With all the hoopla surrounding Josh Freeman and the back and forth of his season, as well as the attempts to place him historically against other quarterbacks, I figured a look at other quarterbacks in similar situations may help us get a better idea of where Freeman is as a player and where he may be headed. (To view all graphs, you can set your viewing to "Wide" as opposed to "Narrow" or you can click on links at bottom)
In Part I (located here if you didn't read it), we looked at a total of 17 NFL quarterbacks and their first year as a starter. We analyzed their stats and compared them to our young quarterback. This did not give us any predictive data, but gave us the comparison of Freeman to other NFL quarterbacks in Year 1. In Part 1.5 (located here if you didn't read) we looked specifically at how offense type or timing of their first start may affect each quarterback and what type of numbers they produce.
In Part II, which will be broken out into several articles due to the amount of data, we will look at our same selected quarterbacks (minus Freeman as we have no season two data on him) as well as historical comparisons and begin to look at the strides, if any, that these quarterbacks made in the next two years (Years 2 and 3) of their starting careers. We'll look at the same basic stats we perused in the last article to get a feeling for how each quarterback progressed.
Using their rookie years as a baseline, lets look at the next few years for these quarterbacks. So without further ado, lets jump into the next few seasons of data collected.
As usual, let me get some of the caveats out of the way. You've heard me go on about sample sizes, this not being an all-inclusive study, and how predictions are essentially a non-guaranteed thing. Guess what? All those same statements still apply. Through these articles we'll continue to sift through the data and see if any extrapolation for Freeman (or any other QB) can be made based on the components and criteria we are using. Specifically, we'll see if we can make any reasonable projections for Freeman in his second and third year of starting.
For illustrative purposes, I've broken down the quarterbacks into two separate classes; starters and career backups. Now, this is speculative on my part, but in order to not have cluttered graphs and tables all over the place, I felt this was necessary. In a sense, I cheated by looking at how each player's career has panned out but I wanted to look at those who have more of a history of starting and compare to those who have a history of holding a clipboard. By breaking our base set into two distinct categories, I'm hoping that certain characteristics will present themselves, allowing us to either project Freeman's performance or analyze his season 2 data and begin to get an idea of where he is headed.
Let me first start by stating who fell into which group. The starters are (in no particular order): Eli Manning, Donovan McNabb, Kerry Collins, Drew Bledsoe, Peyton Manning, Joey Harrington, Tom Brady and Ben Roethlisberger. The backups are (again, in no particular order): Alex Smith, Kyle Boller, David Carr, Gus Frerotte, Byron Leftwich, Brian Griese, Charlie Batch, and Daunte Culpepper. Using these two "classes", we can look at how each performed in years 2 and 3. For the record, I'm not lumping Freeman into either category at this point.
In this portion of Part II, we will look at the accuracy component of a quarterback. We can only do this using data on hand, meaning we are only looking at complete passes and incomplete passes, not degrees of incomplete passes like drops, throwaways etc. This also does not account for the length of the throw, so a completion (or incompletion) for a 5 yard pass or 50 yard pass counts just the same. By analyzing completion percentage for both specific quarterbacks and specific groups, we can determine if their is a trend from years 1 to 2 and 2 to 3 and how each quarterback fits into the puzzle when assessing their numbers.
I feel compelled to mention a few numbers here that weren't covered earlier. To determine baselines for each group, I took a look at the mean completion percentage for all quarterbacks in our study, as well as for each group. Additionally, if we all can jump back to high school or college math, you might remember a term you hoped to forget, standard deviations. Now, without jumping too far in and losing the 4 people who have made it this far, I just bring this up to remind you that 95% of all data can be expected to be found within two standard deviations in a normal distribution. We'll discuss this more in later (in simplified terms, I promise), but for now, lets look at the actual numbers.
I pulled every quarterback from 1970-2009 who played as a rookie and started 5 or more games (same as our criteria). Those quarterbacks (of which our group is included) featured an average completion percentage of 52.21% (standard deviation of 5.47%). Our "starters" had an average completion percentage of 54.21% (7.26%) while our "backups" had a completion percentage of 54.48% (4.71%). You can see that not only did our groups slightly outperform the NFL average from 1970-2009, but our "backups" did slightly better with less deviation than our "starters." Using a scatter plot, we can view where each quarterback fell in relation to the others.
That gives us a visual idea of where our selection stands when compared to every other rookie starter who started at least 5 games. Remember, this is just the baseline data for the rookie year with the trendline indicating the trend for all NFL quarterbacks we pulled.
We'll go ahead and look at years 2 and 3 in terms of completion percentage for the rest of our league as well as our own data points. Once we have looked at the mean completion percentages for each group (rookie, year 2 and year 3), we will start charting the movements from year to year and see if there is any way to project or categorize Freeman moving forward.
To just jump into right into the data, here are the average (mean) completion percentages for all NFL, our starters, and our backups: 55.2%, 58.43%, and 58.19%. Take note that the mean completion percentage for each group was higher in Year 2 than Year 1. You can also see that again, the quarterbacks I selected have outperformed the NFL mean. I'd also like to point out here that the mean NFL completion percentage over the last 10 years has been 59.36%. This means we can expect most quarterbacks to regress towards this number over periods of time.
Finally, Year 3 numbers for our groups (All NFL, starters, backups) are as follows: 56.8%, 57.15%, and 57.36%. These are much more closely grouped with very little deviation between them. To bludgeon you with some stats, the standard deviation (in order) of each group are 4.72%, 3.93%, and 5.42%. Again, our groups slightly outperform the NFL mean for Year 3 starters, but there is less room between each group.
Now that I've thrown a good bit of data at you on this front, lets get into what we came here for, a look into how each quarterback performed year over year in the early portion of their career. As a reminder, we have separated each quarterback into one of two groups (starters and backups) and for this article we are solely looking at completion percentage in years 1 through 3. I've taken our quarterbacks, pulled their stats from their first three years as a starter and looked at the change from Year 1 to Year 2, Year 2 to Year 3 and Year 1 to Year 3. With only 8 quarterbacks in each grouping, an unusually high or low number can skew the results. I've also included the NFL mean which allows us to look at a larger population. Here are the base numbers, which we will begin to analyze.
|Name||Category||Comp % (1)||Comp % (2)||Comp % (3)||Year 1 to 2||Year 2 to 3||Year 1 to 3|
|Sample Set Mean||-||54.35||58.31||57.26||3.90||-0.89||3.01|
|Back up Mean||-||54.49||58.19||57.36||3.70||-0.82||2.88|
The first thing that jumps out to me is how, when comparing completion percentage, the starters and backups for our study move right together, varying by minute amounts. The next trend that jumps out is the lowering of the completion percentage when going from Year 2 to Year 3. Drew Bledsoe and Alex Smith skew these numbers, but even with them removed, the average move is downwards for our sample groups and relatively flat for the NFL. (Note: To get the NFL averages, I took every player who fit our criteria in Year 1 and tracked him for 3 years, as long as he continued to meet our criteria. We ended up with 42 quarterbacks, who from 1970-2009, met the same criteria as those we selected based on playing time. The numbers will not add up when you look at the NFL averages and then the change column).
To visualize these numbers, lets put them in chart form, and look at the progression from Years 1 to 3 for our two groups.
Like we saw in the data chart, most of the quarterbacks saw an upswing from year 1 to year 2 and then a downturn or plateau from year 2 to 3. To refresh , the average completion percentage in the last 10 years has been 59.36%, which most of these quarterbacks are regressing towards. Players like Brady and Roethlisberger, who started out in the high 60's (percentage-wise), made a move back towards the mean, which is what we would expect. They were the only two in our starter category who saw a downturn from Year 1 to Year 3 overall which can be attributed to their higher starting point.
In order to further understand this concept, I've borrowed a snippet from Lookout Landing and a quick look they put together discussing regression (Note: This is a pretty good and simplistic piece for those who want to get their feet wet). Please note that this is a baseball piece, but regression to the mean is something that can be universally used. It simply means that while individual result may be exceedingly good or bad, we can expect most participants in any pool to fall back towards the average for the group.
'Regression' in this case carries no positive or negative connotation. It can push numbers up or drag them down.
Data cannot always be taken at face value. Since we cannot always be completely confident that we've measured what we want to measure, we can apply an expected regression to the mean to get a true idea of talent
Now, lets look at the same chart, only for the backups of our study. We'll plot the points in the same fashion as we did above and get the visual to accompany the numbers.
We see the upswing in years 1 to 2 as expected and then a similar move in years 2 to 3 as with our starters. Lastly, another representation of completion percentage across our groupings, this time with all three groups represented by their mean completion percentage.
Utilizing our sample set and the numbers for each player in our categories, we can see that there isn't much, if any difference between the "starters" and "backups" as we analyze their first three years of starting in the NFL. I think it's safe to say that based on the players we researched, that completion percentage is not something, at least from years 1 through 3, that determines your place in the league hierarchy. But lets keep in mind, this does not speak to all the quarterbacks out there. Each of our players had significant playing time and were given a chance to start once upon a time. The difference in completion percentage may vary as we get further into careers. It's also likely that as we are looking at players with significant playing time in years 1 through 3, their numbers will be close to the NFL mean, whereas those who truly fell off the map (like some of our players in later years) are what drags down the mean.
Even though we didn't see much variation between our groups doesn't mean we can't offer some predictions as to what we can expect from Freeman. When looking at Freeman's second and third seasons, we'll assume no injuries (at least 10+ starts in each year) and that he will fall within standard parameters. This means we can expect him to fall within two standard deviations of the mean. For completion percentage in year 2, this means we can expect him to fall between a completion percentage of 43.97% and 66.43%.
While we expect that 95% of the data will fall in between the numbers listed above, we can look at past quarterbacks in the NFL and see where their year 2 completion percentages fell. Once we have this graphed, we can then see the distribution pattern and where the bulk of the data falls. To illustrate the distribution of completion percentages for year 2 quarterbacks (including our sample set), see the graph below.
The two inside black lines indicate one standard deviation from the mean. For this grouping (Year 2 quarterbacks from 1970-2009 with at least 10 starts in their second year), the mean was 55.20% with one standard deviation being from 49.59% to 60.82%. This is significant because in assuming a normal distribution, we can expect approximately 68% of all data to fall within one standard deviation. As outlined above, two standard deviations spans 43.97% to 66.43%, which is where 95% of the data can be expected to fall.
Now, that seems fairly obvious that most quarterbacks will fall between a completion percentage of 43.97% and 66.43%. But rather than just assuming that, we now have some statistical data to go from. Freeman's completion percentage in year one was 54.50%. The mean increases from year 1 to 2 based on our numbers were 2.93% (NFL), 4.21% (starters) and 3.70% (backups). So while we know the range Freeman should fall into, we can also utilize the mean increase and predict that Freeman's completion percentage should increase somewhere around the 2.93% range and continue to regress towards the mean over the last 10 years of 59.36%. This is not a guarantee as Freeman may see a larger increase, no increase or decrease, or a large decrease. But based on past trends for starting quarterbacks witha certain number of starts and those quarterbacks we consider a fair comparison, the numbers indicate an approximate 3 percentage point jump in his completion percentage.
Now, lets run through the same drill with year 3 numbers.
Now that we are looking at Year 3 data, we can follow the same formula as we did for Year 2. Lets start with completion percentage again. We can kind of backdoor where Freeman will land in Year 3 by looking at Year 1 to 3. Using Year 2 to 3 data isn't possible, as Freeman has no Year 2 data yet. We know that in Year 3, the mean completion percentage for our criteria is 56.80%. We also know the standard deviation (One SD = 52.09% to 61.52%, two SD = 47.37% to 66.24%). We also know the mean increase from Year 1 to 3 as well as variations. This should allow us to guesstimate where Freeman will land in Year 3. The average increase from year 1 to 3 was approximately 3-3.5%, which targets Freeman right around 58% completion in his third season. Coincidentally, this is right around the long term league average of 59.36%. Of course, this is if Freeman follow the normal distribution patterns. Most quarterbacks, as our charts indicate above see a decent move from year 1 to 2 and then stay relatively flat from year 2 to 3. While I'd like Freeman to become a superstar in year 2, I would settle for the standard upward move of approximately 3.00%. Again, these is no guarantee that Freeman will follow these patterns, but by analyzing 50+ quarterbacks over 40 years who fit our criteria, our sample size has increased from the 16 other quarterbacks we analyzed. Luckily, our group and the NFL numbers are all relatively close.
Will Freeman end up improving his 54.5% completion percentage? The answer is simply we don't know at this point. Any number of things could sway his numbers, including injuries, drops, study habits, coaching, etc. There is no guarantee that if Freeman stays 100% healthy, studies 24 hours a day and every WR catches every pass that he will improve. There are some items that can't be quantified to date, like ability to read a defense and decision making. But assuming he follows the path of the 16 quarterbacks we reviewed and the mean completion percentage for those who played between 1970 and 2009, it seems we can expect Freeman to make an improvement in years 1 to 2, which is something we all would welcome.
Given that we are looking at stats and trying to project Freeman, as well as other quarterbacks, I figured we could use this as a learning experience for some new numbers. Now, I've used DVOA in the past, but for this analysis, we can't use it as every quarterback does not have available data for their first few years. We can, however, utilize pro-football-reference. They have a few proprietary stats that they use to evaluate players in the context of a given season. See the quote below, with my emphasis added
First, for each stat for each year for each league, we computed two things:
- the league average for that stat in that league during the three-year period with the given year in the middle. For example, the "league average" for the 1963 AFL would be the aggregate average of the stats accumulated in the AFL from 1962 to 1964. (NOTE: the 1960 AFL and the 1969 AFL, as well as the current season, will be based on only two years worth of data rather than three.)
- the standard deviation of the stat for all individuals who had 14 or more pass attempts per scheduled game during the three-year period.
Next, we computed how many standard deviations away from the league average each player was in each of his seasons. We multiply that number by 15 and add it to 100, and that is the number you see.
- On all stats, 100 is league average.
- On all stats (including sack percentage and interception percentage), a higher number means better than average
What this is saying is that the numbers we look at will have 100 as "average", anything under 100 (for any category) is worse than average, anything higher is better. Given that we are looking at completion percentage in this article, lets look at what they called Comp % +. We'll look at all of our quarterbacks and see how, when scaling against a league average based on numbers as opposed to opinion, each quarterback fared.
|Name||Category||Year 1||Year 2||Year 3|
Remember, these aren't percentiles. For example, in Year 2, Donovan McNabb (in regards to his completion percentage) was an average quarterback with a rating of 100. In Brian Griese's second year, he was pretty high, at 125. As a point of reference, anything in the 140 range is phenomenal and the league leader typically falls in the high 120's to low 130's. While we are looking at only completion percentage, it is pretty amazing to see Daunte Culpepper up there with the Peyton Manning's and Tom Brady's (and Brian Griese???) of the world.
This metric, along with the other "+" stats we'll look at will be able to give us a context as to where each player stands historically for each season. They are proprietary so they aren't something that ESPN or NFL.com will track, but in terms of comparing a quarterback to his peers for that season it will prove to be useful. We can't use these stats across years to compare quarterbacks (meaning a rating in 2009 can't be compared to 2001) simply because it's a sliding scale based on that particular year. We can, however, analyze how each quarterback fared compared to his peers in that season (i.e. rookie year) and look at how other rookies did their rookie year.
The chart above does not say that Alex Smith is better than Donovan McNabb in their rookie years, just that Alex Smith had better numbers (completion percentage only at this point) in his rookie year when comparing against the rest of the league that year than McNabb did compared to his rookie year and other quarterbacks that year. Clear as mud, right?
By looking at our classes of quarterbacks, as well as the broader strokes of NFL quarterbacks over a 40 year period, we have given ourselves a good bit of information and perhaps a clearer picture of how Freeman will progress in terms of completion percentage. Thus far, it doesn't look like we can distinguish which class he will fall into, based on the lack of variation between our two classes, but we can determine this; If Freeman will fall into the category of the players above (i.e. players who start 10 games or more), we can expect his completion percentage to rise on average approximately 3% in year 2 and another .5% in years 3. If it falls or stays flat, this increases the likelihood that he won't be starting 10+ games the next year, as quarterbacks who failed to meet our criteria did not start that number of games.
We've also now established some general parameters as what to expect from young quarterbacks overall. As we've seen from the data above, any one quarterback can have a particularly high or low number, but when looking at larger sample sizes, or by attempting to predict where a quarterback will fall, we now have a statistical means to do so, rather than just throwing darts. Based on year 1 numbers, Freeman was in the middle of the pack when looking at completion percentage. We now have an idea of what year 2 could hold for him.
In the upcoming portions of Part II, we will look at yards per attempt, passer rating, attempts per touchdown and attempts per interception to see if we can both project Freeman and see if we have any signs as to what puts a quarterback in elite company, or sends him to the bench with a clipboard. Be sure to leave any comments or feedback in the comments section below.
In case any of the graphs above were not viewable for you, here are direct links to photobucket to use. Once you hit the link, the picture should show up. For optimal viewing, click the "Zoom In" button on the top left of picture.