Chief Delphi

Chief Delphi (http://www.chiefdelphi.com/forums/index.php)
-   Extra Discussion (http://www.chiefdelphi.com/forums/forumdisplay.php?f=68)
-   -   pic: Top 10% OPRs compared via standard deviation distribution - 2008-2015 (http://www.chiefdelphi.com/forums/showthread.php?t=136227)

Citrus Dad 30-03-2015 15:44

pic: Top 10% OPRs compared via standard deviation distribution - 2008-2015
 

nuclearnerd 30-03-2015 15:51

Re: pic: Top 10% OPRs compared via standard deviation distribution - 2008-2015
 
Thanks for this. The chart really shows how the top 2-3% (maybe 50 teams) earn a much larger share of the points than the average team, more so than any other year (and a huge jump from 2014).

It's good to see that there doesn't appear to be a broader trend towards less equal OPR distributions over time. RR seems to be an outlier.

IKE 30-03-2015 16:01

Re: pic: Top 10% OPRs compared via standard deviation distribution - 2008-2015
 
Thanks for posting. I have made of looked at this sort of trend for a handful of years, but use a different normalizing function.

To a certain degree, I expect that the 2015 data might decrease a bit thanks to weeks 5&6. There will be more teams playing their 2nd event during those weeks which will raise the median pervormance quite a bit.

JamesBrown 30-03-2015 16:30

Re: pic: Top 10% OPRs compared via standard deviation distribution - 2008-2015
 
Quote:

Originally Posted by IKE (Post 1463898)
Thanks for posting. I have made of looked at this sort of trend for a handful of years, but use a different normalizing function.

To a certain degree, I expect that the 2015 data might decrease a bit thanks to weeks 5&6. There will be more teams playing their 2nd event during those weeks which will raise the median pervormance quite a bit.

It is possible, but I think the real driving force this year is the lack of defense. I think the top teams have always been this much better at scoring than everyone else, however facing regular defense from other teams dragged down their average scores significantly.

nuclearnerd 30-03-2015 16:41

Re: pic: Top 10% OPRs compared via standard deviation distribution - 2008-2015
 
Here's another way of looking at some of the same data. I'm just showing histograms of OPRs between 2015 (the most skewed game) and 2014 (one of the least skewed games), normalized by standard deviation as Richard has done:
https://drive.google.com/file/d/0B16...ew?usp=sharing

The difference between the games isn't as stark as I thought. Certainly there were more people with OPRs in the 0.5 to 1.5 SD range in 2014 than there are this year, but the curves are otherwise remarkably similar. That said, 1114's placement way out at 6.25 SD looks all the more impressive on this chart!

Citrus Dad 30-03-2015 18:45

Re: pic: Top 10% OPRs compared via standard deviation distribution - 2008-2015
 
Quote:

Originally Posted by nuclearnerd (Post 1463929)
Here's another way of looking at some of the same data. I'm just showing histograms of OPRs between 2015 (the most skewed game) and 2014 (one of the least skewed games), normalized by standard deviation as Richard has done:
https://drive.google.com/file/d/0B16...ew?usp=sharing

The difference between the games isn't as stark as I thought. Certainly there were more people with OPRs in the 0.5 to 1.5 SD range in 2014 than there are this year, but the curves are otherwise remarkably similar. That said, 1114's placement way out at 6.25 SD looks all the more impressive on this chart!

The 2015 distribution is flatter and more spread than I expected with extended tails at both ends. The number of teams beyond the top of 2014 is interesting as well.

Citrus Dad 30-03-2015 18:49

Re: pic: Top 10% OPRs compared via standard deviation distribution - 2008-2015
 
Quote:

Originally Posted by JamesBrown (Post 1463925)
It is possible, but I think the real driving force this year is the lack of defense. I think the top teams have always been this much better at scoring than everyone else, however facing regular defense from other teams dragged down their average scores significantly.

I would go further to say its in part a function of the lack of interdependency, both offense & defense. Our highest score came with only 2 robots on the field (although we would have gotten a higher score if our robot had grabbed a 2nd center can in one of the Sacramento finals with 3 bots on the field.)

But that's further enhanced by the technical challenge of capping stacks which is akin to climbing beyond the first level of the pyramid in 2013. Not many teams could do that back then either.

Abhishek R 31-03-2015 09:01

Re: pic: Top 10% OPRs compared via standard deviation distribution - 2008-2015
 
Quote:

Originally Posted by Citrus Dad (Post 1463995)
I would go further to say its in part a function of the lack of interdependency, both offense & defense. Our highest score came with only 2 robots on the field (although we would have gotten a higher score if our robot had grabbed a 2nd center can in one of the Sacramento finals with 3 bots on the field.)

But that's further enhanced by the technical challenge of capping stacks which is akin to climbing beyond the first level of the pyramid in 2013. Not many teams could do that back then either.

Yup, we're back to a game where a single team can score a massive amount of points by itself, as opposed to something like 2014, where even without defense teams relied on each other to get those assists if they really wanted to score high, and it was sort of a limiting factor.

nuclearnerd 31-03-2015 11:04

Re: pic: Top 10% OPRs compared via standard deviation distribution - 2008-2015
 
Here are the two histograms again, but this time I've clipped the top of the scale so we can focus on the "tails" (the 50 or so teams at the top and bottom of the distribution). I also screwed up the last plot and didn't show 2056 with an OPR at ~6 SD above the mean. 2056 and 1114 look lonely out there :)

https://drive.google.com/file/d/0B16...ew?usp=sharing

Citrus Dad 31-03-2015 17:16

Re: pic: Top 10% OPRs compared via standard deviation distribution - 2008-2015
 
Quote:

Originally Posted by nuclearnerd (Post 1464264)
Here are the two histograms again, but this time I've clipped the top of the scale so we can focus on the "tails" (the 50 or so teams at the top and bottom of the distribution). I also screwed up the last plot and didn't show 2056 with an OPR at ~6 SD above the mean. 2056 and 1114 look lonely out there :)

https://drive.google.com/file/d/0B16...ew?usp=sharing

What's absurd is that there are 60 teams BELOW the bottom of last year's distribution.

Caleb Sykes 31-03-2015 22:09

Re: pic: Top 10% OPRs compared via standard deviation distribution - 2008-2015
 
Quote:

Originally Posted by Citrus Dad (Post 1464515)
What's absurd is that there are 60 teams BELOW the bottom of last year's distribution.

Not really, it is easier to accidentally hurt your own alliance this year than it was last year.

Woolly 31-03-2015 22:15

Re: pic: Top 10% OPRs compared via standard deviation distribution - 2008-2015
 
Quote:

Originally Posted by Caleb Sykes (Post 1464653)
Not really, it is easier to accidentally hurt your own alliance this year than it was last year.

G40 would like a word with you.

Caleb Sykes 31-03-2015 23:27

Re: pic: Top 10% OPRs compared via standard deviation distribution - 2008-2015
 
Quote:

Originally Posted by Woolly (Post 1464658)
G40 would like a word with you.

You're absolutely right, I had forgotten about penalties. I wonder how these graphs would compare if the OPRs from last year incorporated penalties.

Pretzel 31-03-2015 23:54

Re: pic: Top 10% OPRs compared via standard deviation distribution - 2008-2015
 
Quote:

Originally Posted by Woolly (Post 1464658)
G40 would like a word with you.

G40 didn't reduce your OPR though. It gave points to your opponents instead of taking away your own hard-earned points. You also were unable to de-score yourself by knocking over stacks, since it was pretty hard to get the ball to go backwards back through the goal after your scored.

Citrus Dad 01-04-2015 16:09

Re: pic: Top 10% OPRs compared via standard deviation distribution - 2008-2015
 
Quote:

Originally Posted by Caleb Sykes (Post 1464685)
You're absolutely right, I had forgotten about penalties. I wonder how these graphs would compare if the OPRs from last year incorporated penalties.

Pretty sure last year's OPR included penalties, as with all previous years. The OPR calculation is simply a regression analysis on the total final score for each match with a dummy variable for each team.

So I go back to my original comment that the low end spread is quite interesting. I think it might be reflective of how difficult it is for a newer or less experienced team to contribute to the game, but I didn't think they could detract so much.

BTW, there are special statistical properties to include when running regressions with a continuous dependent variable (score) and 0-1 dummy variables (i.e., whether a team is present on the alliance). I haven't looked at that issue for quite a while so I don't remember much beyond that but it is a consideration in the OPR estimation.

Caleb Sykes 01-04-2015 16:30

Re: pic: Top 10% OPRs compared via standard deviation distribution - 2008-2015
 
Quote:

Originally Posted by Citrus Dad (Post 1464926)
Pretty sure last year's OPR included penalties, as with all previous years. The OPR calculation is simply a regression analysis on the total final score for each match with a dummy variable for each team.

So I go back to my original comment that the low end spread is quite interesting. I think it might be reflective of how difficult it is for a newer or less experienced team to contribute to the game, but I didn't think they could detract so much.

BTW, there are special statistical properties to include when running regressions with a continuous dependent variable (score) and 0-1 dummy variables (i.e., whether a team is present on the alliance). I haven't looked at that issue for quite a while so I don't remember much beyond that but it is a consideration in the OPR estimation.

I should have been more clear. What I meant to convey was that penalties last year had no direct impact on the offending team's score. Thus, OPR would have no way to tell that a team was hurting their alliance because of penalties. Since penalties this year are subtractive, they are reflected in the OPR calculations.

What I am curious about is an alternate OPR calculation for last year which would use the following formula in the match score matrix instead of the nominal score:
adjusted score = (nominal score) - (penalty points incurred by opposing alliance) + (penalty points incurred by this alliance)

Actually, I do remember uOPR (unpenalized OPR) being calculated last year. Was this the method of obtaining that? Or did it just subtract out all penalty points incurred without adding any back in?

Ether 01-04-2015 18:10

Re: pic: Top 10% OPRs compared via standard deviation distribution - 2008-2015
 
Quote:

Originally Posted by Caleb Sykes (Post 1464939)
What I am curious about is an alternate OPR calculation for last year which would use the following formula in the match score matrix instead of the nominal score:
adjusted score = (nominal score) - (penalty points incurred by opposing alliance) + (penalty points incurred by this alliance)

Actually, I do remember uOPR (unpenalized OPR) being calculated last year. Was this the method of obtaining that? Or did it just subtract out all penalty points incurred without adding any back in?

Is this at all helpful?
http://www.chiefdelphi.com/media/papers/download/4037

Column headings are as follows:
Code:



L2s        Final Score OPR
L2f        Foul OPR
L2h        Hybrid OPR
L2t        TeleOp OPR
L2os        Opponent's Final Score OPR
L2of        Opponent's Foul OPR
L2oh        Opponent's Hybrid OPR
L2ot        Opponent's TeleOp OPR
As        Average Final Score
Af        Average Foul
Ah        Average Hybrid
At        Average TeleOp
Aos        Opponents Average Final Score
Aof        Opponents Average Foul
Aoh        Opponents Average Hybrid
Aot        Opponents Average TeleOp
uOPR        unpenalized OPR = L2s - L2f
uDPR        unpenalized DPR = L2os - L2of
uCCWM        unpenalized CCWM = uOPR - uDPR
EPA        uOPR - (As - Af)
M        Matches Played

N.B. : The usual Twitter data caveats apply


Conor Ryan 01-04-2015 18:49

Re: pic: Top 10% OPRs compared via standard deviation distribution - 2008-2015
 
There are more teams this year than ever before. The 1% is getting bigger and more elite ever year.

More than anything I think this statistic is showing the complete absence of defense from this years game.

smurfgirl 01-04-2015 18:53

Re: pic: Top 10% OPRs compared via standard deviation distribution - 2008-2015
 
Thanks for sharing this - it's really interesting!

Caleb Sykes 01-04-2015 22:29

Re: pic: Top 10% OPRs compared via standard deviation distribution - 2008-2015
 
Quote:

Originally Posted by Ether (Post 1464972)
Is this at all helpful?

Thank you, that is indeed very helpful. I will play around more with this after my team competes this weekend.

Citrus Dad 02-04-2015 16:48

Re: pic: Top 10% OPRs compared via standard deviation distribution - 2008-2015
 
Quote:

Originally Posted by Conor Ryan (Post 1464982)
There are more teams this year than ever before. The 1% is getting bigger and more elite ever year.

Actually 2014 was one of the most flat years in distribution so there's not an obvious trend. The SD calculation normalizes for the size of the pool of teams--it should be the same regardless of whether there are 100 or 10,000 teams. The difference in the distribution is really driven by the nature of the game.

Quote:

Originally Posted by Conor Ryan (Post 1464982)
More than anything I think this statistic is showing the complete absence of defense from this years game.

See my points above: It's not only lack of defense but the particular difficulty of gaining the highest point activities. There's been posts on that issue in other threads.


All times are GMT -5. The time now is 01:30.

Powered by vBulletin® Version 3.6.4
Copyright ©2000 - 2017, Jelsoft Enterprises Ltd.
Copyright © Chief Delphi