Evaluating InjuriesThe NFL is a tough game to play. Each week that a pro football player steps on the field is another series of damaging shots to the body. The question is, can we find a way to use the injury report information released by the NFL to devise a method for predicting how teams will perform with injured players?The first thing is to look at what data we have on hand. The official NFL injury report for each team tells us which players are hurting, to what degree they are unlikely to be able to play in the next game, and their position. The classifications used for the significance of an injury is as follows:
Our goal is to somehow quantify football injury report data and to do this we will begin with a controversial assumption: all injuries count the same regardless of the player’s position or starting status. It’s easy to think that the starting QB should be worth more than some backup, but given that the backups may play important roles on special teams and elsewhere, it may not be so. Teams can be better prepared to replace a QB than say a special teams leader. An old study we did looked at NFL Injury Report impact by position (and we will be publishing a new version of this soon with additional years of data), but the second benefit to treating everything the same (other than ease of use) is that we will be going against conventional wisdom. Everyone notices a star RB being out, few notice a minor player who may actually have a major role! With this assumption we can then translate the official NFL injury report for a team into a “player units lost” number along the lines of their own stated likelihood of playing:
Out - worth one point So a team with one player out, one player doubtful and one player probable would score 2.0 player units lost. A simple way to use these ratings would be to play a team in a matchup with a significantly lower “player units lost” number than its opponent on the theory that a healthier team could be a good value bet. Results playing team with lower "player units lost" reported A footnote on the above: the results reflect week 3 through the Super Bowl. The predictive value playing all games would appear to be low, although there is some promise for the higher delta games with three 60%+ years in the six year span! If we had happened to choose 4+ difference as the cutoff instead of 4+ it would look like an excellent handicapping indicator: 84-59 for 59% overall (14-8 in 2001, and 13-6 in 2002, 16-18 in '03, 17-9 in '04, 11-9 in '05 and 13-9 last season). The problem with simply using this week’s injury level is twofold. First, it doesn’t account for the fact that some teams are much quicker to list a player as injured than others, and at a much more severe level (one team’s doubtful is another team’s questionable). Second, and more importantly, what really would seem to matter is how a team’s injury level this week compares to the week or weeks before. A second rating to calculate could be the “change in player injury units” for this game from the last time the team played. A team with a 5.5 rating two weeks ago and a 3.5 rating this week has improved by 2.0 player units. Conversely a team with a 2.5 rating its last game and a 5.0 coming into this has seen an additional 2.5 player units hit their injury list. Example from the TwoMinuteWarning.com NFL Injury Report Analysis Oakland at Kansas City
This week Score: Oakland 4.5 , Kansas City 1.5 [Diff: 3 ]
Change Score: Oakland 1.5 , Kansas City-1 [ 2.5 ] Testing out this second rating scheme (we'll call it the "Change Score") over the same game sets as the "this week" rating produces the following results: Results playing team with lower "Change Score" This is a pretty impressive forecasting performance given all we are looking at is the injury report gradings (out, doubtful, etc) without regard for the player's position and role (starter, reserve)! It suggests that when there's a dramatic difference from this week to the last time the team played in its injury levels, you had better take notice! For a specific game that emphasizes this point, in week ten of 2002, Detroit played at Green Bay and the injury report showed the Lions had almost six full player units more on the report than the prior week (the equivalent of six full players completely out for the game!) while the Packers had 1.5 player units less reported, or the equivalent of one and a half players back to 100%. The final score on the day reflected Detroit being "undermanned" not just by talent but by injuries, as the Packers won in a rout 40-14. Still, there could appear to be some weaknesses in the change number alone as well, so our final tweak to the “injury capping method” is to use both the “player unit” number for this week and the “player unit” number change from this week to last. Let’s see how the teams fare when both these factors are put into play: Results playing team with 2 point lower "Change Score" and 2 point lower "This Week" score We selected the 2+ cutoffs before the tests were run, although raising the minimums required to a 3 player unit difference in both ratings winds up with a 39-20 (66%) record -- even stronger than the 2+ range, but with far fewer plays. The one concern is the results have been less dominating the last three seasons, and the number of plays is small. Suffice it to say though, this has proven to be a handicapping tool that's shown both historic predictive value and excellent performance in live usage over the past several seasons. While we will give samples of the 'injury report' column during the season, to have access to the predictions for every game of the NFL 2007 campaign you will need to Subscribe, which at only $99 for the full season if you sign up by July 31st has to be regarded as great value! After all, the injury report is just one of many proven handicapping methodologies we publish each week for subscribers.
Additional Injury Research articles:
Also see:
|