By Bill Lotter

How would you determine who had the best performance at the NFL Combine? If one player had a faster 40, but another was quicker in the shuttle, who would you say performed better? While it might seem like a subjective question, there’s really a principled way we can answer it. Ultimately, we’d need a measure of the importance of each event, then we can simply weight each event by its importance. Of course, each event might have a different importance per position, so we’d need to account for this as well. Last year, I built a model that estimates precisely this information. Using historical combine data and a surrogate for NFL success, I trained a model to predict how successful a player would be given combine data alone. While far from being perfect, the model had significant predictive power for all positions, except, interestingly enough, wide receivers. The parameters of the model, which tell us, for instance, how much better we’d expect a player to be if they were x much better in the 40, give us an unbiased estimate of the importance of each event by position. For reference, here are the results from last year:

Using these importance values, we can quantitatively say who had the best combine. Let’s start by looking at DE’s. Below is a chart illustrating the raw (unweighted) performance of 11 DE’s over the 8 combine measurements. The values are based with respect to historical averages: black is average, green is better than average, and red is worse. How much better or worse is dictated by the historical standard deviation per event. You can think of the standard deviation as a reference frame for the variability/scale in the data. If you decrease your weight by 1, it’s much different than decreasing your forty time by 1. So to normalize a measurement, we subtract the average and then divide by the standard deviation. (For the means and standard deviations per event and position, see the bottom of http://harvardsportsanalysis.org/2015/02/the-combine-actually-matters-part-2/). The chart below shows these normalized values for DE’s.

Notice that weight and 40-yard-dash time tend to be anti-correlated. Players that weigh more than average tend to have below average 40 times. Makes sense.

Next, we want to weight each measurement by its importance, i.e. take each value in the plot above and multiply it by its corresponding importance contained in the first chart. This is shown in the plot below. The weight and 40 columns remain relatively bright because they are the most important variables for predicting future performance of DE’s, whereas the bench is darkened out because it’s not important. In the leftmost column, the values are summed up for each player to get an overall score.

According to this approach, Dean Lowry of Northwestern had the best combine out of any DE. If we look at why, it’s because he weighs much more than the average DE, yet he still had a decent 40 time, bucking the common trend. In fact, he weighs about 30 pounds more than Joey Bosa, yet ran only one one-hundredth of a second slower, with a time of 4.87. Impressive stuff.

Below is the same plot for CB’s, ordered by total weighted score. Sean Davis and Jalen Ramsey consistently had good numbers in the measures that matter.

The combine generally has more predictive power for defensive players than offensive players. Here are the plots for the top 10% performers in the remaining defensive positions.

Next, let’s look at the offensive skill positions. The rankings for WR’s should especially be taken with a grain of salt since the combine isn’t predictive of their future success.

Last but not least, here are the plots for O-Line.

Finally, since the measurement I used to predict success (3-year approximate value, see previous posts for details) is designed to be normalized by position, we can ask who had the best combine across positions. The crown goes to… wait for it… QB Paxton Lynch. Lynch has exceptional size, yet he also had a great vertical leap and a decent shuttle, all of which are also correlated with success. On the defensive side of the ball, the top performer was DE Dean Lowry. As mentioned above, Lowry is a big dude, but he ran and jumped as well as guys who are much smaller.

While there is much more that determines how successful a prospect will be besides a few numbers measured in February, it’s interesting that we can still significantly predict future success through the combine alone. What is particularly useful is that we have a measurement of the importance of each event, which can vary greatly depending on position. Using this importance, we can say who had the best combine. While consensus favorites such as Jalen Ramsey and Keith Marshall top the list for their positions, less heralded performances by prospects such as Dean Lowry and Paxton Lynch are perhaps surprising. Far from guaranteeing all-star careers, such superb performances should nonetheless be appreciated and warrant further scouting consideration.

Ben Braunecker listed as the top tight end, surprise surprise Harvard sports blog!