Every year around draft time, tidbits like this one start floating around twitter and NFL scouting blogs:
There has never been a QB who threw less than 54 mph w/ any long term success in NFL. http://t.co/v0iw6Q9XoK— Benjamin Allbright (@AllbrightNFL) March 2, 2014
That seems pretty surprising right? It supposedly gives us a nice, easy cutoff that we can use to cull the crop of draftable quarterbacks. And at a very basic level, that tidbit is true. People will use this piece of information to completely cross quarterbacks off their draft charts, even quarterbacks they may have otherwise graded in the first two or three rounds. This was brought up in the comment section yesterday in regards to Brett Hundley, and we even had a short article here at DBN last year about this very subject. I think all of this attention being paid to ball velocity is misguided for a number of reasons. Allow me to explain why.
First, we have an issue of sample sizes. We only have velocity numbers going back to 2008, which covers 90 quarterbacks. That seems like a lot of data, but we have to consider the result we're looking for: long term success. Out of those 90, how many have had "long term success" in the NFL? Here's how Tim Miller put it last year:
If they are basing this solely on QB’s that have thrown at the combine since 2008, then the statement could also read: "There has never been a QB not named Joe Flacco w/ any long term success in NFL."
Now, I'm going to expand that group a little just for the sake of conversation, but I think the point is a good one. For the purposes of this article, I am going to consider these players in the group of quarterbacks who threw at the combine and have established themselves as NFL starters: Joe Flacco, Russell Wilson, Cam Newton, and Andy Dalton. Now, already, this should be raising some red flags. We're looking at four guys out of 90. Trying to draw any correlation between four data points is an uphill battle at best and a complete waste of time at worst. But that correlation has been drawn, so let's roll with it. One way to test whether we've actually detected a signal in this noise is to compare our selection of established quarterbacks to what we might expect if we drew quarterbacks at random from the whole sample. Out of our 90 quarterbacks, 50 had velocities over 54 mph. That's about 57% of the total sample. So, to see the probability of drawing four quarterbacks at random with over 54 mph, we simply take 0.57^4, which comes out to 10.5%. Hey! that actually seems pretty small. It seems our velocity parameter may not be random after all.
But let's look at that sample of 90 QBs a little more closely. Over a third of those guys weren't even drafted, and it includes guys like Marquise Gray. These are players that no one ever expected to succeed, for reasons well beyond their ball velocity. So let's throw them out. Now we're down to 54 quarterbacks, 32 of whom threw 54 mph or harder. The probability of our four QBs being randomly drawn from this sample is about 12%, so it's still pretty low. But again, a lot of these guys were never expected to succeed, and most of them never even got a shot to start for more than a game or two. To really get a good idea of what's going on here, let's look at players drafted in the first three rounds. These are players that were expected to contribute and many of them end up getting real playing time. There were 22 quarterbacks who threw at the combine at were drafted in rounds 1-3. Out of those 22, 17 threw 54 mph or harder. The probability of picking four hard throwers at random from this sample is about 35%. There's over a one in three chance that this "rule" is nothing more than random chance. So, sure, if you're looking at a guy who might get a sixth round grade and you see he doesn't throw hard, maybe you drop him off your board. But if you think he's worth a first or a second? Don't sweat it. And remember, these numbers I've just gone through don't even account for any factors outside of ball velocity. Did Pat White fail because his ball velocity was 52 mph, or because he was small, inaccurate, and couldn't read a defense?
All of that math we did up there rests on two assumptions: that our sample is representative, and that our measurements are accurate. If those two things aren't true, then it's even less likely that this "trend" is real. Well guess what? I don't think either of those assumptions hold up to scrutiny. Let's look at the first one. We want to know what makes a successful quarterback, and the numbers we have are from the combine. But not all successful quarterbacks threw at the combine. Here is a list of QBs who chose not to throw at the combine, and thus never got a velocity reading: Matt Ryan, Matt Stafford, Andrew Luck, and Ryan Tannehill. These are successful QBs for whom we have no data. If all four of these quarterbacks also threw over 54 mph, then all of the sudden the probability of this "rule" being a result of random chance drop to 16% at best. However, it only takes one of them throwing below 54 mph to disprove the rule. Would it surprise anyone if Ryan Tannehill had a weaker arm than Landry Jones, who threw 53 mph? If you wanted to know what makes a pitcher successful, you wouldn't ignore four of the last eight Cy Young winners, but that's essentially what we're doing here.
Finally, let's move on to the accuracy of the measurement. Internal error of radar or lidar guns can be up to two or three miles per hour, depending on the model of the gun and the angle at which the reading is taken. But rather than get into the nitty-gritty of how radar guns work, let's just look at some readings and see if they make sense. I mentioned Landry Jones above, and he makes a pretty good case study. He was widely regarded as having one of the best arms in the 2013 draft. He could sling it all over the field, the problem was his accuracy and decision making. The radar gun clocked him at 53 mph, below our threshold. Joe Flacco, who is widely regarded as having one of the strongest arms in football, threw 55 mph. That ties him for 35th among quarterbacks who threw at the combine. Some names of QBs who reportedly threw as hard or harder than Joe Flacco: Andy Dalton, Casey Keenum, Curtis Painter, Colt McCoy, Austin Davis. That's right, if we're to believe these measurements, Austin Davis has one of the ten strongest arms ever measured at the combine. Conversely, here's a list of QBs who supposedly have a weaker arm than Colt McCoy: Joe Flacco, Geno Smith, Russell Wilson, Jake Locker, Landry Jones, and EJ Manuel. Now, as Browns fans, that should ring some alarm bells. Clearly, either something is wrong with the measuring device, or what we're measuring doesn't translate from the combine to the playing field. Long time readers will know I'm not usually one to advocate the "eye test," but I think we can all agree that Joe Flacco throws harder than Colt McCoy.
Each of these reasons for skepticism might not be enough to change your mind on its own, but taken together I hope these three factors make you hesitate the next time you want to cross a quarterback off of your draft board due to ball velocity. I think we could pull together similar rules about pretty much any combine event, if we really wanted to, simply because we're only looking at a small handful of data points. To be clear, I'm by no means suggesting that arm strength isn't important. I'm not even arguing against the existence of some threshold arm strength necessary for NFL success. What I'm arguing against here is the specific number and method of measurement that seems to have permeated the NFL scouting community. I love the way people are increasingly turning to real data to inform their football opinions. It's my favorite subject to write about and it drives much of my interest in the sport. But we have to be careful about how we apply this data, and make sure we know exactly what it's telling us. In this instance, I don't think enough thought was put into the result before jumping to a conclusion.
Author's Note: As Benjamin Allbright has pointed out on twitter, the Colt McCoy point was taken at a pro day using both a radar gun and a micro-chip in the ball. I initially left this out because it doesn't alter the conclusions made here, but it does warrant inclusion.