks. The above-average cornerbacks had a mean of 4.42 seconds for the 40-yard dash which is 7 seconds faster than the other group. From the table below, we can see that the confidence interval does not contain 0 as well as a p-value below .05. However, all the other t-tests I performed came out with insignificant differences between the means. So the 40-yard dash means were the only predictor means I could significantly say had a difference between an above-average cornerback and an average to a below-average one.

Table 7

Two -Sample T-Test for 40 Yard Dash

Conclusion

From the data collected, it leads us to a couple of conclusions. The first conclusion I came up with is that most athletic testing numbers in the NFL combine are not great predictors of where a cornerback will be taken in the NFL draft. Broad jump happened to be the only significant predictor in that case. A possible conclusion from that fact could be that NFL scouts and GMs tend to value broad jump data. They believe that the better the broad jump, the better that corner can possibly be. In the speed/agility model, I tried as well, it could be a possibility that the 3-cone time is important for scouts and GMs as well. Despite the fact that I will never be able to enter their draft rooms and understand their draft boards, the speculation is rather fun and interesting.

Regarding actual in-game performance, again it showed that none of the athletic testings from the combine was very predictive on cornerback performances through their first three years. From the t-tests, I can possibly conclude that a faster 40-yard dash is definitely preferred, but not the “be-all and end-all”. A lot of the athletic testing numbers seem to be overly hyped up in terms of actual on-field production. From the models I tested, it seems to me that the combine numbers do not mean as much as we tend to think they do.

One of the big limitations with the combine data is that a lot of players decide to not participate in some of the drills. Sometimes it may be because they are hurt, or maybe they just do not feel like doing it, but regardless it still dampens the full data that can be collected. I would like to go even farther back possibly, and look at cornerbacks from the entirety of the 2010s to possibly help circumvent this issue. The next logistical step would be take a look at all of the position groups.

Answering my question from the beginning, the combine data for cornerbacks is largely overhyped and overrated. Looking possibly to further investigate other position groups in relation to the combine, it could be largely media-manufactured as a way to keep their NFL fanbases interested during the offseason. Coaches and GMs seem to overlook some of the combine tests when drafting, and the athletic testing data does not seem to reflect too heavily in in-game performance as well.

--

--

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
3rd Quarter Analytics

3rd Quarter Analytics

Musings on Sports Analytics and whatever else