Each week this season, I will publish an analysis of the opposing QB for the upcoming Colts game. This will use a lot of charts and metrics to describe things like the QB’s style of play, how well he has performed and how he achieved that performance.
Normally, I publish these the day before the game and I keep the commentary to a minimum. However, I’m going to split this first one of the 2023 season into multiple parts, so that I can explain in detail how to read these charts and why the measures are important.
When I analyze a QB, the first thing I look at it is what I call my QB dashboard. This is a chart that shows a variety of different metrics and how the QB performed in those measures relative to other 31 team QBs(1). The first QB the Colts will face this season is the Jaguars’ Trevor Lawrence. The dashboard below summarizes all of his 2022 regular season games.
The x-axis contains 23 different metrics (use the mouse-over text above the graph to get a definition of each one). Beneath each metric label is his season total in that metric, but unless you are familiar with what is a “high” or “low” number for that measure, the values themselves can be pretty meaningless. Therefore, they are displayed on the y-axis as percentile scores(2), ranging from 0 - 100 and given a ranking (out of 32 QBs) to provide a sense of the level of performance.
Values above the mean are not necessarily “good”. Sack rate, for example, is better when the number is below the mean. Some numbers have no inherent good or bad values in isolation, but gain their meaning when compared to other numbers (e.g. time to throw).
From this dashboard, I can tell a lot about a QB’s season. For Lawrence, looking at the metrics from left to right I know that:
OPD - He faced harder than average opponent defenses. Opponent Passing Defense ranked 9th which means only 8 QBs faced tougher opponents, based on passing EPA given up by the opponent defenses over the season.
EDP - He led a relatively balanced offense. He passed on 51% of 1st and 2nd downs which gives an Early Down Passing rank of 16th.
ARSR - The Jags had an average run game to compliment his passing. Their Adjusted Rushing Success Rate ranked 18th, so 17 teams were better and 14 were worse.
PR% - He didn’t face much pressure. He was pressured on only 29.1% of his dropbacks, which is a lower Pressure Rate than what 24 other QBs experienced. Obviously less pressure is good for a QB, but this is a stat that takes on a lot more meaning when compared to some of the stats listed below (TTT, aDOT, AA%).
TTT - He got rid of the ball quickly. Pressure has 2 main drivers: O-Line performance and how long the QB holds the ball. Lawrence’s Time to Throw was the 3rd quickest of any QB and certainly was a key factor in his low pressure rate. By itself, this does not necessarily mean that the O-line didn’t also give good protection, but it certainly means Lawrence did not invite his own pressure, like other QBs do . . . Justin Fields, I’m looking at you.
To gain insight into the O-Line’s pass protection, I compare the relative rankings of Time to Throw and Pressure Rate. For example, last year, Lamar Jackson took a long time to throw the ball (2nd TTT), but he faced only average pressure (14th). That suggests that his protection was good, giving him a lot of “extra” time to pull the trigger. In other words, if the protection was just average the pressure would have been commensurate with TTT. This conclusion is backed up by ESPN’s Pass Block Win Rate which ranked the Raven’s O-Line 6th best. Conversely, Ryan Tannehill threw the ball quicker than average (19th TTT), but he was rushed to make those throws against the 5th most pressure (the Titans O-Line ranked 26th in Pass Block Win Rate).
Trevor Lawrence’s 25th ranked pressure is low but it is still relatively higher than his 30th ranked TTT, suggesting the low pressure was more about Lawrence’s quick throws than it was about O-Line protection.
ADOT - He thew shorter passes. Lawrence’s Average Depth of Target ranked 21st, meaning 20 other QBs had longer average pass attempts (air yards). Low ADOT is not necessarily a bad thing, but it gains meaning when compared to Time to Throw. Lawrence threw short and threw quick which lines up, but that isn’t always the case. Justin Herebert threw short (31st ADOT), but took a long time to do it (12th TTT). That is evidence of a QB that can’t find open receivers downfield.
AY/C - His completions were on shorter passes. It makes sense that shorter target depths lead to shorter completions, so a 20th ranked Air Yards per Completion is perfectly in line with his 21st ranked ADOT. However, sometimes QBs have a lot of long attempts but can only complete shorter passes, which isn’t good . . . Carson Wentz.
SCK% - He doesn’t take sacks. His 4.2% Sack Rate ranked 29th
SCR% - He’s not a “run-first” QB. His 3.8% Scramble Rate ranked 17th
TA% - He doesn’t throw the ball away much. His 3.4% Throw-Away Rate ranked 23rd.
AA% - He doesn’t give up on plays. The previous 3 stats are the different outcomes of when a QB abandons the pass (sack, scramble or throw-away). Add them together and you get an Abandon Attempt Rate, which for Trevor Lawrence was 11.4%, ranking 26th (7th lowest). Recall that Lawrence faced the 25th ranked pressure, so the 26th AA% is right in line with that.
QBs with a high AA% either have a pressure problem or a decision problem (or both). Looking at metrics like Pressure Rate can help explain AA% a bit more. Kyler Murray abandoned plays a lot (13th AA%) even though he wasn’t pressured much (29th PR%). That demonstrates a willingness to give up on the play too early. At the other end of the spectrum, Justin Herbert was pressured a lot (9th PR%), but didn’t give up on the pass (25th AA%). Remember his short passing depth? That tells me he threw a lot of check-downs, rather than take a sack, scramble or throw the ball away.
You can also gain insight by comparing the components of AA% directly. Lawrence deals with pressure using throw-aways as a last option, as opposed to someone like Tom Brady who used throw-aways as a first choice (3.7% ta, 2.9% sck, 0.4% scr). Ryan Tannehill’s 2.2% throw-away rate, 3.5% scramble rate and 8.9% sack rate tells me he’s not very mobile and he reacted poorly to pressure.
AAY - He uses his legs well. Lawrence ranked 8th in Average Abandoned Yards, which is sack yards + scramble yards per all abandoned attempts. This is a pretty good measure of how good a scrambler the QB is (low avg sack yardage and/or high avg scramble yards). With the rise in scrambling QBs, this metric will be critical.
You don’t want your QB bailing on the play until it’s necessary and when he does, you want him to maiximize yaradge. Ideally, you want a QB with a low AA% and a high AAY. Last year, Patrick Mahomes was 18th in AA% and 4th in AAY. Justin Fields had the 2nd highest Average Abandoned Yards, which is great, but he also had the league’s highest Abandoned Rate, which is very much not great.
It is likely that Anthony Richardson will have a high AAY number, but if that comes with a high AA%, it won’t be a good thing.
CMP% - Lawrence had a good completion rate, but so what. He had a 13th ranked CMP%, but that came on short passes (ADOT), which are easier to complete. CMP% in isolation doesn’t tell us much.
CPOE - He is very accurate. When you adjust CMP% for situation (down, distance, field position, etc.), you get Completion Percent Over Expected, which is a much better estimate of accuracy than CMP%. Lawrence finshed 9th best in this measure.
CPOE will likely be a problem for Anthony Richardson in his rookie year. Not a lot of stats carry over from the college game, but accuracy is one that does. That really doesn’t bode well for Richardson, so we’ll see if he can be “coached into accuracy”. I’m hopeful, but skeptical.
YAC - His receivers got about average yards after the catch. YAC is like CMP%, it really depends on passing depth (and QB accuracy), so in isolation it doesn’t mean much.
YACOE - His receivers get more than average YAC when accounting for situation. Lawrence’s receivers got the 10th most Yards After the Catch Over Expectation, which adjusts for variables like passing depth and field position.
Many people categorize this as a receiver stat and it is, but the data suggests that it is more related to the QB than the receiver (decision, timing, accuracy). If YACOE is low but accuracy (CPOE) is also low, then that may be a QB issue.
YPA - His average pass gets average yards. Yards per Attempt is simply that: pass yards divided by attempts and Lawrence ranked 17th in that measure. Total passing yards doesn’t tell you much about a QB’s performance, because leading teams stop passing to burn clock and trailing teams pass more to try and catch up. This causes passing yards to have very little relationship to wins or points scored. YPA solves that problem as it is an efficiency measure (average) not a volume measure.
YPA is an important number and looking at its its components will tell you how it was achieved and where potential weaknesses are. An alternate way to calculate YPA is:
YPA = (AY/C + YAC) * CMP%.
Lawrence was 20th in completion depth (AY/C), 14th in YAC and 13th in CMP%. Average those rankings and you get around 16th, which uncoincidentally is close to his 17th YPA rank. Raising YPA requires raising one or all of those numbers. We know his CMP% is good relative to passing depth (9th CPOE) and similarly, YAC is not a problem (10th YACOE), so it’s completion depth that is the weak link.
NY/d - His average dropback gains more yards than most other QBs. While his YPA is only 17th, adding in sacks, scrambles and throw-aways raises his overall yardage efficiency to an 8th best Net Yards per Dropback. We could have guessed that it would rank higher than YPA as we know that he doesn’t abandon attempts much (AA%), so sacks/scrambles/throw-aways aren’t going to dilute his YPA as much as it will for other QBs. We also know that he is a good scrambler and on those abandoned attempts, he gains more yardage than most others (AAY). So he is very efficient, not so much by being a great passer, but rather by making quick throws to avoid pressure and using his legs to maximizing yards on broken plays.
1st% - He gets a lot of first downs. First downs have more value than just the yardage gained to get them and 30.7% of Lawrence’s dropbacks result in first downs (7th best). This is a critical stat to a QB’s success. If you can’t get first downs on passing plays, then you will have trouble moving the ball down the field.
This will be a key stat for Anthony Richardson. If he can’t be generally accurate (most likely), he will have to make up for it by getting a lot of first downs.
TD% - He throws TDs at any average rate. Need I say throwing TDs is important? Well, it is. 3.9% of Lawrence’s dropbacks resulted in a TD (18th best). Not really his strong suit.
TO% - He doesn’t have a turnover problem. Only 2.2% of his dropbacks resulted in an INT or QB fumble, which is a little better than average (19th). This is also a stat to watch for Richardson. If he isn’t accurate, will that result in a lot of INTs? Probably.
EPA/d - His average dropback gains more value than most QBs. A 1 yard gain on 3rd & 1 has a much different value than if it were on 1st & 10. Expected Points Added captures that situational difference and is the fundamental stat to measure play value. EPA per Dropback tells you the average value a QB adds per designed passing play. This is the king of QB stats as it not only includes yardage efficiency, but also impacts from first downs, TDs and turnovers. It basically incorporates the impact of every stat in the dashboard.
The bottom line is that if you can’t manage good average value on pass plays then you are not a good QB. Trevor Lawrence added the 8th highest average EPA per dropback last year.
PSR - He is successful on more plays than most every other QB. Instead of measuring average EPA, you can measure the percentage of dropbacks that added positive value (EPA > 0). That is called Passing Success Rate and Lawrence was 4th best at it. You can think of PSR as the quantity of succesful dropbacks, whereas EPA/d is the quality (value) of dropbacks. So, Lawrence was about as good at consistently being successful (4th PSR) as he was at having high average success (8th EPA/d).
Since both metrics use EPA as their component measure, they are strongly correlated with each other (+0.84) and will plot closely to a clear trend line. I like to graph PSR and EPA/d against each other to identify the QBs that deviate from that line showing a dependence on either quantity or quality.
The upper-right quadrant contains QBs that have both a high EPA/d and high PSR and Trevor Lawrence is firmly in that group. He was easily a top 10 QB last year, especially when you consider he faced tough passing defenses. He’s pretty close to the trend line, showing that his value (EPA/d) comes from consistant success (PSR).
One QB who deviated a lot from the trend line is the former Raider Derek Carr. He had league average EPA/d but low PSR. This means that his succeses were infrequent, but of high value. In other words, he thrived on explosive, larger EPA plays. He had the 4th most attempts over 20 yards and the 9th largest average EPA on those passes.
That is usually not a sustainable model and Carr will most likely regress back to that trend line. So, either he will become a better overall QB leading to more success on his average play and thus a higher PSR (moving north on the chart) or the infrequent explosive plays he relied on willl dry up reducing his EPA/d (moving left on the chart). The latter is a far more likely an outcome than the former.
That’s a lot of infomation so I’ll end part 1 here. I know that a lot of these stats seem confusing and obscure to many, but they contain so much descriptive power that it’s really worth the effort to understand them.
Most of these stats are inter-dependent and don’t provide “answers” as much as they do context. As such, the individual stats are much less meaningful in isolation than how they relate to each other. Together they form a solid picture of how the QB plays the game.
(1) The data includes the 32 QBs with the most dropbacks. In rare cases, this could include the same team twice.
(2) Percentile scores are caluclated from z-scores converted to percentiles using a normal cumulative density function.