Total Quarterback Rating (Total QBR) is a measure for QB performance created by ESPN in 2011. It was intended to overcome obivous drawbacks of passer rating, which is purely based on passing stats. Total QBR evaluates each play and assigns a value according to many factors (like outcome of the play) including subjective ones, such as how clutch the throw was, how much pressure the passer was under, etc.
Passer rating is much simpler, it only measures the ‘hard stats’: passing yards, TDs, etc. Which rating system is more representative of QB performance is debatable. While total QBR has received a lot of critics (partly because ESPN has not released the model), it is arguably a more valuable measure of the value of a QB (after all, a WR who catches a ball travelling 5 yards in the air and runs 80 yards for a TD should receive more credit than the QB).
So the question is how are these two compared with each, and what might be the takeaway from this comparison.
Scrape data from ESPN
PLAYER columns have different values for same player
The data on ESPN website is different from stat.nba.com or Car2go, which I worked on previously. They are stored on the server and can be extracted rather easily using readHTMLTable(). Note player name values are different in each table. Therefore I made a short string and merge() the two tables into one with by = ’playerShort’.
Total QBR vs passer rating
The data is rbinded to include years from 2006 -2015. Below is a gif loop over the comparison plot for each year. The overall trend is expected: total QBR and passer rating are positively correlated. I plotted the linear regression and 95% confidence interval as well. If we make the prediction of QBR from passer rating based on this curve, this is not too far off. Of course ESPN has more advance stats as input data to find a better regression (basically the QBR model), but passer rating is also robust and much simpler. It is essential for a model to be easy to understand, even sometimes, this means a little accuracy is sacraficed (again, this is debatable here). Without realeasing the actual model makes QBR hard to explain to someone, not to mention the subjective metrics: how clutch is clutch…
The underrated and overrated
Never the less, let’s say the ‘clutchness’ of the world really makes a positive impact to player evaluation and since passer rating fails to capture it, we can make a judgement whether a player is under- or overrated. For example: Ryan Tannehill’s passer rating this season is 88.3, which is about average. However, his QBR is only 34 ( voice of Skip Bayless: over the scale of 0 to 100), which is pretty bad. As a result, he is overrated as an average passer. On the flip side, Andrew Luck’s passer rating in his rookie season was 76.5, about 10 points below average, but his QBR is 67.4, way above average. So in his rookie season, Andrew Luck was underrated as a below-average passer. In fact, he was very clutch.
Next, I plot the top 15 most under- and overrated passers in the past decade. The criterion is the residuals() of the aforementioned regression in each year (QBR may only be compared in single season).
This is a very interesting plot, especially when you are very familiar with NFL QB stats and performance in the last decade.
For me, I only became a fan since recent years, so bear with me. 1: As good as we think Peyton Manning was from 06-10, he’s not getting enough credit 2: Ryan Tannehill is really bad this season, like much-worse-than-we-thought bad. 3: Kevin Kolb was even worse back in 2011 in Arizona. 4: Philip Rivers posted best passer rating in 2008 with 105.5, but he was not as good as many other QBs that year in terms of value added to the team (Charges won AFC west with 8-8 record.) 5: Jay Cutler appeared twice on the underrated passer list, the latest was 2013 season, after which he signed a seven-year deal with the Bears 6: Ryan Fitzpatrick is clutch this year! Enough said.
Over the last decade, the average passer rating has increased from 80 to over 90. No doubt NFL is becoming a passing friendly league (We can look at other metrics like increasing QB salary/cap, and the flip side of RB, penalties called to protect QB, average play per game and etc., but this is another article ). Average QBR shoule be 50 (per definition?), but here I am only presenting the QBR averaged over player (I guess more accurately, I should average over per play, but I don’t have the data).
The entire code is published here. I partly used Mr. Todd W. Schneider’s theme for my plots.
To leave a comment for the author, please follow the link and comment on their blog: Jun Ma - Data Blog.