Nice work. I referenced drive efficiency in an earlier post, but this is more useful.
Tempo-Free Defense Points Per Posession Update: Includes Offense PPP as well
Last week we saw Points Per Posession for the offense as a tempo free metric to see how good our O is. With that in mind I wanted to look at PPP for our defense. This is a little tough as the NCAA doesn't put it all together so you have to go back to each drive and pull in the drive numbers. So I went ahead and did that for 2010 so far and got this...
|1||-9||0||Punt||71||7||Rush TD||53||3||FG Good|
|5||48||3||FG Made||19||0||Punt||79||7||Rush TD|
|8||49||0||TO on Downs||24||0||Punt||24||0||Punt|
|9||42||0||TO on Downs||77||0||EOH||70||7||Rush TD|
|11||66||3||FG Made||26||7||Pass TD|
|4||64||7||Rush TD||99||7||Rush TD|
|9||69||7||Rush TD||33||0||TO on Downs|
|10||8||0||Punt||50||0||TO on Downs|
Note that the items in italics were not counted as I decided they shouldn't be counted - a couple other EOH drives were counted because, at least to me, it seemed obvious the other team was definitely trying to score.
All that data chrunched in this:
Ok, so thus far our D is giving up just over 2 points/posession - hmm (and OMG, Thank God IU wasn't as efficient as UMass). That doesn't sound that good - rather than compare it to tOSU or MSU I thought I'd compare it to our 2009, since most of us have a pretty firm grasp on what we thought of that D (I'll update with the rest of the Big Ten next week but I don't have the time just yet). Also, for 2009 I used our first 4 and Delaware State. Chart...
|Opponent||Western||Notre Dame||Eastern||IU||Delaware St|
|1||6||0||Punt||69||0||FG Miss||8||0||Punt||80||7||Rush TD||-1||0||Punt|
|3||6||0||Punt||56||3||FG Made||79||7||Rush TD||-5||0||Punt||5||0||Punt|
|4||0||0||Int||76||7||Pass TD||8||0||Punt||67||7||Rush TD||2||0||Punt|
|6||-14||0||Punt||17||3||FG Good||36||7||Rush TD||52||3||FG Made||14||0||Punt|
|9||80||0||TO on Downs||17||0||Punt||6||0||Punt||8||3||FG Made||14||3||FG Made|
|11||85||7||Pass TD||36||7||Rush TD||55||0||TO on Downs||72||3||FG Made||76||3||FG Made|
Again the italized EOH drives were not counted in the following:
Ok, great, so now we know exactly much worse our D is this year than last year but we also played Sparty last year so let's see how that turned out...
|10||45||0||TO on Downs|
So using that same Sparty 137% over achieving you end up with the D doing this...
Eeek! So how many point are we looking at? Well, the average number of drives faced thus far in 2010 per game is 12 and the average number in 2009 was 13. Last year Sparty had 11 countable drives so I'll call that a wash compared to this year. With that in mind where does that leave this years D vs Sparty?
Well - 2.91*11 = 32 points and if we give them another posession they get 35 points.
Summary - well, it looks like comparing this years data to last years data our D might be giving up another 14 points in this game but our O is also much better. The next natural step is to look at how MSU is performing this year compared to last and merge the two sets of data but, as I said, that's for another week as it's already Thursday and I've got work to do!
My prediction (knocking on wood, throwing salt over shoulder, every other non-jinxing thing you can think of) UM 38-MSU 35 (I think we'll get a 27 yard FG at some point along the way).
I welcome any suggestions/additions and I'll try to update this weekly and expand it to all of the Big Ten and just have summary data in the future so as not to make it too long.
Update: I've updated with the fixed numbers for the IU EOH TD and pushed the prediction to 38-35 - maybe that 27 yard FG will come at the end of the game?
Update 2: For Mat - I've run the Offensive numbers comparing ourselves to last year and, in a word, NNNOOOO!!!!
Keeping UMass and DSU in the calc I ended up with our O only getting 57% of it's expected output (OUCH!) with about 1 posession more per game. Even if we throw in that extra posession our expected offensive output is only 2.058 pts/posession leaving us at 25 points. Of course 09 was bouyed by the DSU game pulling in over 5pts/possession so if we take that out that game you end up with an expected offensive output of 26 for 11 drives or 28 for 12 drives.
Basically, our O is going to have to do MUCH better against their D than they did last year if this is going to be a win because last years O didn't do squat (60% <= squat).
Also, something of note to give hope for this years O vs last years (even after accounting for drive efficiency) is the massive decrease in number of negative yardage drives (not just plays but entire drives!).
2009 negative yardage drives through 4 OOC + IU = 9
2010 negative yardage drives through 4 OOC + IU = 2.
I'm still sticking to my prediction but, hopefully, the UM O will do better against MSU than it did last year otherwise we're going to be hurting.
In case you're curious our O, as mentioned in another diary, is at 3.614 PPP. If Sparty is going to get the 35 points predicted above and we get 12 posessions we'll need 3.166PPP to get to 38 or 3.5PPP to get to 42 - both below our season average.
Again - knocks on wood, throws salt over shoulder, yada yada yada...
Didn't Indiana score 5 td's? I only count 4 on the chart.
Yeah, I think their td at the end of the first half is missing.
I was looking at the drive chart and rather than say "Pass TD" it just said EOH - I'll update that and update the numbers. Basically it increases Sparties expected points by 2 so 38-35 I guess...
Definitely an interesting way to look at it. Nice work bigmc6000, hopefully your prediction is correct.
Thanks for doing this - I wanted to go through the UFR's to look at what our defensive stops looked like, but I can use your data to do most of what I had in mind.
I took a look at all the drives that we got stops on (other than EOH drives), breaking them down by how long the drive was. I think this will show how much we're bending, and if we bend do we break.
|Drive Length||Drives||% of Drives||Stops||% of Stops||Stops/Drives||Stops w/ TO|
|<= 10 yards||20||31%||16||42%||80%||19%|
|11 to 20 yards||3||5%||3||8%||100%||33%|
|21 to 30 yards||12||19%||10||26%||83%||30%|
|31 to 40 yards||3||5%||3||8%||100%||33%|
|41 to 50 yards||5||8%||3||8%||60%||0%|
So, we actually get a fair number of quick stops. However, the longer the drive is, the more dependent we are on getting a stop via a turnover (interception, fumble or missed field goal) rather than just stiffening.
Another way to look at that is to ask, given that a drive has already gone X yards, what is the probability of a stop and/or a turnover?
|Drive of at least||Prob(Stop)||Prob(TO)||Stops w/ TO|
Not surprisingly, the longer the drive goes, the less likely we are to get a stop, and the more likely we are to need a turnover to get a stop. With Brian now including the number of rushers in UFR it would be interesting to see whether we're more likely to blitz the longer a drive has gone in order to get a TO.
With your permission I'd like to, possibly, incorporate this into next weeks numbers (should I find the time). I think it could pair nicely with the oppositions numbers for a pretty complete statistical analysis.
Sure, go ahead. The other thing I was hoping to pull from the UFRs was whether on defensive stops we got other kinds of "big plays" in the last set of downs - sacks or TFLs.
Oops, I left in the EOH drives for the total drives, but not for stops. Here are the corrected tables. I also switched the Indiana EOH touchdown to be a scoring drive.
|Drive Length||Drives||% of Drives||Stops||% of Stops||Stops/Drives||Stops w/ TO|
|<= 10 yards||16||27%||16||39%||100%||19%|
|11 to 20 yards||3||5%||3||7%||100%||33%|
|21 to 30 yards||12||20%||10||25%||83%||30%|
|31 to 40 yards||3||5%||3||7%||100%||33%|
|41 to 50 yards||4||7%||3||7%||75%||0%|
|Drive of at least||Prob(Stop)||Prob(TO)||Stops w/ TO|
rather than sorting Drives by yards, sort them by the number of plays. Two reasons for that. First it would mesh better with the previous Mathlete work. I'm basing that on thinking that statistically each play is an opporunty to succeed or fail. In general yes a longer drive is made up of more plays, but that's not always the case. And I know that field position is measure out in terms of yards, but again I think his analysis is dependent on the number of plays/opportunites.
the second reason is that this would separate out the ND long pass plays from the grinding drive plays of Indiana. Again it is two different kinds of defensive failure on those different drives that are likely both inside your 51+ bin of stats.
Finally it seems like your sorting also skews the results to concluding that our defense is worse at stopping long drives versus short drives. Again going to the number of plays in a drive and sorting on that could help quite a bit. And your comparison bins need to be equal in weight (events) so that your percentages have enough sample size to cancel out the outliers.
Your 11-20 bin and your 31-40 bins have two few events to really conclude that the defense is super awesome at stopping drives of that distance.
I'm betting that the less than 10 yard category has many successes for the defense at a very different field position than the defensive failures (TD's). Basically cause the only defensive failures that could exist in that category are drives that started in the redzone. Which Michigan has been very good at preventing this year by hanging on to the ball and being adequate on returns. (knock on wood)
But definitely very cool work, and please take my comments as contructive criticism in the hopes that you continue doing this analysis.
I wonder if there should be a special wiki section for all the math geeks to collaborate in on this Blog? Sharepoint to save the files? Rules for contributing data?
I'd be willing to do some data scraping if someone can draft a process and provide somewhere to dump the files.
I actually do this kind of analysis at work, and have started things for football as a hobby, but it's just daunting to attempt to complete this kind of stuff in real time as a hobby.
I definitely think it would be interesting to break drives down by number of plays rather than yardage - I used yardage because that was the information bigmc6000 had in original table. I do think the yardage breakdown is useful if you think about three groups - immediate stops, short drives and long drives. About a third of the drives against our defense stop in 10 yards or less (i.e. one set of downs), about a third go for a while and then stop, and a third drive for 50+ yards and mostly score. Field position certainly plays into this, but again wasn't in the data yet.
I think the other thing that you can tell from yards is that when a drive has gone a long way, we basically need a turnover to stop it. We're not very good at just forcing a stop when the other team has already gotten deep in our half of the field. You see that clearly in the second table, which looks cumulatively rather than by total drive yardage (so avoids the problem of categories with few observations). The longer a drive has already gone, the more we need a turnover to stop it.
I would definitely be in favor of someplace where basic scraped data can be centrally located. bigmc6000 posting the individual drives in addition to his analysis made it easy for me to add my two cents. It would be very useful, for example, to have the basic play-by-play numbers from the UFRs in a spreadsheet. On the other hand people like Mathlete who do a lot to transform the raw data into something else like PAN shouldn't feel like they should have to share that (though if he wants to that'd be great!).
Nice work. All of the diaries about this game are starting to calm my nerves. However, I do have to ask...38? Really? You think we'll kick a FG?
* = Kicked by Forcier, held by Denard ;) haha
Well, then -- 3.6 PPP for Michigan's O and 2.1 for Michigan's D. Based on an average of 12 possessions per team per game -- that is an 18 point advantage for Meeechigan.
Not too shabby.
Only if it were equally distributed...
Our offense so far has rolled on everybody, but the PPP for the defense has varied quite a bit. If you figure they give up more like 3 PPP except against missing-starting-QB teams (BGSU, most of ND, and UConn is an honorary member,) that's a 0.6 PPP x 12 possessions = 7.2 point advantage.
the indiana touchdown at the end of the half is listed as EOH, the indiana points per possession should be 2.91666
the total points per possession would go up to 2.11666
Everyone we are playing this year is having to keep up with us in scoring. The play calling is more aggressive against our D and with good reason....our offense is disturbing to all in college football save UM fans. If our D was on a team that had a weaker offense their numbers would not be as bad. The problem is our defense is not at a point where we can stop an offense in panic mode yet. It will come. I would argue that this defense is no worse than last years, but rather they are facing more reckless opponents and not able to capitalize yet.
I'm really glad to see these tempo-free metrics gaining some traction, at least here on mgoblog. It's so much more interesting and many of the common metrics, like (oh god) time of possession.
Looking forward to you crunching the numbers for the other games.
Fantastic work. You will never see such detail and numerical accuracy on any other team's blog site. I continue to be blown away by the metrics and the in-depth analysis provided on this site.
One question. What is this "work" you speak of? And, how is it remotely important compared to the mathematical destruction of lil' brother prior to the physical destruction that will take place Saturday, beginning in 50 hours!
Thanks to the mods for fixing the formatting, looks so much better in Maize-n-Blue :)
Brian Fremau talks about some of these same issues in his Football Outsiders column this week (http://footballoutsiders.com/fei-ratings/2010/fei-better-late-never). He handles things a little differently, since he strips out "garbage time" posessions and he ignores games with FCS schools. He only lists a couple of top 10 tables, but he lists Michigan as 9th in terms of points per possession on offense with 3.386. Top 10 defenses allow somwhere in the 0.5 to 1.3 points per posession range, so we're letting teams score twice as often as that. He also shows that top 10 defenses are able to make a stop for a drive that has already gone 10+ yards at least two-thirds of the time, where we're only doing it half the time (see below).
Overall he's got us ranked 16th with a rating of 0.142 (I don't have a good feel for the units on his metric, but the #1 team is 0.297 and the #25 team is 0.122. projects us to win on average 4.4 of our remaining games (so end with a total of 9.4 wins). For comparison Michigan State is ranked 31st with a rating of 0.95. He doesn't have a prediction yet (will probably come out later today), but for reference last week he had MSU at 0.73 and Wisconsin at 0.69 and gave MSU a 60% chance to win. He had Iowa at 0.148 and Penn State at 0.102 and gave Iowa a 71% chance to win. So I'm guessing we'll be like 65-70% favorites in his system.