2019 ACF Regionals detailed stats

Elaborate on the merits of specific tournaments or have general theoretical discussion here.
Post Reply
User avatar
women, fire and dangerous things
Tidus
Posts: 604
Joined: Mon Feb 16, 2009 5:34 pm
Location: Örkko, Cimmeria

2019 ACF Regionals detailed stats

Post by women, fire and dangerous things » Wed Jan 30, 2019 10:54 am

The every.buzz detailed stats aren't quite ready, but I'm releasing the detailed stats spreadsheets for each individual site and for the combined sites. Please read all the information below before you view the spreadsheets.

Overview

The aim of detailed stats is to provide more data than standard stats, which only collects summary data (player statlines and final team scores) for each game. Once each question is tagged and each complete scoresheet is entered, it becomes possible to compile aggregate data by question, category, team, or player. This form of detailed stats is usually called conversion stats.

The Categories and Authors sheets list aggregate tossup and bonus conversion by subcategory and author. (In this case, since it's a packet sub tournament, there are no author tags so the Authors sheet is empty.) Tossups and Bonuses list conversion for every question. The performance of each team by subcategory is in Teams–Categories. The performance of each player is in Player, which is broken down by category and subcategory in the remaining tabs.

Because this is a large, complicated spreadsheet, I recommend not leaving it open in your browser while you are not viewing it, as it may cause your computer (or others’) to slow down.

Spoilers

The Tossups and Bonuses sheets contain every answer in the set. If you do not want to be spoiled on packets that you haven't heard yet, be careful!

Errata

Please do not report corrections or problems accessing the spreadsheet in this thread; contact Ophir instead. See http://minkowski.space/quizbowl/manuals ... layer.html for details.

Links to the spreadsheets

All sites combined ·
CMU ·
Duke ·
Georgia Tech ·
Kansas State ·
Minnesota ·
MIT ·
Oxford-Brookes ·
Stanford ·
Swarthmore ·
Toronto ·
UCF ·
UIUC ·
UT-Austin
Will Nediger
-Proud member of the cult of Urcuchillay-
University of Western Ontario 2011, University of Michigan 2017
Emeritus member, ACF
Writer, NAQT

Iamteehee
Wakka
Posts: 100
Joined: Sun Sep 10, 2017 3:05 pm

Re: 2019 ACF Regionals detailed stats

Post by Iamteehee » Wed Jan 30, 2019 6:35 pm

Here are the BPA stats for 2019 ACF Regionals that Ryan Rosenberg (not me!) made.
Attachments
player_bpa_by_cat.xlsx
(67.72 KiB) Downloaded 253 times
player_bpa.xlsx
(21.59 KiB) Downloaded 171 times
team_bpa_by_cat.xlsx
(21.77 KiB) Downloaded 126 times
Geoffrey "with a G" Chen
Wayzata High School '19
UMN (dual enrollment) '19
Cornell '23

Iamteehee
Wakka
Posts: 100
Joined: Sun Sep 10, 2017 3:05 pm

Re: 2019 ACF Regionals detailed stats

Post by Iamteehee » Wed Jan 30, 2019 6:36 pm

The forums won't let me attach more than 3 files it seems, here's the team_bpa
Attachments
team_bpa.xlsx
(9.17 KiB) Downloaded 131 times
Geoffrey "with a G" Chen
Wayzata High School '19
UMN (dual enrollment) '19
Cornell '23

User avatar
hftf
Lulu
Posts: 54
Joined: Sat Apr 26, 2008 7:28 pm

Re: 2019 ACF Regionals detailed stats

Post by hftf » Thu Jan 31, 2019 12:45 am

Sorry for the long wait. Static visualizations, online packets, and raw data for tossups and bonuses are now available.


If you find this tool valuable and would like to support me creating more quizbowl technology, please consider donating to my Patreon.

Special thanks to all my supporters on Patreon: Alex Damisch, Alston Boyd, Aseem Keyal, Auroni Gupta, Cody Voight, Emmett Laurie, Eric Mukherjee, Evan Lynch, Fred Morlan, Hidehiro Anto, Jakob Myers, Joe Su, Jordan Brownstein, Kevin Wang, Matthew Bollinger, Matthew Lehmann, Mike Etzkorn, Najwa Watson, Oliver Clarke, Rob Carson, Ryan Rosenberg, Tejas Raje, Victor Prieto, Vishwa Shanmugam, Will Alston, Will Nediger, William Golden.
Ophir
Maryland
Walter Johnson
I use email.

User avatar
CPiGuy
Tidus
Posts: 612
Joined: Wed Nov 16, 2016 8:19 pm
Location: Ann Arbor, MI

Re: 2019 ACF Regionals detailed stats

Post by CPiGuy » Thu Jan 31, 2019 12:53 am

hftf wrote:
Thu Jan 31, 2019 12:45 am
Sorry for the long wait. Static visualizations, online packets, and raw data for tossups and bonuses are now available.


If you find this tool valuable and would like to support me creating more quizbowl technology, please consider donating to my Patreon.

Special thanks to all my supporters on Patreon: Alex Damisch, Alston Boyd, Aseem Keyal, Auroni Gupta, Cody Voight, Emmett Laurie, Eric Mukherjee, Evan Lynch, Fred Morlan, Hidehiro Anto, Jakob Myers, Joe Su, Jordan Brownstein, Kevin Wang, Matthew Bollinger, Matthew Lehmann, Mike Etzkorn, Najwa Watson, Oliver Clarke, Rob Carson, Ryan Rosenberg, Tejas Raje, Victor Prieto, Vishwa Shanmugam, Will Alston, Will Nediger, William Golden.
Thanks as always for the great visualizations!

The static visualizations appear not to include stats from the CMU site.
Conor Thompson
Bangor HS (Maine) '16
Michigan '20

User avatar
Smuttynose Island
Forums Staff: Moderator
Posts: 490
Joined: Wed Oct 21, 2009 9:07 pm

Re: 2019 ACF Regionals detailed stats

Post by Smuttynose Island » Thu Mar 07, 2019 4:45 pm

I was inspired by Fred's post on the purpose of first line clues to revisit this year's ACF Regionals data. Below you'll find a hodgepodge of results along with comments.

Highest Neg% (Min ~15 rooms heard)
  1. Volsunga Saga - .368421
  2. Uyghers - .3666
  3. saddle points - .365591
  4. Uranus - .358025
  5. Hitler Youth - .353659
These are the tossups where negs made up the highest percent of total buzzes. So it includes bouncebacks. You could also find the TUs with the highest percent of first buzzes being negs. The results would be a little different.


Latest Correct Buzz Location% (Min ~5 rooms heard)
  1. enthalpy of vaporization - .947286
  2. diatomic molecules - .942840
  3. viral capsid - .938238
  4. Maori - .929250
  5. halakha - .927781
In order to perform this computation, I pruned the data by eliminating all data points whose buzz location percentage was greater than 1 or less than 0. Using Ophir's excellent data visualization, its clear that a TU can have a late correct buzz location% for several reasons. The most obvious is that the TU is hard. The Maori mythology TU is a great example of this phenomenon. There were 44 total buzzes on this TU. Of the 32 correct buzzes, all of them were in the last fifth(!) of the TU. Of course, one reason why such a large percent of correct buzzes were late is that many of the early buzzes were negs. This is the other big reason why a TU may appear on this list. If a lot of people neg on the early clues, there are going to be more bouncebacks and fewer early gets. Because of this confounding explanation, it might be better to look at the TUs with the latest first buzz location%.

Least Area Under the Curve of Correct Buzz Location% excluding Bouncebacks CDF (Min ~15 rooms heard)
  1. diatomic molecules - .076441
  2. viral capsid - .082125
  3. halakha - .089962
  4. Aram Khachaturian - .090133
  5. the seven deadly sins - .0913333
  6. river gods - .098937
Unlike the other lists, this list requires a bit more explanation. A good primer on the intuition behind area under the curve is Ryan Rosenberg's excellent BPA post.To find these TUs, I examined the cumulative distribution function (cdf) of correct buzzes on a given TU excluding bouncebacks. The more often a question was correctly answered late, the more squished this cdf is to the right and the smaller its area is. The "seven deadly sins" tossup is a great example of this. If you look at Ophir's data visualization for this TU, you see that many people answered this question late. There are some drawbacks to this computation though. As you can imagine, this statistic is correlated to how often a TU was negged. The more often a TU was negged, the more bouncebacks at the end of the TUs there will be. Additionally, if an early clue was misleading many players who otherwise would have gotten the question early will have negged instead. This will push the cdf of the TU to the right. The "diatomic molecule" tossup is a good example of this phenomenon. About 20% of buzzes on this tossup were negs! This led to a lot of bouncebacks and missed opportunities for early gets. Another drawback is that this statist does not measure how often a tossup went dead. Still it gives editors a very quick way to identify potential problem tossups. Tossups that score low on this statistic are likely to be too hard or to be too misleading. With that being said, I don't believe it is the best way to measure how hard a TU is. I'm working on devising a statistic that does a better job of this and I'll post those results when I am done.

As always, I include a warning. There's a good chance that I messed up some computation somewhere.
Daniel Hothem
TJHSST '11 | UVA '15 | Oregon '??
"You are the stuff of legends" - Chris Manners
https://sites.google.com/site/academicc ... ubuva/home

Post Reply