Forums
New posts
Search forums
What's new
New posts
New media
New media comments
New resources
Latest activity
Media
New media
New comments
Search media
Resources
Latest reviews
Search resources
Members
Current visitors
Log in
Register
What's new
Search
Search
Search titles only
By:
New posts
Search forums
Menu
Log in
Register
Install the app
Install
Forums
Open Football Discussion
Green Bay Packers Fan Forum
How Should We Evaluate a Defense?
JavaScript is disabled. For a better experience, please enable JavaScript in your browser before proceeding.
You are using an out of date browser. It may not display this or other websites correctly.
You should upgrade or use an
alternative browser
.
Reply to thread
Message
<blockquote data-quote="HardRightEdge" data-source="post: 793332"><p>The general approach makes a lot of sense, but it's another black box. We cannot see inside as to how the varying outcomes of varying plays are weighted in the rankings. We've seen oddities in PFF grading, such as the aggregate grade being higher than any of the more granular grades, without explanation. Whether PFF or Football Outsiders, the scoring has to be taken on faith.</p><p></p><p>We encounter black boxes all the time, particularly in the world of finance. An algorithmic ETF or hedge fund, for example, claims a certain objective. We may not see inside how the algorithms operate, but we can evaluate its performace over time against external benchmarks. For example, does an ETF which claims low volatility outformance against the S&P 500 in a down market offsetting underperformance on the upside actually perform against that externally measurable objective?</p><p></p><p>The thing about PFF or Football Outsiders black boxes is there is no external objectively determined reference against which to judge the effectiveness of their algorithms. We're left to compare the grading of one black box to that of another, a closed loop. We know we can't judge the grading of a defense, for example, on yards or points with all of the extenuating circumstances, not the least of which is performance of the offense. We're left to judge the performance of the algorithms by the eye test.</p><p></p><p>I'm also skeptical of "level of competition" adjustments in grading algorithms such as Football Outsiders. There's an element of infinite regress or circularity in those computations. For example defense A has a "bad" performance against Offense B's "good" performance in week 1. Then Defense A has a "good" game against Offense C in week 2 whereby Offence C is downgraded for having played a previously "bad" defense. But maybe Defense A was actually "good" in week 1 but Offense A was "really, really good" against them. Again, there's no external reference by which to judge and it has an element of chasing one's tail.</p><p></p><p>Further, a thought occurs after watching KC last night. They get a 28th. ranked defense from Football Outsiders after week 5. Whoever they play next week or next month, and whatever that offense might do, we presume that offense's results will be downgraded by having played against a "bad" defense. However, Houston has been hobbled with a hamstring, with limited snaps in week 5 and none in week 6. Eric Berry has not played since week 1 last season with an Achilles injury. What happens if, in a few weeks, these guys step on the field together and return to something resembling their previous All Pro form? As a predictive tool, no algorithm is going to account prospectively for the sudden improvement in that defense under those conditions. And it would skew the grading of the opposing offense that week based on the "bad" KC defense coming into that game.</p><p></p><p>The approach I took in the OP, admittedly rudimentary, has the benefit of transparancy. The 3 elements capture most of what constitutes a good defense, without getting into the regress involved in gauging level of competition, or the pitfalls of ranking on points or yards:</p><ul> <li data-xf-list-type="ul">passer rating against rank</li> <li data-xf-list-type="ul">"stop" rankings (3rd. down, 4th. down, red zone TDs)</li> <li data-xf-list-type="ul">fumbles recovered</li> </ul><p>What it noteably lacks is a measure of whether a defense has a propensity to surrender long runs or contain them. Those runs are uncommon enough to not have a large affect. If there was an easy way to rank that I'd add it to the list. Rushing defense performance, where it most often and most critically comes into play in the aggregate is accounted for in the stop rankings. The ability to cause and recover fumbles might have a run defense contribution imbedded in it.</p><p></p><p>I didn't create any algothim to weight these factors. My C+ grade for the Pettine defense to date, at the upper end of the 14th. - 19th. ranking "C" range, was an eye test balancing of the three factors with a little Kentucky windage. Others might come up with a slightly different overall grade based on those factors.</p><p></p><p>Interestingly, the Football Ousiders overall Packer defensive rank of 17th. is a virutal tie for 15th. with Pittsburgh and Dallas. All three are shown as 0.0%:</p><p></p><p><a href="https://www.footballoutsiders.com/stats/teamdef" target="_blank">https://www.footballoutsiders.com/stats/teamdef</a></p><p></p><p>One presumes the difference between 15th. and 17th. is a matter of taking their calculation out to 2 decimal positions which they do not display. That difference is presumed trivial.</p><p></p><p>If we regard their ranking as effectively a tie for 15th., that's squarely in alignment with my C+ grade. Again, that's a blunt performance grade. Pettine's performance grade is higher for improvement over last last season's dismal performance by the same measures. Football Outsiders graded the Packers defense 20th. at the conclusion of 2017, which would be a D+ grade in my scheme. I give that defense an F+, to do with as one sees fit.</p><p></p><p>One thing we should all agree on is this, from Football Outsiders:</p><p></p><p>"The popularity of fantasy football only exacerbates the problem [of grading and ranking]. Fans have gotten used to judging players based on how much they help fantasy teams win and lose, not how much they help <em>real</em> teams win and lose."</p></blockquote><p></p>
[QUOTE="HardRightEdge, post: 793332"] The general approach makes a lot of sense, but it's another black box. We cannot see inside as to how the varying outcomes of varying plays are weighted in the rankings. We've seen oddities in PFF grading, such as the aggregate grade being higher than any of the more granular grades, without explanation. Whether PFF or Football Outsiders, the scoring has to be taken on faith. We encounter black boxes all the time, particularly in the world of finance. An algorithmic ETF or hedge fund, for example, claims a certain objective. We may not see inside how the algorithms operate, but we can evaluate its performace over time against external benchmarks. For example, does an ETF which claims low volatility outformance against the S&P 500 in a down market offsetting underperformance on the upside actually perform against that externally measurable objective? The thing about PFF or Football Outsiders black boxes is there is no external objectively determined reference against which to judge the effectiveness of their algorithms. We're left to compare the grading of one black box to that of another, a closed loop. We know we can't judge the grading of a defense, for example, on yards or points with all of the extenuating circumstances, not the least of which is performance of the offense. We're left to judge the performance of the algorithms by the eye test. I'm also skeptical of "level of competition" adjustments in grading algorithms such as Football Outsiders. There's an element of infinite regress or circularity in those computations. For example defense A has a "bad" performance against Offense B's "good" performance in week 1. Then Defense A has a "good" game against Offense C in week 2 whereby Offence C is downgraded for having played a previously "bad" defense. But maybe Defense A was actually "good" in week 1 but Offense A was "really, really good" against them. Again, there's no external reference by which to judge and it has an element of chasing one's tail. Further, a thought occurs after watching KC last night. They get a 28th. ranked defense from Football Outsiders after week 5. Whoever they play next week or next month, and whatever that offense might do, we presume that offense's results will be downgraded by having played against a "bad" defense. However, Houston has been hobbled with a hamstring, with limited snaps in week 5 and none in week 6. Eric Berry has not played since week 1 last season with an Achilles injury. What happens if, in a few weeks, these guys step on the field together and return to something resembling their previous All Pro form? As a predictive tool, no algorithm is going to account prospectively for the sudden improvement in that defense under those conditions. And it would skew the grading of the opposing offense that week based on the "bad" KC defense coming into that game. The approach I took in the OP, admittedly rudimentary, has the benefit of transparancy. The 3 elements capture most of what constitutes a good defense, without getting into the regress involved in gauging level of competition, or the pitfalls of ranking on points or yards: [LIST] [*]passer rating against rank [*]"stop" rankings (3rd. down, 4th. down, red zone TDs) [*]fumbles recovered [/LIST] What it noteably lacks is a measure of whether a defense has a propensity to surrender long runs or contain them. Those runs are uncommon enough to not have a large affect. If there was an easy way to rank that I'd add it to the list. Rushing defense performance, where it most often and most critically comes into play in the aggregate is accounted for in the stop rankings. The ability to cause and recover fumbles might have a run defense contribution imbedded in it. I didn't create any algothim to weight these factors. My C+ grade for the Pettine defense to date, at the upper end of the 14th. - 19th. ranking "C" range, was an eye test balancing of the three factors with a little Kentucky windage. Others might come up with a slightly different overall grade based on those factors. Interestingly, the Football Ousiders overall Packer defensive rank of 17th. is a virutal tie for 15th. with Pittsburgh and Dallas. All three are shown as 0.0%: [URL]https://www.footballoutsiders.com/stats/teamdef[/URL] One presumes the difference between 15th. and 17th. is a matter of taking their calculation out to 2 decimal positions which they do not display. That difference is presumed trivial. If we regard their ranking as effectively a tie for 15th., that's squarely in alignment with my C+ grade. Again, that's a blunt performance grade. Pettine's performance grade is higher for improvement over last last season's dismal performance by the same measures. Football Outsiders graded the Packers defense 20th. at the conclusion of 2017, which would be a D+ grade in my scheme. I give that defense an F+, to do with as one sees fit. One thing we should all agree on is this, from Football Outsiders: "The popularity of fantasy football only exacerbates the problem [of grading and ranking]. Fans have gotten used to judging players based on how much they help fantasy teams win and lose, not how much they help [I]real[/I] teams win and lose." [/QUOTE]
Insert quotes…
Verification
Post reply
Members online
DABIGZ
Latest posts
Prospects you are A LOT higher on than consensus…
Latest: OldSchool101
12 minutes ago
Draft Talk
S
Free Agency Thread
Latest: sschind
Today at 3:48 PM
Green Bay Packers Fan Forum
Rodgers for Vice President?
Latest: Pokerbrat2000
Today at 3:47 PM
Aaron Rodgers Discusson
2024 Draft Prospect Discussions
Latest: Pokerbrat2000
Today at 3:42 PM
Draft Talk
Final Annoying Mock/Prediction
Latest: tynimiller
Today at 3:28 PM
Draft Talk
Forums
Open Football Discussion
Green Bay Packers Fan Forum
How Should We Evaluate a Defense?
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.
Accept
Learn more…
Top