Cal in the NFL Draft, Evans Hall Football edition: What is a draft pick worth?

There are many formulas for calculating the draft pick value. The most famous chart was developed by Jimmy Johnson when he was the HC for the Dallas Cowboys. Another attempt at calculating the value of each draft pick was conducted by the "The Harvard Sports Analysis Collective". The HSAC used standardized Career Approximate Value overtime and conducted a simple regression on prick value to calculate the pick value.

In this article I will conduct a more indepth analysis of 8,411 draft picks from 1970 to 2015 from pick #1 to pick #224. All of the players were chosen from pro-football-reference's database of players picks and values. In my case I decided to use R and STATA. The R code was used to clean-up the data set from the initial saved output from the website and then the data was processed by STATA for the process of statistical analysis.

tl;dr My chart emphasizes the value of the midlevel picks more than the JJ or HSAC. This falls in-line with my idea that it is important to build the core of the team through the draft, surely you can find blue-chip talent in the draft. However, when we evaluate the picks in the draft, the value of a few solid contributors from the middle of the draft can make a huge difference in the fortunes of a team. It may not be a 1:1 difference, but I would argue that outside of a handful of truly great players, few are worth more than 3-4 good players.

Methodology

Dependent Variables

I will first start with the process that I was thinking about: what makes a successful draft pick? Of course we have advanced data such as AV for per player long-term performance. However, I think that there are more dimensions to the analysis that can point towards a successful career. So from the available data from PFR, besides the Career AV (CarAV), I picked the following:

• Pro Bowl Selection (PB)
• All-Pro Selection (AP)
• Years Played (YP)
• Games Started (GS)

The first two are imperfect data points when measuring the performance of the player. We often have players getting those accolades without merit (ex. Jeff Saturday in his last year getting the Pro Bowl nod) or being snubbed completely (ex. Lavonte David last year in the All Pro, or Delanie Walker). However, in general it should give us a glimpse at performance levels.

Years played points more towards the longevity of a player, especially in the mid to late rounds, can mean a lot for a team to have a player that lasts a long time even if they are not exceptional players. Likewise with game started, having a player that is a starter but not PB or AP level, can give a team mid-level stability for a team.

However, I wanted to normalize the relationship between all variables to see the relationship better. Value of 100 for each of the variables indicates the average for the variable. Here are the averages for each dependent variable for reference:

Independent Variables

Thus our dependent variables are such : CarAV, PB, AP, UP, and GS.

Additionally, we will have these following independent variable : Log of pick (which then will be translated into a relationship between the value of a pick and a pick).

The reason I added the is to adjust for the curvature of the data that is constrained by the pick value as well as the minimum value of each dependent variable. See the following charts to see why:

Piotr

Piotr

Finally, I will use the QB as the benchmark for the data controlled by the different positions. This means that the coefficients from the regression on the positions will indicate the adjustment we will need to take on a per position level.

Model Outcomes

Simple Model with Pick Value

First I started with a simple model that tries to predict the value of the pick for each of the dependent variable using only the natural log of the pick.

Piotr

What we can see that the Log of each pick is statistically significant and negative. This is a linear-log model where the relationship between the dependent and independent variable is... complicated and not relevant to the conclusion. Which is the following table:

We can see here, from this very long table, that teach of the indicators show a fluctuating set of values.

The value of the 1st overall pick:

• in my table is two 22nd (15th if  using the Pro Bowl measure) overall picks or three 63rd (37th if using the Pro Bowl measure) overall picks, or four 107th overall picks
• in JJ's table is two 7th overall, or three 16th overall, or four 25th overall,
• and in HSAC's table two 20th overall, or three 39th overall, or four 69th overall.

Conclusions From the Model

Class Dismissed