Best, Bears Add Consistency To Their Running Game
by ragnarok on Sep 9, 2009 1:00 PM PDT
in Football
43 comments
Jahvid Best has long been a highlight reel machine running the football, but one of the knocks on him has been a reputation for a lack of toughness. It was felt that, for all his speed, when you needed tough yards up the middle, Best wasn't the guy to give them to you. For every 20+ yarder he broke off, there were far too many runs up the middle for a yard or less, the sort of maddening inconsistency that can really stall an offense. Saturday night vs. the Maryland Terrapins, I think he took a good step forward towards remedying that.
To illustrate what was, I picked a game at random from last year: Cal's Emerald Bowl victory over Miami (FL). Jahvid Best finished that game with 186 yards and 2 TDs on 20 carries, a stellar 9.3 yards per carry average. In busting runs of 15, 19, 25, 28, 32 and 42 yards, he left television audiences breathless with his talent, cementing his status over the offseason as a darkhorse Heisman candidate. However, those 6 carries listed above also accounted for 87% of his yards on the night. Meanwhile, fully half of Best's carries vs. Miami (10 of 20) went for a yard or less -- a recipe for 2nd & long, 3rd & long, and stalled drives all night.
What I'd like to do now is illustrate a more concrete way of measuring the consistency of the running game. To do that, I'm going to introduce the concept of 'success rate'.
Here I'm going to crib from RockMNation's Bill C., who defines it thus:
Success Rate is the standard for a lot of the stats I use. I've modified it slightly so that there's roughly a 44% chance of 'success' on any given play. Here are the rules according to down:
1st Down: 50% of necessary yardage. If it's 1st-and-10, you need 5 yards for 'success'. Football Outsiders use 40% for 1st down, but with the games I've entered, that led to a 1st down success rate of about 51%. Bumping the requirements to 50% led to the 44% rate for which I was aiming.
2nd Down: 65% of necessary yardage (rounded up to the nearest yard, of course). If it's 2nd-and-10, you need 7 yards for 'success'. 2nd-and-15? 10 yards. This makes sense, really, because to succeed regularly on 3rd downs, you need to stay at 3rd-and-5 or less. Getting most of the way there on 2nd downs sets you up infinitely better for 3rd down.
Football Outsiders uses 70%, but the success rate for that was around 42%. Weakening the requirements slightly got me into the range I was looking for.
3rd and 4th Downs: 100% of necessary yardage. I figure this requires no explanation.
I think this describes a fairly intuitive definition of what most football observers would determine a 'successful' play to be, at least under normal conditions. Applying this standard to Best's Emerald Bowl numbers, we find that just 11 of his 20 rushes (55%) were successful; the rest of them left the Bears in a lurch, facing 2nd or 3rd & long and needing a big play to keep the drive alive. Not that you could really pin all the blame on Best, as the rest of the Bears' running game wasn't any better. Add in Shane Vereen's 8 carries and Jeremy Ross' end-around, and Cal was still successful on just 16 of 29 rushing attempts, still a 55% average.
And as I did more research, it turns out that, despite all the misfires, a 55% success rate is actually pretty good, at least for our Bears. Only 3 times last season did the Bears have success at a greater rate, all against terrible teams (Colorado State, Stanford, and an amazing 71% success rate vs. Washington, who had by that time given up on both their season and their coaching staff). The mean was about 46%, and that included some terribly unsuccessful days vs. Maryland (36%), USC (25%) and Oregon (22%, with 3 fumbles in the rain).
Versus Maryland last Saturday, however, the Bears found a lot more success on the ground, and more importantly, they found it more consistently. All the major media outlets will focus on Best's 73-yard touchdown run, but I was equally happy to see Best gain 4, 5, 6 yards up the middle. Much of the credit, whenever you're running the ball, has to go to the offensive line, and this time was no exception. They opened holes and pushed defensive linemen around all night long, whether the ball carrier was Best, Vereen, Covaughn Deboskie-Johnson, Brian Holley or Peter Geurts.
Jahvid Best only had 10 carries Saturday night vs. Maryland before taking a seat, but those were some highly successful carries, picking up great yardage on 7 of them, an excellent 70% success rate. While Shane Vereen was less of a gamebreaker than Best, with rushing totals that weren't nearly as impressive (48 yards on 10 carries), the carries he did get were successful at an equally impressive rate (7 of 10). Throw in a couple more miscellaneous carries from Holley and Isi Sofele before Cal pulled all the starters and went into 'run straight into the line' mode, and Cal's running game posted a very impressive 73% success rate (16 of 22) for the game, better than any performance last year.
Speaking of Brian Holley, I was particularly impressed with the one carry he did get, wherein he drove forward straight into the line to pick up a yard on 3rd and 1, and then kept his center of gravity low and his feet moving to carry would-be tacklers forward for an additional four yards. Just like a fullback is supposed to do. As a group, Cal's run game was a perfect 4/4 on 3rd and short conversions, a statistic that makes me feel all warm and fuzzy inside.
There are certainly nits you could pick with Cal's overall performance on Saturday night, but not much you could say against the run game. We don't yet know how good (or bad) Maryland's defense is relative to the rest of Division I-A, and we certainly don't want to get ahead of ourselves after one game, but I think this years' group could be better than last years', even without Alex Mack. Add in a confident Kevin Riley and an improved receiving corps, and the 2009 edition of the Bears offense could end up being very, very special.
6 recs |
43 comments
|
Email
Print
Comments
re?’d
"We lose to Stanford in many sports, but if you want to make a Cal team quit, bring a weapon."
--Coach Clark
by carp on Sep 9, 2009 1:27 PM PDT reply actions 0 recs
Huh. The scientist in me wonders two things regarding ‘success rate’ — what is the significance of 44% (is it tied to overall 1st down achievement?), and also how to define or measure the 44% success rate, i.e., the measurement is not independent of Best’s above average success rate. Maybe it’s measured across all of college football…?
But I’m too lazy to look this up, so I’ll never know.
by sec119 on Sep 9, 2009 2:15 PM PDT reply actions 0 recs
I know that the 44% measurement was done across all of college football, at least for the 2007 season. As to why Bill C. wanted to achieve that number? I don’t know. Something I should ask him, perhaps.
So, basically, you gotta Go Bears!
by ragnarok on Sep 9, 2009 2:26 PM PDT up reply actions 0 recs
Great post – but I too am wondering about the 44% number.
dboneisloose
by HolmoePhobe on Sep 9, 2009 3:01 PM PDT up reply actions 0 recs
It sounds to me like we can think of a 44% success rate as “par”. By your analysis above, Rags, Cal had a very uneven season last year, with lots of bogies and birdies to leave the Bears at even par for the year.
Why 44% and not 40% or 50%? Who knows? But if 44% is the national average given a defined set of criteria (e.g., half the distance on first down), then 44% becomes our analytical par.
Go Bears!
by California Pete on Sep 9, 2009 7:30 PM PDT up reply actions 0 recs
flag’d for writing “scientist” without specifying your field or micro-fields. Or nano-fields. I give up.
"We lose to Stanford in many sports, but if you want to make a Cal team quit, bring a weapon."
--Coach Clark
by carp on Sep 9, 2009 6:01 PM PDT up reply actions 0 recs
Those 3rd down conversions made me happier than almost anything I saw on Saturday night. I feel like our offense this year is just…way more offensive, as stupid as that sounds.
What happens in Vegas stays in Vegas. What happens in California makes the world go round.
by Spazzy Mcgee on Sep 9, 2009 5:13 PM PDT reply actions 0 recs
Those 3rd down conversions made me happier than almost anything I saw on Saturday night.
Major flag’d.
"We lose to Stanford in many sports, but if you want to make a Cal team quit, bring a weapon."
--Coach Clark
by carp on Sep 9, 2009 6:01 PM PDT up reply actions 0 recs
It’s interesting to think that Cal probably had the 2nd or 3rd most powerful run attack in the Pac-10 last year (behind Oregon and/or USC) and still only had a mean of 46% success on running plays. It makes me wonder what the conference and nation averages are. How bad was Wazzu? Was Toby Gerhart as successful as everybody makes him out to be?
Great stuff Rags.
The #1 greatest threat to America: BEARS
by norcalnick on Sep 9, 2009 6:33 PM PDT reply actions 0 recs
I was extremely happy to see us converting 3rd and short and Jahvid getting the 4-6 yard runs (between the tackles, no less), in addition to the longer ones. It’s interesting to see this put numerically.
by dchu on Sep 9, 2009 6:40 PM PDT reply actions 0 recs
Fullbacks
My observation might have been selective, but it sure seemed to be that Cal’s running game got a big boost from the fullback position. Holley’s third-down run was great, but so was his lead blocking all night long. And when Kapp came in, he picked right up where Holley left off.
If Danzig doesn’t have anything better to do, how about a highlight package of Cal’s run game versus Maryland, focusing on the work done by the O Line and the fullbacks to open up those holes.
Go Bears!
by California Pete on Sep 9, 2009 7:34 PM PDT reply actions 0 recs
Props to the fullbacks
I agree. Holley’s refusal to give up and to get the yards they needed was some serious football.
I'd like to smell the Roses before I die.
by BTown85 on Sep 9, 2009 8:46 PM PDT up reply actions 0 recs
This is the whole point of standard deviation
Football stats people often crack me up in how they reinvent the wheel that the mathematicians already created. If you want to find run consistency of a back use average – 1 std dev. If you want to be REALLY picky, use median – 1 std dev.
That’ll tell you all you need to know about consistency.
by kencraw on Sep 9, 2009 10:12 PM PDT reply actions 0 recs
I think the problem with doing that is that you lose the context in which yards are gained. A 4th and 1 carry that goes for 3 yards is far more valuable than a 3rd and 19 carry that goes for 15, for instance. I think that was the big dichotomy between Tatum Bell and Mike Anderson a few years ago on the Broncos – people were up in arms saying that Football Outsiders had to be insane saying that Mike Anderson was more valuable as a runner than Tatum Bell was when Bell has the higher YPA, but Anderson was far more consistent at picking up yards necessary to get first downs while Bell was a very boom/bust runner.
No longer wanting an interview with Ryan Anderson.
by yellow fever on Sep 10, 2009 6:23 AM PDT up reply actions 0 recs
Exactly
Standard deviations are nice, but they ignore the context in which those yards were gained; success rates try to correct for that. In addition to the example given by Yellow Fever (not an uncommon occurrence), I’ll add one from Best’s Emerald Bowl performance. I mentioned above that half (10 of 20) of Best’s carries went for a yard or less vs. Miami. This is true, but it is also misleading. One of those carries was a 1-yd. run on 1st and Goal on the 2, which, despite the minimal gain, can be considered a success (50% of the necessary yards gained). Best’s next carry was also a 1-yd. run, and this one went for a touchdown — another success! Both of those carries lower the Yards Per Carry average, both lower the median and the standard deviation, but both were successful runs, and as such raise Best’s success rate.
So, basically, you gotta Go Bears!
by ragnarok on Sep 10, 2009 8:09 AM PDT up reply actions 0 recs
But in converse, the metric suggested tells you nothing about how closer or far a player came to meeting the somewhat arbitrary success metrics. I consider 4 yards on 1st down a success and if you give me a runner who gets 4 yards every 1st and 10 I’ll be a very happy man. The guys who gets 1 yard half the time and 7 the other half, would actually do far better in the suggested metric.
If we insist on coming up with a benchmark where success is a necessary component of judging a runner (which, while I see the value in, I think the arbitrary-ness of it on anything but 3rd and 4th down limits its value), it can’t be a binary metric. Perhaps re-formulate each play in terms of a percentage of success (so 4 yards on 1st and 10 would be 80% of the 5 yards/50% of success based the the suggested metrics) and then take the average percentage – 1 standard deviation. It may also be necessary to cap the maximum percentage (a 1 yard 3rd down conversion that goes for 15 yards will unfairly skew things even with 1 standard deviation) as well (at something like 300%). That might give a number I could get behind, even with the arbitrary components.
All of that said, the law of averages makes it so it’s pretty safe to ignore all the 3rd and 1 caveats for every-down backs. Particularly considering there are plenty of other caveats that aren’t covered by the success metric. To use Fever’s example, what if it was 3rd and 19 from the opponents 35 yard line and the goal of the play was to ensure they got in field goal range while taking an opportunistic shot at the 1st down? I’d say the 15 yard run met the goal perfectly. One just can’t account for every scenario and it feels like a fools errand to try to even account for some of the caveats from my vantage point.
As such, for every-down backs, average – 1 std dev will give you an exceedingly good metric for what is being measured without introducing any arbitrary components.
by kencraw on Sep 10, 2009 8:59 AM PDT up reply actions 0 recs
Ken, out of curiosity, what did you study at Cal/what is your job?
What happens in Vegas stays in Vegas. What happens in California makes the world go round.
by Spazzy Mcgee on Sep 10, 2009 10:28 AM PDT up reply actions 0 recs
I’m a fake and a fraud. I didn’t study anything at Cal… but I did study Electrical and Computer Engineering at Chico State. My Dad is a Cal Engineering Alum and I was raised going to Lair of the Bear every summer, so that’s why I’ve got strong Cal loyalties.
I currently work for Hewlett Packard doing system board design (more management these days than actual design) for our big servers.
by kencraw on Sep 10, 2009 11:59 AM PDT up reply actions 0 recs
Cool.
What happens in Vegas stays in Vegas. What happens in California makes the world go round.
by Spazzy Mcgee on Sep 10, 2009 3:04 PM PDT up reply actions 0 recs
While I would agree that the thresholds for success are somewhat arbitrary, I don’t know of how you would do it any other way. And the specific example you cited of trying to get into field goal range is something akin to saying that a sacrifice bunt is as good as a hit in baseball. It isn’t, in the baseball example, because you’ve made an out, and in football, it isn’t a complete success because you haven’t made the first down.
No longer wanting an interview with Ryan Anderson.
by yellow fever on Sep 10, 2009 10:53 AM PDT up reply actions 0 recs
True, but a sacrifice bunt isn’t counted against your batting average where in this case it would be counted against your “success average”.
by kencraw on Sep 10, 2009 12:00 PM PDT up reply actions 0 recs
This is true too. Which is why it’s important to look at the context of these things. In doing this post, I was careful to ignore things like sacks (which occur on passing plays) and runs obviously designed more to run out the clock than to win the game. The 73% success rate I cited above for Saturday’s Maryland game ignored every carry by Deboskie (although his carries were still pretty successful, as it turns out, which is a great positive indicator for our offensive line).
So, basically, you gotta Go Bears!
by ragnarok on Sep 10, 2009 12:18 PM PDT up reply actions 0 recs
I really like this idea. It goes the rest of the way to including situational data but does so entirely objectively.
Still having trouble with the parsing of the play-by-play files and their lack of consistency for your database?
by kencraw on Sep 10, 2009 2:40 PM PDT up reply actions 0 recs
I won’t argue that the success rate metric is above reproach; far from it. This essay is a first stab at illuminating some overlooked and underappreciated points in the game — specifically, getting away from using the Yards Per Carry metric altogether, a statistic that is easy to produce but fraught with false implications. Whether you use success rate or a simple standard deviation, I think it’s easy to see that Best’s game had some holes in it last year, holes that weren’t present vs. Maryland on Saturday.
To answer your criticisms, yes, the line between success and failure is arbitrary. I think that in most situations, it’s in about the right place, but I too feel like 4 yards on 1st and 10 could be considered a success. I’m not sure why exactly Bill C. came up with the standard he did, although given the thoroughness of his other work, I feel pretty confident that he had some good reasons behind it. However, I too would like to get away from an arbitrary benchmark, and I’ve got some idea of how to do it, although working it out is going to take some time (hint: it involves another metric of Bill’s, Points Per Play).
As for the binary nature of the success rate metric, well, you’ve got me there. Like any complex statistic, this one has some holes, and I feel like the inability to discern between a 3-yard gain and a 6-yard loss on 1st down is the biggest weakness of success rates. It was something that bugged me even as I wrote this post, and you’ve definitely given me some more to think about where to take this next.
So, basically, you gotta Go Bears!
by ragnarok on Sep 10, 2009 11:05 AM PDT up reply actions 0 recs
I think what could be done next is a look at the value of each play. There is already data which looks at the average points scored by a team that has a first down at a certain position on the field, not unlike the transition matrices all over BP that describe the average runs scored by a team in each situation (i.e. no outs, no runners on; no outs, bases loaded; two outs, runner on second, etc.).
So you could say that a play that takes a team from their 20 to the opponent’s 20 is worth (expected value of points from opponent’s 20 – expected number of points from team’s own 20), but then, not all players get similar types and numbers of opportunities, not to mention it would require a massive amount of data to even know what those expected number of points from each position on the field are.
No longer wanting an interview with Ryan Anderson.
by yellow fever on Sep 10, 2009 11:22 AM PDT up reply actions 0 recs
not to mention it would require a massive amount of data to even know what those expected number of points from each position on the field are.
This is what’s taking me so long. I’m building a database, but getting it all right takes time. But essentially, you’ve described what I’m aiming to do — place a value on any combination of down, distance & field position (DD&P). Then, instead of picking some arbitrary line where ‘success’ happens, we could say a ‘successful’ play is any play that moves the ball from a DD&P of some value to a DD&P of greater value. This would cover situations such as moving into field goal range, and moving away from potential safety range.
So, basically, you gotta Go Bears!
by ragnarok on Sep 10, 2009 11:46 AM PDT up reply actions 0 recs
I really like this idea. It goes the rest of the way to including situational data but does so entirely objectively.
Still having trouble with the parsing of the play-by-play files and their lack of consistency for your database?
by kencraw on Sep 10, 2009 2:40 PM PDT up reply actions 0 recs
Well, part of my trouble is finding time at all to work on this project.
The parser is essentially done. If bugs exist, I don’t know about them yet, and I’m fairly confident that what bugs do exist are of the fairly esoteric variety. What’s taking more work is dealing with all of the, for lack of a better word, “unparsable” inputs, play-by-play data that is broken in strange and unique ways, such that it would either be almost impossible for a computer to automatically correct such errors, or writing a fix for them would be so stupendously time-consuming that it’s easier to do it all by hand. Which is still time-consuming.
At this point, I’m thinking I’ll need about the equivalent of a week off with no distractions to really finish this thing off.
So, basically, you gotta Go Bears!
by ragnarok on Sep 10, 2009 2:52 PM PDT up reply actions 0 recs
That’s exciting, it should be a great analysis tool. I assume there exist similar databases, at least at the NFL level, since some website (espn, maybe? I forget) posts a winning percentage during the game, which must evaluate the average points from a given field location.
by Kai on Sep 10, 2009 3:58 PM PDT up reply actions 0 recs
I’d surely bet that similar databases exist, at least for the NFL. Whether anyone is sharing that data with the generla public is another question.
So, basically, you gotta Go Bears!
by ragnarok on Sep 10, 2009 4:09 PM PDT up reply actions 0 recs
That’s true. I’m sure you would become very popular if the database was easily accessible and/or the parser was open source.
by Kai on Sep 10, 2009 4:57 PM PDT up reply actions 0 recs
Heh, I’m sure.
I’m actually more inclined to share the database than the parser (and I was definitely more inclined to share earlier in the project, before I realized how much time I was going to put into this), but it’s still up in the air. I’ve been talking to someone about possibly hosting/contributing to a massive database of this sort. When something is finalized, the CGB community will be the first to know.
So, basically, you gotta Go Bears!
by ragnarok on Sep 10, 2009 5:59 PM PDT up reply actions 0 recs
ohh, can I get admin access and run a few choice scripts setting random fields to NULL?
by turkey on Sep 11, 2009 10:57 AM PDT up reply actions 0 recs
BTW, Rags, I feel like I’ve been unduly picking on your posts lately… I didn’t even read who wrote this one before I started on my screed. ;-)
Sorry about that.
by kencraw on Sep 10, 2009 9:02 AM PDT up reply actions 0 recs
Heh, no problem, Ken. I’ve got no problem with people criticizing what I write. And besides, I’d much rather have an interesting discussion about these sorts of things than just having people accept what I write without really thinking about and challenging my ideas.
So, basically, you gotta Go Bears!
by ragnarok on Sep 10, 2009 11:14 AM PDT up reply actions 0 recs
This reminds me of the NFL player—I forget who—who described his running style as follows:
“If you need two yards, I’ll get you three. If you need seven yards, I’ll get you three.”
Go Bears!
by California Pete on Sep 10, 2009 9:06 AM PDT up reply actions 0 recs
I’m guessing it was LenDale White.
No longer wanting an interview with Ryan Anderson.
by yellow fever on Sep 10, 2009 9:12 AM PDT up reply actions 0 recs
Ken – any chance you can explain in layman’s terms why average – 1 standard deviation is a good measure of consistency?
dboneisloose
by HolmoePhobe on Sep 10, 2009 3:56 PM PDT up reply actions 0 recs
I'll give it a shot
Since standard deviation (sd) is the average amount that individual observations—in this case, rushes—vary from the mean, looking at mean minus sd provides a lower boundary of what might informally be called “normal”.
For example, imagine two running backs that average 6 yards per carry. Back #1 does so with a standard deviation of two, meaning that his average run is 2 yards more or less than 6. It’s reasonable, then, to describe his “normal” rush to be between 4 and 8 yards. Back #2, meanwhile, has a standard deviation of five, so his “normal” range would be much wider, from 1 to 11.
If one is looking for a metric that provides a sense of a back’s consistency, how reliably he produces significant positive yardage, then Ken’s suggestion isn’t bad at all. Alternatively, one might look at percentiles (e.g., yards gained or exceeded on 90% or 75% of the back’s carries)—in essence, redefine “average” as the median, rather than the mean. For a guy like Jahvid, who so frequently busts runs of 20+, even 70+, yards—extreme values that can really inflate a standard deviation, the percentile route is probably the better way to go.
Go Bears!
by California Pete on Sep 10, 2009 8:47 PM PDT up reply actions 0 recs
Thanks for the explanation. So in a very rough sense, you’re looking for the number of yards that the back can somewhat consistently be counted on to deliver?
dboneisloose
by HolmoePhobe on Sep 11, 2009 12:28 AM PDT up reply actions 0 recs
Comments For This Post Are Closed
Secondary Sidebar
User Tools
Related FanPosts on California Golden Blogs
FanPosts
Community blog posts and discussion.
Certain photos copyright © 2010 by Associated Press or Getty Images. Any commercial use or distribution without the express written consent of Associated Press and Getty Images is strictly prohibited.
