Model Validation: Tight ends from the 2013 draft class

To follow up on a previous post, let's look at some test data.  The training set (for model building) was all combine participants in the tight end position from 2005-2012.  As such, we'll look at the list of combine tight-end participants from 2013, and apply our logistic regression model for size adjusted speed and high point ability from the 2005-2012 data to separate players into two groups:  those with a higher predicted chance (>0.06) of pro-bowl success:

                Tyler Eifert (Bengals: 1^st Round)

                Vance McDonald (49ers: 2^nd)

                Travis Kelce (Chiefs: 3^rd)

                Dion Sims (Dolphins: 4^th)

                Nick Kasa (Raiders: 6^th)

                Chris Gragg (Bills: 7^th)

                Joseph Fauria (Lions:  Undrafted)

And those with a smaller probability of success (<0.06):

                Zach Ertz (Eagles: 2^nd Round)

                Gavin Escobar (Cowboys:  2^nd)

                Jordan Reed (Redskins: 3^rd)

                Levine Toilolo (Falcons: 4^th)

                Mychal Rivera (Raiders: 6^th)

                Justice Cunningham (Colts: 7^th)

                Jake Stoneburner (Packers:  Undrafted)

                Matt Furstenburg (Ravens:  Undrafted)

                MarQueis Gray (49ers:  Undrafted)

If we follow both groups of these players forward in time 3 years (to the present), we find 2 players (Tyler Eifert and Travis Kelce) from the top group were recently elected to a Pro-bowl, with Jordan Reed perhaps narrowly missing out on the honor (an 87-952-11 stat line from 14 games is certainly PB worthy).  With 7 players predicted to eventually make a pro-bowl, and 2 of them elected, we observe a true positive rate of 28.5% (=2/7) according to the derived algorithm, and a seemingly false positive rate of 72.5%.  Still, time may tell if more individuals from the either group make their mark as pro-bowlers.  Regardless, it does appear that the predictive algorithm based on size adjusted 40 time and high point potential did fairly well at predicting which players had a PB ceiling based on the data above.

You like that? Evaluating Kirk Cousins' 2015-2016 performance

After 4 long years, it seems that the Redskins may have finally found some stability at quarterback after mortgaging their future for the injury prone Robert Griffin III.  Interestingly, the rise of Cousins and Fall of Griffin might have been predicted by the 26-27-60 rule, though like any predictive rule, it has its limitations.

At first glance, Cousins had a good year, averaging a passable rating of 100.2 over the course of the 15 meaningful games that he started this season.  With a win over the Philadelphia Eagles in Week 16 locking down 1st in the NFC East, Cousins has earned some richly deserved rest in Week 17 which is the justification for leaving week 17 out of the calculations.  

However, with a standard deviation of 31 points in his passer rating, it appears that Cousins may have some room to improve in terms of consistency.  As a picture is worth 1000 words, we'll examine a time series plot of his passer rating by date:

It's pretty clear from the plot that there is a significant improvement in Cousin's passer rating over the course of the year (r=0.55, 95% CI: 0.05-0.83, p =0.034), though there appears to be a marked difference between his home and away performances (means ± SD of 115.4±30.1 vs. 83.1±22.6, p = 0.035).  This difference isn't alarming, as this trend is quite common in recent franchise QBs:

Peyton Manning: 101.6±26.9 vs. 94.9±26.6, p = 0.045
Aaron Rodgers:  107.3±23.4 vs. 99.0±26.1, p = 0.013
Drew Brees:  99.7±26.4 vs. 93.4±27.2, p = 0.084

Still, the late season improvement is comforting to the Redskins fan, particularly as it manifested itself in the W/L column at a time when the Redskins need to decide how much to shell out to retain Cousins for the 2016-2017 season.  However, the difference may be due to the fact that DeSean Jackson sat the first 7 games of the season (but that's an entirely different post).