Monday, June 4, 2007

The QB Rating Is Ridiculous

This blackboard makes much more sense than the NFL's QB rating.

I'm gonna start by listing a stat set for three QBs and let you guess which one has the highest QB rating. I'm then going to tell you which one was the highest before I allow you to tell me your guess, thereby making useless your actual guess. Got it? Here goes

QB A
27/35, 280 yards, 3Tds, 6 Ints

QB B
27/35, 250 yards, 0Tds, 1 Int

QB C
25/48 305 yards, 3Tds, 1 Int

Ok, now you guess which Quarterback has the highest QB rating. You ready? I'll give you another second. Well to start, it's not QB B. His rating is the second highest (84.2). If that was your pick, pick again out of A or C. If you picked C, you are wrong. Pick again (just kidding. If you picked again, put your head down on your desk and nap it up for a little bit. The rest of this post may give you a headache.). So QB A, who threw for 5 more Ints and less yards than QB C while throwing the same amount of TDs as C, has a higher "rating" (88.7) than C (84.1). Well that makes sense, right? No, actually it doesn't at all.

The QB rating is often cited by broadcasters, pundits and "those in the know" as the defining statistic for Quarterbacks. It is a mysterious formula conceived by some stat dorks that attempts to compare how well Quarterbacks performed on a particular afternoon as well as over the course of the season. It is an "all-encompassing" statistic (the exact explanation can be found here) that takes into account percentage of completions per attempt, average yards gained per attempt, percentage of touchdown passes per attempt and percentage of interceptions per attempt. In reality, the QB rating is essentially a modified efficiency rating (as opposed to the college game's "QB Efficiency Rating". A comparison can be found here) and if it were used in the same vein as the college rating (to compare efficiency of the passing game) as opposed to the NFL's described more universal purpose:
The NFL rates its passers for statistical purposes against a fixed performance standard based on statistical achievements of all qualified pro passers since 1960. The current system replaced one that rated passers in relation to their position in a total group based on various criteria.

The current system, which was adopted in 1973, removes inequities that existed in the former method and, at the same time, provides a means of comparing passing performances from one season to the next....
I would have no problem with it. Instead the QB rating has morphed into this uber statistic by which all QBs are measured. Well I'm here to tell you that while it is a useful statistic to measure a QB's efficiency year in and year out, and is helpful in comparing how efficient different QBs are, it has little use beyond that and has no use in determining if one QB "performed" better than another in a given game, season or career.

One of the main problems with the QB rating is that it is too heavily weighted towards completion percentage and does not weight heavily enough interceptions. To illustrate this concern and because my first example went so well, allow me to give another example to help flush out my point:

QB A - 25/40, 300 yards, 3 Tds, 5 Ints = QB rating of 70.8

Let's give QB B +10 more completions:
QB B - 35/40, 300 yards, 3 Tds, 5 Ints = QB rating of 83.3

Now instead of the +10 completions, let's give QB C all of QB A's stats but give him -2 Ints:
QB C - 25/40, 300 yards, 3 Tds, 3 Ints = QB rating of 79.2

Interesting. So the yards were the same, the touchdowns were the same but because passer completed 10 more passes, QB B's rating is still higher than QB C's despite the fact that QB B through two more interceptions. QB B was more efficient, but is it really fair to say that a QB who threw 5 interceptions had a better day than a guy who threw 3 if the only thing he did better that day was complete more passes? Give me the 3 Int guy any day. Especially considering that in the NFL the team that wins the turnover battle has about an 75% chance of winning the game. And shouldn't a universal QB rating be a statistical reflection of the positive impact a passer has on his team's ability to win? I would think so. A quarterback's interceptions are far more impactful on a team's ability to win than 5 or 10 extra completions and should be weighted accordingly.

Another more common complaint about the rating is that the maximum rating you can possibly have is arbitrarily 158.33. As a matter of convenience and statistical aesthetics, I can understand the problem with it but this is coming from a country that measures distances in terms of 5280 feet when there's a perfectly reasonable system available that is a little more pleasing to the Ol' calculator. Where I do see this being a problem is that the maximum QB rating can be achieved whether a guy goes 5/5 for 80 yards and 1 td or 40/40 for 500 yards and 5 tds. That seems a little odd to me. A good rating would take into account the percentages as well as the cumulation of statistics. I don't think it should be possible to obtain a perfect rating.

A final problem that I have with the QB rating is that it doesn't take into account how much better than the average a QB is playing. One of the measurement factors I would take into account would be the deviation from the mean for all QBs across the league. I don't think you can have a QB rating without using a system that has built in comparisons for what is going on in the current season or is at least reflective of the historical percentages. For instance, if in a given year most QBs are throwing 1 or 2 tds per game and one or two guys are throwing 3 a game, that difference should be reflected in whatever rating is created to compare QBs. This is important in terms of historical context. If the rules change and in the future QBs are throwing 50 Tds per year consistently, a year like Peyton Manning had in 2004 won't be properly understood unless there is some stat to show how far above average he was that year. The QB rating is capable of just that type of comparison if properly created (I am likely not the man to do it, though I may give it a shot in the coming weeks).

In conclusion, the biggest problem I have with the NFL's QB rating is that people use it improperly. If you want to use it to show how efficient a QB has been in a game and how that efficiency has helped a team win, that's fine. If you want to use it to justify why one QB is better or has performed better over a given period of time, it's almost useless by itself. You need context. And any stat that cannot stand by itself isn't really worth using.

4 comments:

Unknown said...

the long division problem on the chalkboard is wrong

K said...

This is really good news for Mike Vick, Rex Grossman and Andrew Walter.

Ben said...

... which is why this site exists: http://www.footballoutsiders.com/stats/qb.php

Luis DeLoureiro said...

Hi. It's interesting that you decided to write this post.

I recently posted something similar - and I've even created an alternative QB ranking system......based primarily on Y/A.

http://www.nflstatanalysis.net/2009/09/current-qb-rating-formula-sucks.html