Fixing the Seeding Algorithm

[Wrestleknownothing]

I have complained long and loud about the problem. Now it is time to attempt to be part of the solution.

I think there are three easy, common sense adjustments that can be made right away, and a fourth to think about.

1. Make H2H H2H. Drop (vs. field) as it is nonsensical. This solves two major problems. It removes a duplication in the formula, and it restores the single most important element to the seeding process. It also removes a potential gaming strategy that rewards teams for not using their starters in a conference dual.
2. Switch RPI and CR weightings. Quality wins (matches?) is very similar to RPI in that they measure the quality of the opponents faced (though in different ways). There is no need to have them both be weighted the same. Drop one down to 5% (could drop QW or RPI) and bump CR up to 10%.
3. Start with binary measures, finish with binary measures. The switch from binary comparisons (who had a better record? who won more categories?) to non-binary is bad math. No bad math.

One other thing that was discussed was the coaches' desire to emphasize showing up and wrestling in conference duals. That could be reinforced by tweaking the H2H a bit, too. If we make H2H a two tiered measure where H2H conference dual results are considered first, then if no conference result is available, H2H in non-conference competition is considered.

There is a potential problem with this, though. If two wrestlers meet at National Duals the winner has a dis-incentive to compete in a conference dual against the same wrestler. One way to solve this is to consider the H2H, for conference dual purposes only, at the team level. If Levi Haines beats whoever Iowa sends out at 174 in the dual, he is considered to have beaten whoever they send out at the conference tournament. Far from a perfect solution, but worth consideration.

 

[Corby2]

I have complained long and loud about the problem. Now it is time to attempt to be part of the solution.

I think there are three easy, common sense adjustments that can be made right away, and a third to think about.

1. Make H2H H2H. Drop (vs. field) as it is nonsensical. This solves two major problems. It removes a duplication in the formula, and it restores the single most important element to the seeding process. It also removes a potential gaming strategy that rewards teams for not using their starters in a conference dual.
2. Switch RPI and CR weightings. Quality wins (matches?) is very similar to RPI in that they measure the quality of the opponents faced (though in different ways). There is no need to have them both be weighted the same. Drop one down to 5% (could drop QW or RPI) and bump CR up to 10%.
3. Start with binary measures, finish with binary measures. The switch from binary comparisons (who had a better record? who won more categories?) to non-binary is bad math. No bad math.

One other thing that was discussed was the coaches' desire to emphasis showing up and wrestling in conference duals. That could be reinforced by tweaking the H2H a bit, too. If we make H2H a two tiered measure where H2H conference dual results are considered first, then if no conference result is available, H2H in non-conference competitions is considered.

There is a potential problem with this, though. If two wrestlers meet at National Duals the winner has a dis-incentive to compete in a conference dual against the same wrestler. One way to solve this is too consider the H2H, for conference dual purposes only, at the team level. If Levi Haines beats whoever Iowa sends out at 174 in the dual, he is considered to have beaten whoever they send out at the conference tournament. Far from a perfect solution, but worth consideration.

I said this years ago on a show with Willie come seeding time whatever team won the head to head that result counts for seeding. It was an idea to stop the ducking issues we were having at the time

 

[gimb14]

I wouldn't mind something about it hurting the guy that doesn't wrestle also. E.g., Haines gets the win over whoever Iowa sends out. Whoever Iowa enters for the postseason gets the loss.

 

[LostInTransition]

I have complained long and loud about the problem. Now it is time to attempt to be part of the solution.

I think there are three easy, common sense adjustments that can be made right away, and a fourth to think about.

1. Make H2H H2H. Drop (vs. field) as it is nonsensical. This solves two major problems. It removes a duplication in the formula, and it restores the single most important element to the seeding process. It also removes a potential gaming strategy that rewards teams for not using their starters in a conference dual.
2. Switch RPI and CR weightings. Quality wins (matches?) is very similar to RPI in that they measure the quality of the opponents faced (though in different ways). There is no need to have them both be weighted the same. Drop one down to 5% (could drop QW or RPI) and bump CR up to 10%.
3. Start with binary measures, finish with binary measures. The switch from binary comparisons (who had a better record? who won more categories?) to non-binary is bad math. No bad math.

One other thing that was discussed was the coaches' desire to emphasize showing up and wrestling in conference duals. That could be reinforced by tweaking the H2H a bit, too. If we make H2H a two tiered measure where H2H conference dual results are considered first, then if no conference result is available, H2H in non-conference competition is considered.

There is a potential problem with this, though. If two wrestlers meet at National Duals the winner has a dis-incentive to compete in a conference dual against the same wrestler. One way to solve this is to consider the H2H, for conference dual purposes only, at the team level. If Levi Haines beats whoever Iowa sends out at 174 in the dual, he is considered to have beaten whoever they send out at the conference tournament. Far from a perfect solution, but worth consideration.

That looks like an improvement over what we got but, as an engineer, I like math. You obviously like math, or stats, if you will, but I'm not so sure that math solves this problem. Logic and reason are more applicable and conference dual record should be the first and highest weighted criteria, then move on to secondary and tertiary criteria.

1. Conference dual record, if you're the only guy that is 8-0 you are the 1 seed, all other factors be damned.
2. Head to head is the first tie breaker
3. Record against common opponents and record against the field outside of B1G duals.
4. I'd go with coaches ranking as final tiebreaker though adding rpi and/or strength of schedule can be included.

 

[Aardvark86]

That looks like an improvement over what we got but, as an engineer, I like math. You obviously like math, or stats, if you will, but I'm not so sure that math solves this problem. Logic and reason are more applicable and conference dual record should be the first and highest weighted criteria, then move on to secondary and tertiary criteria.

1. Conference dual record, if you're the only guy that is 8-0 you are the 1 seed, all other factors be damned.
2. Head to head is the first tie breaker
3. Record against common opponents and record against the field outside of B1G duals.
4. I'd go with coaches ranking as final tiebreaker though adding rpi and/or strength of schedule can be included.

And don't forget though, that the current process has a last step - the black box - that we've haven't seen yet. But in advance of that, I'd add (5) Don't put limits on the black box like point differentials that would preclude the correction of egregious errors.

 

[gimb14]

That looks like an improvement over what we got but, as an engineer, I like math. You obviously like math, or stats, if you will, but I'm not so sure that math solves this problem. Logic and reason are more applicable and conference dual record should be the first and highest weighted criteria, then move on to secondary and tertiary criteria.

1. Conference dual record, if you're the only guy that is 8-0 you are the 1 seed, all other factors be damned.
2. Head to head is the first tie breaker
3. Record against common opponents and record against the field outside of B1G duals.
4. I'd go with coaches ranking as final tiebreaker though adding rpi and/or strength of schedule can be included.

Just a scenario...3rd and 4th seeds....one guy beat the other by tech. but also lost 2 SV to top 2 seeds. whereas the guy he teched didn't face them. One is 7-1, the other is 6-2. You rank 6-2 below the guy he teched?

 

[HikeNatParks]

Old hiker’s fix:
Every programmer on these threads knows that if actual coding is involved in the Wrestlestat data collection, one final path of logic fixes the glaring H2H misses. After temporary seeds are established based on straight numbers (current state), a top-to-bottom last pass loops through a comparison of adjacent seeds for any H2H. This roughs it out:

Set curr-seed = 1
Set next-seed = 2

While curr-seed < 14
... If seed-pts (curr-seed) – seed-pts (next-seed) < 16
...... And H2H-results-exist
...... And H2H-result (next-seed) > H2H-result (curr-seed)

...... Write Adjusted-file with lower wrestler as current seed
...... Reset higher wrestler as lower one to continue in comparisons
... Else
...... Write Adjusted-file with higher wrestler retaining seed
... End-if

... Add 1 to curr-seed
... Add 1 to next-seed
End-while

After the loop completes, the adjusted file holds the automated H2H fixes. This simple logic can be adapted to whatever computer language being used. To the B1G’s, I say you’re welcome.

 

[98lberEating2Lunches]

And don't forget though, that the current process has a last step - the black box - that we've haven't seen yet. But in advance of that, I'd add (5) Don't put limits on the black box like point differentials that would preclude the correction of egregious errors.

15 point differential limit is both arbitrary and capricious, and so doesn't meet administrative law basics.

 

[Aardvark86]

15 point differential limit is both arbitrary and capricious, and so doesn't meet administrative law basics.

You never know in a post-Loper Bright world...

 

[Dogwelder]

15 point differential limit is both arbitrary and capricious ..

And gormless and incognizant

 

[Nitlion1986]

And gormless and incognizant

Damn execrable post!!!!

 

[pish69]

Issue is there HAS to be a human component. Kyle Snyder came into tourney with 3 matches. He got 1 seed and deservedly so (he just won Olympics). Whatever matrix used would have him seeded 10th or worse. Who does that help? Certainly not the 2 seed who would get Snyder in Round 2

 

[pawrstlersinpa]

I have complained long and loud about the problem. Now it is time to attempt to be part of the solution.

I think there are three easy, common sense adjustments that can be made right away, and a fourth to think about.

1. Make H2H H2H. Drop (vs. field) as it is nonsensical. This solves two major problems. It removes a duplication in the formula, and it restores the single most important element to the seeding process. It also removes a potential gaming strategy that rewards teams for not using their starters in a conference dual.
2. Switch RPI and CR weightings. Quality wins (matches?) is very similar to RPI in that they measure the quality of the opponents faced (though in different ways). There is no need to have them both be weighted the same. Drop one down to 5% (could drop QW or RPI) and bump CR up to 10%.
3. Start with binary measures, finish with binary measures. The switch from binary comparisons (who had a better record? who won more categories?) to non-binary is bad math. No bad math.

One other thing that was discussed was the coaches' desire to emphasize showing up and wrestling in conference duals. That could be reinforced by tweaking the H2H a bit, too. If we make H2H a two tiered measure where H2H conference dual results are considered first, then if no conference result is available, H2H in non-conference competition is considered.

There is a potential problem with this, though. If two wrestlers meet at National Duals the winner has a dis-incentive to compete in a conference dual against the same wrestler. One way to solve this is to consider the H2H, for conference dual purposes only, at the team level. If Levi Haines beats whoever Iowa sends out at 174 in the dual, he is considered to have beaten whoever they send out at the conference tournament. Far from a perfect solution, but worth consideration.

Take heed, all you people who argued with me over the years about Wrestlestat rankings. "Let me see YOUR rankings and algorithm," I would say, only to be told, "Wrestlestat sucks." Never an attempt at a solution, just tell Wrestlestat to fix theirs. "Let them do the work." Now, we finally have someone stepping up to fix this B1G travesty. Still waiting on the other.

<rage bait complete>
<snicker>

 

[Nitlion1986]

Issue is there HAS to be a human component. Kyle Snyder came into tourney with 3 matches. He got 1 seed and deservedly so (he just won Olympics). Whatever matrix used would have him seeded 10th or worse. Who does that help? Certainly not the 2 seed who would get Snyder in Round 2

Just my thought, using the Kyle Snyder example as justification for not punishing non participation in dual conferences is somewhat disingenuous. There are very few examples of similar concerns. Hell, write an Olympic/senior World medal into the seeding considerations if necessary.
However if it was me and I had an 8-0, a 7-1, a 6-2 and a 6-0 scenario then first time through they would be seeded; 8-0 first, 7-1 second, 6-2 third and 6-0 fourth. If HTH results were available between third and fourth then I would with those 2 otherwise they woukd stay the same

 

[Dogwelder]

Take heed, all you people who argued with me over the years about Wrestlestat rankings. "Let me see YOUR rankings and algorithm," I would say, only to be told, "Wrestlestat sucks." Never an attempt at a solution, just tell Wrestlestat to fix theirs. "Let them do the work." Now, we finally have someone stepping up to fix this B1G travesty. Still waiting on the other.

<rage bait complete>
<snicker>

Right on! In the same spirit, I never just tell somebody they suck. I give them plenty of advice on how they can improve themselves. I tell them exactly which cosmetic surgeries they need, which types of self improvement courses they should take, which opinions never to express again, etc.

 

[pish69]

Just my thought, using the Kyle Snyder example as justification for not punishing non participation in dual conferences is somewhat disingenuous. There are very few examples of similar concerns. Hell, write an Olympic/senior World medal into the seeding considerations if necessary.
However if it was me and I had an 8-0, a 7-1, a 6-2 and a 6-0 scenario then first time through they would be seeded; 8-0 first, 7-1 second, 6-2 third and 6-0 fourth. If HTH results were available between third and fourth then I would with those 2 otherwise they woukd stay the same

The super talents should be punished but not as much as they are now.

Example..Angelo..no way he should be worse than 4th..same with Cannon and Peterson (I'm talking B1G seeds)

 

[Nitlion1986]

The super talents should be punished but not as much as they are now.

Example..Angelo..no way he should be worse than 4th..same with Cannon and Peterson (I'm talking B1G seeds)

I get what you are saying, but if we are going to insert subjective overrides for those "really good kids" then just have a coaches meeting. Everybody submit their lineup, every coach then submit a 1-14 ranking and tally up the votes and go from their.
If we are going to use as much objectivity as possible and Cannon ends up 7th and Ferrari gets 8th so be it. If they didn't wrestle then they haven't earned a higher seed.

 

[Aardvark86]

And gormless and incognizant

which is a contrast to WKN's very cromulent suggestions

 

[Wrestleknownothing]

which is a contrast to WKN's very cromulent suggestions

Take it back. I never lent anyone a crom in my life.

 

[Nitlion1986]

I get what you are saying, but if we are going to insert subjective overrides for those "really good kids" then just have a coaches meeting. Everybody submit their lineup, every coach then submit a 1-14 ranking and tally up the votes and go from their.
If we are going to use as much objectivity as possible and Cannon ends up 7th and Ferrari gets 8th so be it. If they didn't wrestle then they haven't earned a higher seed.

I get Rocco as the first seed earned an easier first match than against Ferrari, but Ferrari didn't earn an easier quarter-final match than the one against the first seed.

 

[CowbellMan]

I get what you are saying, but if we are going to insert subjective overrides for those "really good kids" then just have a coaches meeting. Everybody submit their lineup, every coach then submit a 1-14 ranking and tally up the votes and go from their.
If we are going to use as much objectivity as possible and Cannon ends up 7th and Ferrari gets 8th so be it. If they didn't wrestle then they haven't earned a higher seed.

Exactly my point in another post in another thread.

Seeding meetings aren’t rocket science. They just require some thought. Unfortunately the coaches wanted an easy way out and found ridiculousness standing right in their way.

If Wrestlestat couldn’t see a Snyder level problem in the algorithm, then it isn’t good enough. It should also have seen a Haines problem baked into it. It can’t handle injured studs either (difficult human consensus too).

Blaze/Byrd….is a philosophy issue.

presuming they are paying wrestlestat…..pay me. I’ll give them pre-seeds with Roar-like research and then they can argue something that is a lot closer to reality.

 

[Wrestleknownothing]

I just took a look at the seeds with regard to H2H.

• There are 21 instances where a wrestler lost the H2H to the wrestler seeded one spot below him.
• Only 141 had zero such instances.
• 174 had four such instances (#1Minto/#2Haines, #3Kennedy/#4Mantanona, #4Mantanona/#5Kharchla, #10Pinto/#11Enright)
• 157, 165, and 285 had three such instances each.
• While I did not keep count, much more often than not, the higher seeded wrestler also won the H2H.

 

[Nitlion1986]

Exactly my point in another post in another thread.

Seeding meetings aren’t rocket science. They just require some thought. Unfortunately the coaches wanted an easy way out and found ridiculousness standing right in their way.

If Wrestlestat couldn’t see a Snyder level problem in the algorithm, then it isn’t good enough. It should also have seen a Haines problem baked into it. It can’t handle injured studs either (difficult human consensus too).

Blaze/Byrd….is a philosophy issue.

I got a great chuckle out of the "in another post in another thread." My mind, which has very little discipline to begin with, went right to "that one time in band camp."
But yep. A group of guys coaching the hardest sport got spunk all over themselves looking for an easy way to deal with a time consuming pain in the asss part of their job.

 

[WV lion]

I just took a look at the seeds with regard to H2H.

• There are 21 instances where a wrestler lost the H2H to the wrestler seeded one spot below him.
• Only 141 had zero such instances.
• 174 had four such instances (#1Minto/#2Haines, #3Kennedy/#4Mantanona, #4Mantanona/#5Kharchla, #10Pinto/#11Enright)
• 157, 165, and 285 had three such instances each.
• While I did not keep count, much more often than not, the higher seeded wrestler also won the H2H.

Ridiculousness

 

[nerfstate]

I’m old and definitely out of touch with the times: I’m team human discernment! Put @RoarLions1, @WillieTheBrain and intermat Earl on a zoom and hash it out in a hour. Make it a live event!

Fans and coaches will be pissed regardless because the goalposts are never in one place for long. People want to “punish” Angelo Ferrari for missing matches until your guy draws him in his first bout. It’s always gonna be a tight rope, and I’d rather complain about Willie and his deep husker bias than @Wrestleknownothing’s amazing work.

 

[manatree]

1. Conference dual record, if you're the only guy that is 8-0 you are the 1 seed, all other factors be damned.
2. Head to head is the first tie breaker
3. Record against common opponents and record against the field outside of B1G duals.
4. I'd go with coaches ranking as final tiebreaker though adding rpi and/or strength of schedule can be included.

When ranking for a conference tournament, why should matches outside of the conference count for anything?

 

[Wrestleknownothing]

At work one of my commands is "no dead astronauts". That is a reference to "2010 The Year We Make Contact", an OK sequel to "2001: A Space Odyssey." In the movie we get an explanation for why HAL killed the crew of Discovery. It was given conflicting orders and the only way it could resolve the conflict was by killing the crew. Oops. Be real careful about the instructions you give these computers. Especially when they are sentient.

Some, not all, of the 21 examples referenced above represent dead astronauts. But rather than get rid of the version of HAL the Big Ten/Wrestlestat built, I say we just stop killing astronauts. The algo is fixable. Given that the vote was 12-1 (1 abstain) to try this, there is clearly demand to save the coaches some labor. Cael Sanderson even referenced that fact in his press availability.

Another guiding principle is that while there is often no one right answer, there are a whole lot of wrong answers. Focusing on a 1 of 1 situation, like Kyle Snyder wrestling an international and collegiate season simultaneously, is missing the forest for the trees. There is no right answer for Kyle Snyder, so stop worrying about solving for one. And since it is unlikely anyone will ever attempt that again, there isn't really much demand for a right answer either. Don't let that extreme outlier example be the reason to backslide to a manual process. Instead focus on eliminating the clearly wrong answers in the existing process.

 

[RoarLions1]

Just a thought, as I've never seen an uproar like this. Go back to the coaches providing rankings, same as last year and many years prior. Then let them duke it out on Friday for the few problematic situations. Not perfect, just better.

 

[Dogwelder]

… Don't let that extreme outlier example be the reason to backslide to a manual process. Instead focus on …

Yeah! Do not slide backward into manually crafted excellence! Instead, bravely march forward into a machine-made future in which the goal is that one day the machine-made product might be almost as good as the handmade product would be!

 

[Wrestleknownothing]

Yeah! Do not slide backward into manually crafted excellence! Instead, bravely march forward into a machine-made future in which the goal is that one day the machine-made product might be almost as good as the handmade product would be!

While we are at it lets get rid of all labor saving devices. What do you need a car for when a horse will do?

Don't forget the part where Sanderson says they spent too much time on the process.

 

[dicemen99]

15 point differential limit is both arbitrary and capricious, and so doesn't meet administrative law basics.

Your fly is open!!!!

 

[Dogwelder]

While we are at it lets get rid of all labor saving devices. What do you need a car for when a horse will do?

Don't forget the part where Sanderson says they spent too much time on the process.

The tail does not wag the dog. If you can save labor without killing the patient, then do it. If you’re gonna kill the patient, then just spend the labor until your technology is good enough. Nobody wants to see the nerds’ version 0.0 stuff. Hey nerds, get it right and then bother me with it. Until then, go back to your lab and stop killing patients.

 

[Wrestleknownothing]

The tail does not wag the dog. If you can save labor without killing the patient, then do it. If you’re gonna kill the patient, then just spend the labor until your technology is good enough. Nobody wants to see the nerds’ version 0.0 stuff. Hey nerds, get it right and then bother me with it. Until then, go back to your lab and stop killing patients.

Apparently the seeding algo is somewhere to the right of the schedule.

 

[pawrstlersinpa]

Right on! In the same spirit, I never just tell somebody they suck. I give them plenty of advice on how they can improve themselves. I tell them exactly which cosmetic surgeries they need, which types of self improvement courses they should take, which opinions never to express again, etc.

"Just make it so it matches my worldview (or so it matches the other 7 ranking services, so we'll have an eighth identical ranking)," was far and away the most frequent take.

 

[Wrestleknownothing]

"Just make it so it matches my worldview (or so it matches the other 7 ranking services, so we'll have an eighth identical ranking)," was far and away the most frequent take.

 

[Tryingtodoitright]

At work one of my commands is "no dead astronauts". That is a reference to "2010 The Year We Make Contact", an OK sequel to "2001: A Space Odyssey." In the movie we get an explanation for why HAL killed the crew of Discovery. It was given conflicting orders and the only way it could resolve the conflict was by killing the crew. Oops. Be real careful about the instructions you give these computers. Especially when they are sentient.

Some, not all, of the 21 examples referenced above represent dead astronauts. But rather than get rid of the version of HAL the Big Ten/Wrestlestat built, I say we just stop killing astronauts. The algo is fixable. Given that the vote was 12-1 (1 abstain) to try this, there is clearly demand to save the coaches some labor. Cael Sanderson even referenced that fact in his press availability.

Another guiding principle is that while there is often no one right answer, there are a whole lot of wrong answers. Focusing on a 1 of 1 situation, like Kyle Snyder wrestling an international and collegiate season simultaneously, is missing the forest for the trees. There is no right answer for Kyle Snyder, so stop worrying about solving for one. And since it is unlikely anyone will ever attempt that again, there isn't really much demand for a right answer either. Don't let that extreme outlier example be the reason to backslide to a manual process. Instead focus on eliminating the clearly wrong answers in the existing process.

We bought a car this past year. In my initial research I read somewhere that you should buy a car based on 95% of your needs, not the 5%. Just a thought.

 

[Coastal2]

I get Rocco as the first seed earned an easier first match than against Ferrari, but Ferrari didn't earn an easier quarter-final match than the one against the first seed.

But did Rowley from Purdue deserve to get a win or don't qualify match against Cartagena-Walsh in his first match. If he doesn't win, he'll get the Welsh/Ferrari loser and is probably going home 0-2 and has to hope for an at large.

At 157, the 8/9 Mechler v Buell match gets the Cannon/Duke loser on the backside.

 

[razpsu]

The super talents should be punished but not as much as they are now.

Example..Angelo..no way he should be worse than 4th..same with Cannon and Peterson (I'm talking B1G seeds)

Did they plan it this way?

 

[Psalm 1 guy]

Cael had this to say after the tournament, especially related to the 184 bracket:

“A great match,” Sanderson said. “Rocco had a heck of tough bracket there. I mean, the seeds, obviously, were really screwed up to start the tournament. It only hurt a couple of teams, so they didn’t do anything about it. But, I don’t think we’ll be going that route again, just because you want to do what’s right for the kids

 

[PortaJ]

Now it is time to attempt to be part of the solution.

Waited too long to join you over here and question the validity of your statistical approach