Because moorehead1976 graciously did a good stats-based Fanpost that generated some excellent discussion, I am not doing a stats analysis this week. Instead, I decided to discuss some meta stats issues I have been rolling around in my mind.
Long-time listeners, first-time callers to The Only Colors (and its predecessor, Spartans Weblog), know that KJ has been a long-time proponent of Tempo-Free Stats.
After my analysis of Drive Success Rate and PPD (raht ovah hyah), I believe that it is not only possible, but necessary, to create a version of Tempo-Free stats for football. This is because two of the statistics most commonly cited as indicator of goodness or badness -- yards per game and time of possession -- are very misleading.
Each drive is an independent grouping of independent tests. (H/T: Smart Football). From the offense's perspective, the more independent tests that are successful, the more likely that the grouping ends in success. The more groupings that end in success, the more likely the overall result is a success. Conversely, the defense's goal is to avoid and/or prevent the opponent's offensive success.
Basically, the logic is this: the more drives (possessions) a team has, the more likely it is to gain more yards and score more points, all things being equal. Just like in basketball [except basketball doesn't have a "shot distance" stat fetish]. So we must look at overall indicators of success on a per-possession basis.
First of all, we look at the rate at which a team's drives end in an offensive score. I was calling this Drive Success Rate, or DSR. I have since discovered that Football Outsiders already has a stat called DSR that they apply to the NFL. FO's DSR is different, though -- rather than measuring the overall success of the drive, it "measures the percentage of down series that result in a first down or touchdown." In other words, FO's DSR is a measure of sub-groupings rather than groupings, and more importantly, excludes Field Goals as a successful outcome. So I need a new stat name with acronym for this concept.
Secondly, we look at Points per Drive, or PPD. Points per drive is indicative of two things: first, the success rate; and secondly, whether the team is scoring touchdowns or resorting to kicking field goals. I am planning on deliberately leaving extra points / 2 point conversion in the points per drive stat, for these reasons:
- Extra points kicked are near automatic, so if a team scores six, it's likely that they will get seven.
- If a team decides to go for the 2-point conversion, the extra two points are attributable to the offense, and therefore the offense should get "credit". Likewise if the conversion fails, the offense should get the "blame" and the defense should get the "credit".
Third, as an indicator of offensive strategy, I always look at Run-Pass Balance. I always express it in terms of percentages with the percentage of run plays given first. We have to adjust "sacks" from being run plays to being pass plays.
Now, here's where I need help and input from y'all. I really would like to narrow down to the "four factors", and make a "quad chart". But I am not quite sure what the four factors should be.
Likely possibilities are:
- Yards per play. Normalized? Net (offense gained minus defense allowed)? Scatterplot offense vs defense?
- Yards per drive. This may be a good indicator of how effective a team is across many independent groupings. I guess the question that sticks in my mind is: yards per drive can be skewed lower based on starting field position. For example, a team that always starts drives at their own 35 yard line (due to the skill of their special teams) would have, on average, 15 less yards per successful drive than a team that always takes the touchback and starts at their own 20. This is because they can't go any farther than 65 yards to be successful [unless, of course, they shoot themselves in the foot]. So, I am thinking about yards per drive as a percentage of yards needed to score a touchdown. In other words, a team that starts at their own 35 and scores a touchdown would have 65 yards gained divided by 65 yards to end zone = 100%. Likewise, a team that starts at their own 35 and only gets to the opponent's 20 before kicking a field goal would have 45 yards gained divided by 65 yards to go = 69%.
- Turnovers. I believe we need to account for turnovers. Turnovers are a relatively rare occurrence -- the turnover leader in 2009, Miami of Ohio, had 36 turnovers out of 886 total offensive plays, or 4% -- but the results are disastrous for drive success when they do occur. [At risk of sounding glib, turnovers are like plane crashes. They rarely happen, but when they do, the results are catastrophic]. However, I am not of the belief that teams have the ability to increase the opponent's turnover rate. I am inclined to scatterplot turnovers lost by the offense vs gained by the defense rather than turnover margin.
- Other things to consider: Breaking down the offense into yards gained vs expected, both rushing and passing, in order to account for the opponent's defensive quality. FO dies this with "DVOA", or Defense-Adjusted Value Over Average. But DVOA requires reviewing each play and assigning "success points", which is a lot of work by itself and is made infinitely more complicated with 120 teams vice 32 teams in the NFL. [The explanation of Success Points is here]. So we need a simpler way of determining expected yards gained that can be calculated from the stats. Penalties can have a significant impact on offensive and defensive success rates. I think we should disregard special teams penalties that only affect field position since the corresponding field position change will be reflected in the yards per drive. Defensive and Special Teams Scores, like turnovers, are rare but catastrophic events, but I don't know that we should put a lot of effort into tracking them. [I know that "kickoff return for a touchdown" would not be a key part of my game strategy if I was a coach].
I look forward to everyone's thoughts and suggestions. If anyone has a desire to contact me privately, please email KJ, LVS, or Pete.
Thanks in advance!