Interesting find. I like the charts and I like that the homer in me can more easily come out when I'm tailgating tomorrow in East Lansing.
Point per Field Positon Breakdown – Week 5
For obvious reasons, field position is critically important to the success of a football team. This is a timeless fact. Teams starting with poor field position have a tougher time scoring because they must move the ball further to get into scoring position. Football geeks wearing green eyeshades and helicopter caps such as myself talk about an obscure stat related to this; I’m not sure it even has a name. I’ll refer to it as points per field position, meaning the average number of points a team has scored (or allowed) from a given starting field position. I thought it’d be interesting to look at this week’s matchup through this prism. For this analysis I’ve split the field into 25 yard chunks for simplicity’s sake and also to boost statistical significance. I’ve excluded Montana St. from MSU’s dataset since they are an FCS opponent and therefore not worthy of analytical scrutiny. I keed…not really. Onward
Advantage: Push with a nod to Michigan. The reason I say push is that the overall averages of both teams are within 0.1 points of each other. I don’t pretend that this analysis is able to reliably detect this level of separation. However, closer scrutiny of the data reveals an apparent advantage for Michigan. That is, in addition to being awesomer.
Starting in their own territory, both offenses perform about the same. The separation comes when the offenses start drives in enemy territory. There are only 2 drives available for MSU, which I take as evidence that their offensive special teams (the return teams) are not sweet, but there’s nothing to say that more data will necessarily improve their average rather than hurt it. So, for now, I’ll believe my flimsy 2 drive average of 3 pts per possession. For equal filed position, Michigan scores an average of 4 points per possession. Hence they get the nod for having demonstrated proficiency in enemy territory. Also from what I’ve seen on the field I think Michigan does show more offensive prowess and schematic advantage than does MSU. Also I’m a homer. Plus we have the Force. And I have a chart…go go gadget CHART!
Advantage: Michigan. This is probably better characterized as Michigan having a lower disadvantage than Sparty because I think it is safe to say that both defenses are pretty damned bad. Think of it as having a lower handicap. I’m actually surprised at Michigan’s overall average of 1.8 points per possession…it felt like it’d be higher. Sparty on the other hand is basically allowing as many points as its scoring, 2.8 points per possession. Again, see the chart.
Advantage: Michigan. Here I’ve split special teams play into coverage and return units with each establishing defensive and offensive field position respectively. The numbers here reconcile nicely. Michigan’s coverage units and State’s return units have both ended up at about the same field position to date; at about the 26 yard line. On the flip side, Michigan’s return units and State’s coverage units have also ended up at about the same field position; at about the 31 yard line. These are just averages so, presumably, Michigan will have a few more drives starting in sector 2 and State will have a few more drives starting in sector 1. No chart here, just a table.
Obviously Zoltan is a huge asset for establishing defensive field position and Darryl Stonum is doing a kick ass job at establishing good field position for the Offense. What’s kind of aggravating is that the woes of last year seem to have the coaching staff content with simply holding onto the ball, which, yeah duh. But damn, we should be able to improve our average field position on punts if Mathews or some satisfactory replacement would be allowed to attempt a return. If the staff insists on not attempting a return then, why not skip even putting a guy back there altogether and sending 11 every time for a block? I know catching the ball stops the roll but, I wonder how much of a difference it makes overall. I dunno, it’s just annoying that we forfeit better field position simply because we’re scared of dropping the ball.
This work is based on past performance and doesn’t account for key injuries, personnel changes, and what not. Another issue is that of unequal opposition. Obviously playing weaker opposition (such as FCS caliber Montana St.) would inflate a team’s numbers so strength of schedule has something to do with the numbers. However, Sparty’s opponent record to date is 5-2 where as Michigan’s opponents are 6-3, and I've already thrown out MSU's game against Montana State so I think the relative strength of schedules are pretty even.
This analysis indicates that Michigan has performed better in all three phases than Michigan State has so far this year (as if the respective win-loss records didn't already say this). What’s not shown here is that both defenses have given up about 100 points so far this year but Michigan State has done it in about 30% fewer possessions; Yowza. Our offense has been more efficient at hitting pay dirt and our special teams have done a good job at setting up field position. These three things should at least neutralize Sparty’s home field advantage. All of this data includes possessions ending in turnovers so, barring another Notre Dame 2008 type scenario, that shouldn’t be a concern.
While analyzing the past is neat, synthesizing the future is what everybody is really interested in. I’ve got a little sumpin sumpin cooking on that but that’ll have to wait until later tonight because I’m tired of writing and you all are probably tired of reading.
Are you going to jail?
we'll see, son. We'll see.
Listened to MGoPodcast this week, and Brian mentioned that a large reason why Mathews hasn't really broken out a punt is because (a) the team seems content to just leave one guy back there and secure the ball, but also (b) the spread punt formation allows the kicking team to send quite a few men down on coverage, negating the possibility of a return. So basically, Mathews is doing what he can with the bad situation, though perhaps putting someone like Grady back there, even as just a blocker, would maybe lead to some better returns.
I like the breakdown, though again with so few data points it might be hard to find any meaningful correlations. That said, I think this reaffirms the notion that a good punter can really help a struggling defense by pinning the offense back. State does not have a particularly good punter but a similar defense to UM, and they are getting killed on the points per possession in comparison.
You might consider dividing the field into fifths (20-yds), then the 0-20 segment matches the redzone and marking it as such will increase readability. You threw me off briefly with M's offense averaging 7.0 in the 76+ region.
From what I have read on this blog I think many people forget to take this into account when analysing our defense. After putting into perspective the number of drives our d faces per game in relation to the overall yards against our defense may not be quite as bad as some people think.
We run a fast paced offense (barely running down the play clock at all) and have often times scored touchdowns after only a couple of plays because of the big play ability of the offense.
It could just be that our defense, in spite of all the Barwis training, is getting tired because of the amount of plays they face in the run of a game.
Also, when you look at the number of plays and figure out the average yards per play, they may not be quite as bad as poeple are making them out to be.
I before E except after C. I like the analysis - thanks.
This is a good analysis, but I think you should just plot all the data points, and fit a linear trendline to them. The way you draw your line is uh... biased in that you drew the trendlines. Let the numbers do the trending.
I thought about doing that, but I don’t think the trend has to be linear. Also, due to the discretization that I needed to impart in order to boost the quality of the conclusions being drawn, I forfeited the resolution require to do real linear regression. It is impossible to start a drive in the interim between sectors, by definition. The dashed lines I show aren’t meant to represent hard and fast trends, just to arbitrate between mathematical theory and football sense. My goal is to gain useful insight into the expected performance of a unit, not to derive a high fidelity model for the expected performance.
For example, on the defensive PPFP chart Michigan's D looks like it is more likely to allow a score when the opponent starts in sector 1 (0 - 25 yd line) than if they started in sector 2 (26 - 50 yd line). These are solid data points too as they are the average of 30 and 17 drives respectively. It doesn't make sense that the defense will give up less points when its opponent is closer to the goal line the defense is trying to defend. It’s more likely that there is no difference whether the opposing offense starts in sector 1 or 2, they’ll have the same amount of success; in this case, about 1.5 points per drive. In fact, the only thing that seems to matter (so far this season) is whether an opponent starts on their side of the field or ours. Hence the Z-shaped line.
Where a linear trend was obvious, I let it stand. Where the ‘trend ‘called for outright disregard for the data (the only things we know for sure) I decided to take over and make the math conform to my football sense. Where there was no data I carried over the known performance from the previous sector. As more drives in each sector become available less intervention will become necessary.
I agree that it's probably not linear, but before you have any more data, it might be better served with a linear model, at least for now.
My point is that the binning might not be necessary, as the individual data points might give better insight. It probably doesn't affect the conclusions (or hypotheses). I like looking at data.
Not binning leaves you with an unintelligible mess. The dependent variable (points scored) is also a discrete variable with values of 0, 3, or 7. Eventually you'll end up multiple data points at each node and a flat line at 3. Regression only goes for best fit, it doesn't comprehend the relative importance of each node. To get this you have to bin and average, whether in 1 or 25 yard increments. Otherwise, the R-squared will always suck.
Now, once you've binned you can do regression to your hearts desire, but at this point (some bins are empty) its no more valid of an assumption than my crayon is. And I like my crayons. So there. /raspberry
Thanks for putting up the raw data. It's interesting how few field goals M has put up, and how many they have forced.
Perhaps a good way to present the data without getting too coarse is to have 0 points, 3 points, and 7 points histograms that are binned. But perhaps that too may be too difficult to understand. Anyway, you're cool, and thanks again.
Nice work - This sort of analysis isn't something you hear about and brings to light some interesting observations of how the Offense and Defense are performing based on the hands their dealt. I'd be interested to see how this plays out over the rest of the season as there are more data points.
Any idea what the numbers looked like for last year?
This is good stuff. You should keep it in mind for the last couple games of the season when the data set is larger. Could make for some interesting analysis.
I like the premise of your analysis, but there just isn't enough data for this to mean much. I wouldn't draw the lines like you did either. Are they trend lines? If so, you should draw one line, probably a curve approaching a asymptote of 6 (or 7 for a td, not sure what you are using).
A few games later into the season and this type of analysis will be much more interesting.
Great info. What interests me is how often the teams start with various field position. For instance, if we most often both start with a long field, then that weights it to a negligible difference. If there are quite a few possessions that start with short fields, then the difference in the teams becomes more pronounced. I have a feeling the former is the case, though. This is hinted at with the average starting field position you show, as well as your comments about the little data you have for State starting with a short field on offense.
I think average starting field position could yield some more useful data if you looked just at the data points around those field positions. In other words, how do our O and their D perform when we and State's opponents start at/near the 31 (maybe do a +/- 3 to 5 yards to get a large enough sample size)? How do our D and their O perform when our opponents and they start at their own 26? I think that gives us a more insightful analysis. Average field position is somewhat limiting in that outliers can skew the data, and the 25-yard segmentations you have make it harder to apply that average, as the trend may not be linear to the other poster's point. Even if you can't do the 26 and 31 yardline analyses, maybe at least you could lump all the data from the 25 to the 35, which would give a big enough sample size.
You got me thinkin', that's for sure. Thanks for the preso.