ADVERTISEMENT

Probability of Winning

Unbiased_football_fan

Unbiased_football_fan

Well-Known Member
Aug 18, 2006
77
741
1
Frederick, MD
Greetings All,

I made a probabilistic wrestling simulation model to project the NCAA tournament (I am a PSU engineering PhD). It was backtested on the last ten years of tournament data. The user enters all the seeds for the top ten teams, then the model simulates the tournament using Monte Carlo simulation with Latin Hypercube Sampling. The user can also enter advanced data like increases to the winning percentages for each seed (vs. every other seed) and increases to bonus percentages (vs. every other seed).

The model was able to give the exact order of finish (#1 - #5) many years and only teams predicted #1 or #2 have won. In fact, the team predicted #1 has won 8 out of 10 times. In 2015 Missouri was predicted #1 and finished a "miserable" fourth; tOSU was predicted #2 and finished #1; PSU was predicted #6 and finished #6.

Of course one can add up the projected points based on seeds, or if you are a bit more sophisticated add in bonus points. But doesn't it matter if you have the #1 seed at a weight when your competitor has the #3 at the same weight? What if that #3 guy pulls an upset? What is the probability a #3 beats a #1? How about a #10 vs. a #2? What if your guy is a super stud (Nolf, Retherford, Taylor, Ruth) compared to a "normal" #1? Modeling can account for that. I see lots of creative ways you may want to use this. Should I redshirt a guy? Should I burn a redshirt? What is the impact of an injury? I think the best use may be to create odds for Vegas and create more interest in wrestling.

Here are the results for this year (probability of winning only, expected order of finish under most scenarios is PSU #1, OkSt #2, tOSU #3, VT/Iowa flip a coin #4/#5. Numbers may not add up to 100 due to rounding. The model simulates all places so you can also get projected scores, score distribution, number of all americans, etc.

Standard data (no boosts to win or bonus):

PSU 77%
OkSt 16%
tOSU 4%
VT 2.4%
Iowa 1.2%

With bonus/win boost for studs (Most accurate, this is how much having a couple hammers matters):

PSU 97%
OkSt 2%
tOSU 1%
VT 0.2%

What if Suriano is a no go? With standard data:

PSU 54%
OkSt 27%
tOSU 11%
VT 5%
Iowa 3%
Mizz 1%


Suriano no go, with bonus boost:

PSU 87%
OkSt 8%
tOSU 4%
Iowa 1%
VT 0.6%

I saw a lot of debate that you must have X scorers, where X is 8 or 9. So I ran scenarios with a hypothetical team with two #1 studs, two other normal #1-2 guys, two #3 seeds, and four non-qualifiers. So basically a six man team. This team went against a hypothetical team of four #3 seeds, three #4 seeds, and three #5 seeds. The ultimate balanced lineup but without stars. The six man team beats the balanced team about two thirds of the time.

Hoping PSU rolls this weekend.
 
  • Like
Reactions: PSUoh90, NoVaLion2, LemonPie and 43 others
Unbiased_football_fan said:
Greetings All,

I made a probabilistic wrestling simulation model to project the NCAA tournament (I am a PSU engineering PhD). It was backtested on the last ten years of tournament data. The user enters all the seeds for the top ten teams, then the model simulates the tournament using Monte Carlo simulation with Latin Hypercube Sampling. The user can also enter advanced data like increases to the winning percentages for each seed (vs. every other seed) and increases to bonus percentages (vs. every other seed).

The model was able to give the exact order of finish (#1 - #5) many years and only teams predicted #1 or #2 have won. In fact, the team predicted #1 has won 8 out of 10 times. In 2015 Missouri was predicted #1 and finished a "miserable" fourth; tOSU was predicted #2 and finished #1; PSU was predicted #6 and finished #6.

Of course one can add up the projected points based on seeds, or if you are a bit more sophisticated add in bonus points. But doesn't it matter if you have the #1 seed at a weight when your competitor has the #3 at the same weight? What if that #3 guy pulls an upset? What is the probability a #3 beats a #1? How about a #10 vs. a #2? What if your guy is a super stud (Nolf, Retherford, Taylor, Ruth) compared to a "normal" #1? Modeling can account for that. I see lots of creative ways you may want to use this. Should I redshirt a guy? Should I burn a redshirt? What is the impact of an injury? I think the best use may be to create odds for Vegas and create more interest in wrestling.

Here are the results for this year (probability of winning only, expected order of finish under most scenarios is PSU #1, OkSt #2, tOSU #3, VT/Iowa flip a coin #4/#5. Numbers may not add up to 100 due to rounding. The model simulates all places so you can also get projected scores, score distribution, number of all americans, etc.

Standard data (no boosts to win or bonus):

PSU 77%
OkSt 16%
tOSU 4%
VT 2.4%
Iowa 1.2%

With bonus/win boost for studs (Most accurate, this is how much having a couple hammers matters):

PSU 97%
OkSt 2%
tOSU 1%
VT 0.2%

What if Suriano is a no go? With standard data:

PSU 54%
OkSt 27%
tOSU 11%
VT 5%
Iowa 3%
Mizz 1%


Suriano no go, with bonus boost:

PSU 87%
OkSt 8%
tOSU 4%
Iowa 1%
VT 0.6%

I saw a lot of debate that you must have X scorers, where X is 8 or 9. So I ran scenarios with a hypothetical team with two #1 studs, two other normal #1-2 guys, two #3 seeds, and four non-qualifiers. So basically a six man team. This team went against a hypothetical team of four #3 seeds, three #4 seeds, and three #5 seeds. The ultimate balanced lineup but without stars. The six man team beats the balanced team about two thirds of the time.

Hoping PSU rolls this weekend.
Click to expand...

Best first post EVER!!!??
 
  • Like
Reactions: PSUer1989, dieselmilk, stacyusaf and 6 others
Unbiased_football_fan said:
I made a probabilistic wrestling simulation model to project the NCAA tournament (I am a PSU engineering PhD). It was backtested on the last ten years of tournament data. The user enters all the seeds for the top ten teams, then the model simulates the tournament using Monte Carlo simulation with Latin Hypercube Sampling. The user can also enter advanced data like increases to the winning percentages for each seed (vs. every other seed) and increases to bonus percentages (vs. every other seed).
Click to expand...

always appreciate data-driven analysis, and you appear to have done a very nice job.

then again, it's sports, and things just do not always go according to the data. I doubt that any simulation model would have had Spencer Lee losing last evening.

models give general trends, and I like the trends that your model is predicting. Only time will tell how many surprises occur at Nationals, for PSU and for other teams/wrestlers.
 
  • Like
Reactions: lionlover and Creek Side
Outstanding UFF!

Any predictions of AA's by weight class?

I love the fact that it correctly picked the winner over eight of the last ten years. That's credibility and also shows that other factors have crept into the final results 20% of the time.

Haven't seen too many experts think it's anything but the Nits vs. the Pokies for the trophy.
 
Fantastic post & information.

Love it. Thanks for your contribution and great luck on your PhD studies. I'm an engineer myself so I enjoy math / analytical info on wrestling tourney scoring.

Please come back to post more often.
 
  • Like
Reactions: yekrut321 and liex26
Unbiased_football_fan said:
Greetings All,

I made a probabilistic wrestling simulation model to project the NCAA tournament (I am a PSU engineering PhD). It was backtested on the last ten years of tournament data. The user enters all the seeds for the top ten teams, then the model simulates the tournament using Monte Carlo simulation with Latin Hypercube Sampling. The user can also enter advanced data like increases to the winning percentages for each seed (vs. every other seed) and increases to bonus percentages (vs. every other seed).

The model was able to give the exact order of finish (#1 - #5) many years and only teams predicted #1 or #2 have won. In fact, the team predicted #1 has won 8 out of 10 times. In 2015 Missouri was predicted #1 and finished a "miserable" fourth; tOSU was predicted #2 and finished #1; PSU was predicted #6 and finished #6.

Of course one can add up the projected points based on seeds, or if you are a bit more sophisticated add in bonus points. But doesn't it matter if you have the #1 seed at a weight when your competitor has the #3 at the same weight? What if that #3 guy pulls an upset? What is the probability a #3 beats a #1? How about a #10 vs. a #2? What if your guy is a super stud (Nolf, Retherford, Taylor, Ruth) compared to a "normal" #1? Modeling can account for that. I see lots of creative ways you may want to use this. Should I redshirt a guy? Should I burn a redshirt? What is the impact of an injury? I think the best use may be to create odds for Vegas and create more interest in wrestling.

Here are the results for this year (probability of winning only, expected order of finish under most scenarios is PSU #1, OkSt #2, tOSU #3, VT/Iowa flip a coin #4/#5. Numbers may not add up to 100 due to rounding. The model simulates all places so you can also get projected scores, score distribution, number of all americans, etc.

Standard data (no boosts to win or bonus):

PSU 77%
OkSt 16%
tOSU 4%
VT 2.4%
Iowa 1.2%

With bonus/win boost for studs (Most accurate, this is how much having a couple hammers matters):

PSU 97%
OkSt 2%
tOSU 1%
VT 0.2%

What if Suriano is a no go? With standard data:

PSU 54%
OkSt 27%
tOSU 11%
VT 5%
Iowa 3%
Mizz 1%


Suriano no go, with bonus boost:

PSU 87%
OkSt 8%
tOSU 4%
Iowa 1%
VT 0.6%

I saw a lot of debate that you must have X scorers, where X is 8 or 9. So I ran scenarios with a hypothetical team with two #1 studs, two other normal #1-2 guys, two #3 seeds, and four non-qualifiers. So basically a six man team. This team went against a hypothetical team of four #3 seeds, three #4 seeds, and three #5 seeds. The ultimate balanced lineup but without stars. The six man team beats the balanced team about two thirds of the time.

Hoping PSU rolls this weekend.
Click to expand...

Wow, that's some pretty slick work - especially the backtesting which is the only thing that will really tell you what you need to know to assess reliability. Pretty much tells us what we intuitively know from looking at past Champions, but still pretty cool to have it quantified like this.
 
This is awesome stuff! I'd be interested to know how you adjusted bonus and win rate for the "super studs" and how you determined who those super studs are.
 
Unbiased_football_fan said:
Greetings All,

I made a probabilistic wrestling simulation model to project the NCAA tournament (I am a PSU engineering PhD). It was backtested on the last ten years of tournament data. The user enters all the seeds for the top ten teams, then the model simulates the tournament using Monte Carlo simulation with Latin Hypercube Sampling. The user can also enter advanced data like increases to the winning percentages for each seed (vs. every other seed) and increases to bonus percentages (vs. every other seed).

The model was able to give the exact order of finish (#1 - #5) many years and only teams predicted #1 or #2 have won. In fact, the team predicted #1 has won 8 out of 10 times. In 2015 Missouri was predicted #1 and finished a "miserable" fourth; tOSU was predicted #2 and finished #1; PSU was predicted #6 and finished #6.

Of course one can add up the projected points based on seeds, or if you are a bit more sophisticated add in bonus points. But doesn't it matter if you have the #1 seed at a weight when your competitor has the #3 at the same weight? What if that #3 guy pulls an upset? What is the probability a #3 beats a #1? How about a #10 vs. a #2? What if your guy is a super stud (Nolf, Retherford, Taylor, Ruth) compared to a "normal" #1? Modeling can account for that. I see lots of creative ways you may want to use this. Should I redshirt a guy? Should I burn a redshirt? What is the impact of an injury? I think the best use may be to create odds for Vegas and create more interest in wrestling.

Here are the results for this year (probability of winning only, expected order of finish under most scenarios is PSU #1, OkSt #2, tOSU #3, VT/Iowa flip a coin #4/#5. Numbers may not add up to 100 due to rounding. The model simulates all places so you can also get projected scores, score distribution, number of all americans, etc.

Standard data (no boosts to win or bonus):

PSU 77%
OkSt 16%
tOSU 4%
VT 2.4%
Iowa 1.2%

With bonus/win boost for studs (Most accurate, this is how much having a couple hammers matters):

PSU 97%
OkSt 2%
tOSU 1%
VT 0.2%

What if Suriano is a no go? With standard data:

PSU 54%
OkSt 27%
tOSU 11%
VT 5%
Iowa 3%
Mizz 1%


Suriano no go, with bonus boost:

PSU 87%
OkSt 8%
tOSU 4%
Iowa 1%
VT 0.6%

I saw a lot of debate that you must have X scorers, where X is 8 or 9. So I ran scenarios with a hypothetical team with two #1 studs, two other normal #1-2 guys, two #3 seeds, and four non-qualifiers. So basically a six man team. This team went against a hypothetical team of four #3 seeds, three #4 seeds, and three #5 seeds. The ultimate balanced lineup but without stars. The six man team beats the balanced team about two thirds of the time.

Hoping PSU rolls this weekend.
Click to expand...
Yes Sir. I agree. Lets get St. Louis rocking. As Cael always request during socials...Stay on those Ref's !!!!
 
Here is a more scientific analysis: Before the season begins 100%, after the B1Gs 33%.
 
giphy.gif

Yep. Checks out.
 
  • Like
Reactions: Creek Side, Ranger Dan, diggerpup and 2 others
Unbiased_football_fan said:
Greetings All,

I made a probabilistic wrestling simulation model to project the NCAA tournament (I am a PSU engineering PhD). It was backtested on the last ten years of tournament data. The user enters all the seeds for the top ten teams, then the model simulates the tournament using Monte Carlo simulation with Latin Hypercube Sampling. The user can also enter advanced data like increases to the winning percentages for each seed (vs. every other seed) and increases to bonus percentages (vs. every other seed).

The model was able to give the exact order of finish (#1 - #5) many years and only teams predicted #1 or #2 have won. In fact, the team predicted #1 has won 8 out of 10 times. In 2015 Missouri was predicted #1 and finished a "miserable" fourth; tOSU was predicted #2 and finished #1; PSU was predicted #6 and finished #6.

Of course one can add up the projected points based on seeds, or if you are a bit more sophisticated add in bonus points. But doesn't it matter if you have the #1 seed at a weight when your competitor has the #3 at the same weight? What if that #3 guy pulls an upset? What is the probability a #3 beats a #1? How about a #10 vs. a #2? What if your guy is a super stud (Nolf, Retherford, Taylor, Ruth) compared to a "normal" #1? Modeling can account for that. I see lots of creative ways you may want to use this. Should I redshirt a guy? Should I burn a redshirt? What is the impact of an injury? I think the best use may be to create odds for Vegas and create more interest in wrestling.

Here are the results for this year (probability of winning only, expected order of finish under most scenarios is PSU #1, OkSt #2, tOSU #3, VT/Iowa flip a coin #4/#5. Numbers may not add up to 100 due to rounding. The model simulates all places so you can also get projected scores, score distribution, number of all americans, etc.

Standard data (no boosts to win or bonus):

PSU 77%
OkSt 16%
tOSU 4%
VT 2.4%
Iowa 1.2%

With bonus/win boost for studs (Most accurate, this is how much having a couple hammers matters):

PSU 97%
OkSt 2%
tOSU 1%
VT 0.2%

What if Suriano is a no go? With standard data:

PSU 54%
OkSt 27%
tOSU 11%
VT 5%
Iowa 3%
Mizz 1%


Suriano no go, with bonus boost:

PSU 87%
OkSt 8%
tOSU 4%
Iowa 1%
VT 0.6%

I saw a lot of debate that you must have X scorers, where X is 8 or 9. So I ran scenarios with a hypothetical team with two #1 studs, two other normal #1-2 guys, two #3 seeds, and four non-qualifiers. So basically a six man team. This team went against a hypothetical team of four #3 seeds, three #4 seeds, and three #5 seeds. The ultimate balanced lineup but without stars. The six man team beats the balanced team about two thirds of the time.

Hoping PSU rolls this weekend.
Click to expand...
This is amazing stuff! I'll admit, after the presidential election I feel like throwing this kinda stuff out the window anymore but damn, this is a sweet read and be fun to go over a few times to think about it all.
 
I am a graduated PSU PhD (98'). I had an engineering class with Dave Hart and Troy Sunderland, both really nice guys. I do make models for a living, or in this case made a model for fun.

Why does it work? I think it is because of the law of large numbers.

To adjust the data on winning percentages and bonus percentages for a "stud", "hammer", "Spyker", whatever you want to call it is a slippery slope. The data suggests a #1 seed beats an unseeded wrestler 98% of the time. There is about a 2% chance of an upset/injury. A #1 seed beats a #2 seed around 70% of the time. Is Nolf's win percentage against Kemerer only 70%? I think it is higher based on the matches I watched. So I added a separate interface where the user could alter these values, but like I said it is a slippery slope and the sample size is small to make these subjective changes. Some real interesting stuff in the data. A normal #1 bonuses a #13 or higher guy at about a 62% rate. A normal #1 bonuses a #3-#4 at about a 15-20% rate. Is the Nolf's bonus rate against the #3-#4 going to be this low? I think not.

Cael can't coach - we hear that from certain rival fans. He gets top recruits then just sits back and let's them do their thing. I looked at how each team did relative to their predicted result for the last five years. Looks to me like Cael has his team performing pretty well relative to his peers.

PSU = +9
OkSt = +4
tOSU = +2
Cornell = -2
Iowa = -12

Dear Jammenz - Sadly Minnesota was not even worthy of entering the data for. Maybe in another ten years or so.
 
  • Like
Reactions: NoVaLion2, manmythlegend1, stacyusaf and 18 others
What the hell kind of message board do you think we're running here, that we use facts and stats and analysis?

Seriously, great stuff. You know you've set an expectation now, right?

Regarding Jammen -- how long until Nebraska is relevant?
 
  • Like
Reactions: manmythlegend1, CSauertiegPSU and diggerpup
Unbiased_football_fan said:
I am a graduated PSU PhD (98'). I had an engineering class with Dave Hart and Troy Sunderland, both really nice guys. I do make models for a living, or in this case made a model for fun.

Why does it work? I think it is because of the law of large numbers.

To adjust the data on winning percentages and bonus percentages for a "stud", "hammer", "Spyker", whatever you want to call it is a slippery slope. The data suggests a #1 seed beats an unseeded wrestler 98% of the time. There is about a 2% chance of an upset/injury. A #1 seed beats a #2 seed around 70% of the time. Is Nolf's win percentage against Kemerer only 70%? I think it is higher based on the matches I watched. So I added a separate interface where the user could alter these values, but like I said it is a slippery slope and the sample size is small to make these subjective changes. Some real interesting stuff in the data. A normal #1 bonuses a #13 or higher guy at about a 62% rate. A normal #1 bonuses a #3-#4 at about a 15-20% rate. Is the Nolf's bonus rate against the #3-#4 going to be this low? I think not.

Cael can't coach - we hear that from certain rival fans. He gets top recruits then just sits back and let's them do their thing. I looked at how each team did relative to their predicted result for the last five years. Looks to me like Cael has his team performing pretty well relative to his peers.

PSU = +9
OkSt = +4
tOSU = +2
Cornell = -2
Iowa = -12

Dear Jammenz - Sadly Minnesota was not even worthy of entering the data for. Maybe in another ten years or so.
Click to expand...
Iowa -12....Fact is the majority are sick and tired of Brands by the end of each season and want it to be over.....Not all but the majority which is sad....Just ask some of them....That is a terrible number....
 
Unbiased_football_fan said:
I am a graduated PSU PhD (98'). I had an engineering class with Dave Hart and Troy Sunderland, both really nice guys. I do make models for a living, or in this case made a model for fun.

Why does it work? I think it is because of the law of large numbers.

To adjust the data on winning percentages and bonus percentages for a "stud", "hammer", "Spyker", whatever you want to call it is a slippery slope. The data suggests a #1 seed beats an unseeded wrestler 98% of the time. There is about a 2% chance of an upset/injury. A #1 seed beats a #2 seed around 70% of the time. Is Nolf's win percentage against Kemerer only 70%? I think it is higher based on the matches I watched. So I added a separate interface where the user could alter these values, but like I said it is a slippery slope and the sample size is small to make these subjective changes. Some real interesting stuff in the data. A normal #1 bonuses a #13 or higher guy at about a 62% rate. A normal #1 bonuses a #3-#4 at about a 15-20% rate. Is the Nolf's bonus rate against the #3-#4 going to be this low? I think not.

Cael can't coach - we hear that from certain rival fans. He gets top recruits then just sits back and let's them do their thing. I looked at how each team did relative to their predicted result for the last five years. Looks to me like Cael has his team performing pretty well relative to his peers.

PSU = +9
OkSt = +4
tOSU = +2
Cornell = -2
Iowa = -12

Dear Jammenz - Sadly Minnesota was not even worthy of entering the data for. Maybe in another ten years or so.
Click to expand...
Unbiased - My son is an engineering student....Send me an e-mail please to touch base with you....
 
Unbiased_football_fan said:
Greetings All,

I made a probabilistic wrestling simulation model to project the NCAA tournament (I am a PSU engineering PhD). It was backtested on the last ten years of tournament data. The user enters all the seeds for the top ten teams, then the model simulates the tournament using Monte Carlo simulation with Latin Hypercube Sampling. The user can also enter advanced data like increases to the winning percentages for each seed (vs. every other seed) and increases to bonus percentages (vs. every other seed).

The model was able to give the exact order of finish (#1 - #5) many years and only teams predicted #1 or #2 have won. In fact, the team predicted #1 has won 8 out of 10 times. In 2015 Missouri was predicted #1 and finished a "miserable" fourth; tOSU was predicted #2 and finished #1; PSU was predicted #6 and finished #6.

Of course one can add up the projected points based on seeds, or if you are a bit more sophisticated add in bonus points. But doesn't it matter if you have the #1 seed at a weight when your competitor has the #3 at the same weight? What if that #3 guy pulls an upset? What is the probability a #3 beats a #1? How about a #10 vs. a #2? What if your guy is a super stud (Nolf, Retherford, Taylor, Ruth) compared to a "normal" #1? Modeling can account for that. I see lots of creative ways you may want to use this. Should I redshirt a guy? Should I burn a redshirt? What is the impact of an injury? I think the best use may be to create odds for Vegas and create more interest in wrestling.

Here are the results for this year (probability of winning only, expected order of finish under most scenarios is PSU #1, OkSt #2, tOSU #3, VT/Iowa flip a coin #4/#5. Numbers may not add up to 100 due to rounding. The model simulates all places so you can also get projected scores, score distribution, number of all americans, etc.

Standard data (no boosts to win or bonus):

PSU 77%
OkSt 16%
tOSU 4%
VT 2.4%
Iowa 1.2%

With bonus/win boost for studs (Most accurate, this is how much having a couple hammers matters):

PSU 97%
OkSt 2%
tOSU 1%
VT 0.2%

What if Suriano is a no go? With standard data:

PSU 54%
OkSt 27%
tOSU 11%
VT 5%
Iowa 3%
Mizz 1%


Suriano no go, with bonus boost:

PSU 87%
OkSt 8%
tOSU 4%
Iowa 1%
VT 0.6%

I saw a lot of debate that you must have X scorers, where X is 8 or 9. So I ran scenarios with a hypothetical team with two #1 studs, two other normal #1-2 guys, two #3 seeds, and four non-qualifiers. So basically a six man team. This team went against a hypothetical team of four #3 seeds, three #4 seeds, and three #5 seeds. The ultimate balanced lineup but without stars. The six man team beats the balanced team about two thirds of the time.

Hoping PSU rolls this weekend.
Click to expand...

I know this is a wrestling board, but I have this basketball bracket that I'm filling out.... can you help?
 
  • Like
Reactions: lionlover, Creek Side, Grove Lion and 1 other person
Unbiased_football_fan said:
...
I made a probabilistic wrestling simulation model ... It was backtested on the last ten years of tournament data ...
Click to expand...
Nice post UFF!

I have two questions:

One. When you backtested on each year of the ten years of data, were the probabilities in your model estimated using data that included that test year's data? In other words, are you testing on your training data, or are you testing on data not used in training (e.g., by jackknifing)?

Two. Do you have insight into why the model failed to predict the winner in those two years?
 
  • Like
Reactions: lionlover
Could you do the Board a favor and run it without Suriano and without Hall. Assuming that coaches might be into your model for coaching decisions, let's see what the data would show.
 
  • Like
Reactions: woodyk and Ranger Dan
ClarkstonMark said:
Could you do the Board a favor and run it without Suriano and without Hall. Assuming that coaches might be into your model for coaching decisions, let's see what the data would show.
Click to expand...
The "without Hall" would also need to be modified by adding back Morelli/Rasheed. I believe the problem would be is where they are seeded, which is the big factor in the calculations.

Else, I probably missed the sarcasm with no smiley face;)
 
CentexLion said:
The "without Hall" would also need to be modified by adding back Morelli/Rasheed. I believe the problem would be is where they are seeded, which is the big factor in the calculations.

Else, I probably missed the sarcasm with no smiley face;)
Click to expand...
No sarcasm. No 174 is the absolute worst case scenario. Just want to see that, and not start another pissing contest, if possible. :)
 
Dogwelder said:
...
In other words, are you testing on your training data, or are you testing on data not used in training (e.g., by jackknifing)?
...
Click to expand...
BTW, I ask because if we want to estimate how well your model will work for 2017, then we will want to know how well your model works on a year that is not already in your training data because, of course, 2017 has not happened yet and is not already in your training data.
 
Last edited:
  • Like
Reactions: lionlover
Tom McAndrew said:
always appreciate data-driven analysis, and you appear to have done a very nice job.

then again, it's sports, and things just do not always go according to the data. I doubt that any simulation model would have had Spencer Lee losing last evening.

models give general trends, and I like the trends that your model is predicting. Only time will tell how many surprises occur at Nationals, for PSU and for other teams/wrestlers.
Click to expand...
A well modeled Monte Carlo simulation would almost surely have predicted the possibility of Spencer Lee losing that match. In Monte Carlo simulation terms, the percentage of simulations in which Lee wins would be something less than 100%. Would it have predicted him losing that match? No, but that's not what they're intended to do.
 
  • Like
Reactions: CSauertiegPSU and CentexLion
Dogwelder said:
BTW, I ask because if we want to estimate how well your model will work for 2017, then we will want to know how well your model works on a year that is not already in your training data because, of course, 2017 has not happened yet and is not already in your training data.
Click to expand...

Thank you for your question. The generic win probabilities and bonus percentages for each seed vs. every other seed was generated from the last ten years of NCAA results. This isn't ideal but I was sensitive to the idea that with all the advanced training and clubs, etc. that kids have today and the limited scholarships that there may be more parity today and the winning percentage of a #1 vs. a #2 from 1973, for example, may have limited relevance. But, I think it is still okay to use all of those percentages to do the backtesting. Otherwise, there was no calibration. The model was built and then run for all the past years. Feel free to message me if you questions that the average guy might not want to read about.

Someone asked what would it look like with an injured Suriano and no Mark Hall. The results are below with Suriano assigned a 28 seed and Hall's replacement assigned a 17 seed. These are win percentages, because to quote Ricky Bobby "If you ain't first, your last."

PSU 50%
OkSt 40%
tOSU 12%
Iowa 1%
Others <1%

Looks like a pretty good idea to me to pull Hall's shirt.
 
  • Like
Reactions: stacyusaf, CSauertiegPSU, lionlover and 6 others
UFF your post to like ratio may never be emulated again and probably has a few around her a little wet in their skinny jeans. Theres a guy over on HR that has the reverse (>200 posts and 2 pathetic likes) and it has affected his online anonymous personality. Maybe someday you could design a way to help him out.
 
Last edited:
Over the last five years, the winner has averaged 120 points, 5-6 AA's, and 2 champs. Hmm, only one team Has that potential. PSU has also averaged 23 bonus points in winning four out the last five years which will once again be the difference.
 
  • Like
Reactions: diggerpup
Ian said:
Over the last five years, the winner has averaged 120 points, 5-6 AA's, and 2 champs. Hmm, only one team Has that potential. PSU has also averaged 23 bonus points in winning four out the last five years which will once again be the difference.
Click to expand...
Too pretentious for my tastes. I'll grant, we're favored, but several teams could win...that's why the matches are wrestled. Different dynamics, but look 2 weeks back.

My take is that the guys are hungry, and will wrestle well, embracing the chance to wrestle for their teammates and fans. Wrestle their best, and the score takes care of itself.
 
  • Like
Reactions: Maddog
Just to be clear, Dunke made a post about UFF's post-to-like ratio and about a guy over on HR, and it looked as if he were quoting a post of mine. But I never posted such a thing. Dunke must've made a simple reply and edit error.
 
  • Like
Reactions: lionlover
You must log in or register to reply here.

Similar threads

J

Big 10 Tourney wrestlestat simulation

Replies
9
Views
2K
The Wrestling Room
Nashville_Hawkeye
Nashville_Hawkeye
J

latest rankings and team points

Replies
21
Views
3K
The Wrestling Room
Pennstate1985
P
mcpat

The Second String Report

Replies
7
Views
2K
The Wrestling Room
mcpat
mcpat
I

Big10 Team Standing

Replies
8
Views
2K
The Wrestling Room
dpd926
D
J

B1G Stats and a few observations

Replies
24
Views
2K
The Wrestling Room
LongJakk
LongJakk
ADVERTISEMENT

Latest posts

ADVERTISEMENT
Top Bottom
Back