Stat Review: Drops

Author: Bill Jones

Introduction

Hey football analytics fam, welcome back to my blog! Today, I want to share another one of my studies. I recently saw on twitter a scout visual showcasing the statistic drop percentage for quarterbacks, but I have always considered drops to be a garbage figure, so this caught my attention. Intrigued by this being shown and my preconceived notion being incorrect, I embarked on a quest to uncover the truth through using some basic data science techniques. Together, we'll peel back the layers of this statistical onion and uncover the hidden truths that lie within the realm of drops. Let's kick off this study and get into the data.

Ground Rules

Before we jump back into the analytics, I would like to let the readers know the ground rules I will be playing with. The data used for this analysis was obtained from Pro-Football-Reference for 2018 through 2022. Additionally, for any quarterback to be included in the analysis they must have had 200 passing attempts and receivers must have 50 catch chances.

Drop Rate

To start, let’s define drop percentage. Pro Football Reference (PFR) calculates drop percentage by dividing drops by targets, but I am going to use a different approach that I believe offers a more accurate assessment. Instead, I calculate drop percentage by dividing drops by catch chances, which is receptions plus drops. I believe this method provides a more accurate view of drop rates as uncatchable passes that are targets are not included in the denominator. 

Pro Football Reference Calculation – Drops / Targets = Drop % (Gabriel Davis 9/93 = 9.68%)

Billy Jones Calculation – Drops / Catch Chances = Drop % (Gabriel Davis 9/57 = 15.79%)

As you can see here in the Gabriel Davis example, there is a pretty big difference in the drop percentages for wide receivers that have a low target to reception conversion profiles. For the remainder of this post, I will be using the Billy Jones Calculation. 

Statistics Refresher

Regression is a statistical method that helps us to understand the relationship between two or more variables. In football, we might use regression to look at how one variable (such as completion percentage) is related to another variable (such as average depth of target) or we may use regression to determine how predictable (or “sticky”) a statistic is year over year. We will be doing the later today.

R-squared is a figure that comes out of a regression analysis that helps us to understand how well the predicting variable (prior year) fits the actual data (current year). R-squared value ranges from 0 to 1, with a value of 1 indicating a perfect fit and a value of 0 indicating no relationship at all. A perfect fit would be highly predictive while a 0 fit indicates the statistic is not predictive year over year. 

Visualizations and Analysis

With that now explained, let’s get into the analysis. To kick things off I’ll start with a year over year regression analysis. This tells us how much last year's statistics predict next year's production. We use a fancy metric called r-squared, which puts a figure to the amount of variation explained. Basically, the higher the r-squared, the better the predictive power last year’s statistic has on the next year’s results. 

Analysis: From a year over year predictiveness perspective, drops (or drop rate) for quarterbacks is not predictive at all while drops (or drop rate) for wide receivers is a bit more predictive yet pretty weak. The technical way to articulate this is… The proportion of the variance in the quarterback drop rate in N year (actual) that is explained by the quarterback drop rate from N-1 year (predicting variable) is 0.0094 whereas the proportion of the variance in the wide receiver drop rate in N year that is explained by the wide receiver drop rate from N-1 year is 0.0606… The less technical way to say this is… QB drops are random as fuck while there is a glimmer of hope that there could be some predictive power in wide receiver drop rates. 

With the regression analysis showing that drop rate lacks strong predictability compared to other key football statistics, an interesting question arises: Where exactly are these drop rates headed? To shed light on this, we turn to the mighty stochastic matrix. This powerful tool allows us to examine the movement of players across different performance buckets—elite, above average, below average, and poor—from one season to the next. By analyzing this matrix, we can uncover the patterns in the drop rate destinations.

Before I can use the stochastic matrix I first need to define the performance buckets and to do this I used the median and interquartile range as the cutoffs for the buckets. This resulted in elite = 0-25%, above avg. = 25%-median, below average = median-75%, poor = 75%-100%. 

Once the buckets are determined, the stochastic matrices can be prepared. In these matrices, the prior year is represented by each column and the following year is represented in each row. In these matrices each column must total to zero.

Analysis: At a macro level the elite and above avg. states are the most common to be transitioned into in the following year. This is cause by incredibly high elite->elite and below avg.->elite figure. The elite->elite phonema in my mind is a function of wanting to keep an elite pass catching core together. Once that group is assembled you hold onto them with dear life. The below avg.->elite phonema is a bit more interesting. I deduce this is a function of two things. First being some postive regression with drops being a highly random and unpredictable statistic. Second being teams seeing the issue and adressing it via free agency or the draft.

Analysis: The wide receiver results are intriguing with the poor state as the most likely outcome by a wide margin. I believe this could be a function of two factors that are causing this pattern. First, achieving an elite or above-average drop percentage as a wide receiver may involve a degree of randomness, making it highly prone to regression in the subsequent period. Drops are a low volume statistic (the NFL leader had 13) and are susceptible to a few random breaks going against a player that could skew the results. Second, it is conceivable that players with a higher drop rate face a greater risk of attrition. This could be due to teams' strategic decisions, whereby players who exhibit suspect hands may be retained based on their exceptional on-field abilities, despite their drop issues, yet those that do not possess the desired production profiles are gradually phased out of the offense in favor of more reliable pass catchers. This hypothesis suggests a potential trade-off between playmaking ability and dependable pass-catching reliability may often be acceptable.

To wrap things up let’s take a look at catch chances against drops on a scatterplot to visualize the drop leaders. Based upon the evidence shown above, there isn’t much substance to quarterback drop rates on a year over year perspective, so I am going to focus on wide receivers. 

Analysis: Not a ton of analysis needed here, just a reminder that the “Should Apply Extra Glove Glue” group is likely to stay poor in terms of drop percentage while the “Sure Hands Squad” is likely for some strong drop regression in 2023. Additionally, I presented the historical drop percentage for 2018 through 2021 compared against 2022 (min. 50 catch chances in both periods) to show who might have had a bit of an anomalous season. Mike Williams, Donovan Peoples-Jones, CeeDee Lamb, and Jerry Jeudy all stand out as players who outperformed their historical averages in 2022. Did these players become better in 2022 or are they in for some negative drop regression? Not sure but I will surely be following these four players drop totals in 2023.

Side note: Having a top of the league drop percentage hasn’t held back back Ja’Marr Chase.

Conclusion

In closing, I want to extend my appreciation to all who have made it this far. This blog was quite heavy, but I hope you found it informative. Stay tuned for more captivating analytics posts as I continue my journey into football analytics. If you have any research ideas or suggestions, I encourage you to connect with me on Twitter (@NFL_Billy_Jones). 

*This blog post was enabled by ChatGPT. The text was generated by me, and the content is my own, but some sentences and wording were provided by the model. I take full responsibility for all information produced in this blog. More information about OpenAI and their technology can be found at https://openai.com.*