Entry #8

NGADM 2017 R2 Stats Analysis!

2017-08-19 19:15:58 by LunacyEcho

Alright guys, let's jump into it! It's time to analyze the tracks and their scores from Round 2 of the NGADM. 

First things first: Here's a list of the overall average scores of everyone from Round 1, sorted from highest to lowest. (Bolded and italicized names won their brackets.)

-----------------------------------------------------------

01. 9.510 | FinnMK
02. 9.210 | bassfiddlejones + joshgawaldomusic
03. 9.092 | etherealwinds
04. 8.824 | johnfn + ethansight
05. 8.744 | JacobCadmus
06. 8.724 | Phonometrologist
07. 8.700 | CloakedSoup
08. 8.680 | garlagan
09. 8.664 | Metallica1136
10. 8.626 | McGorilla42
11. 8.552 | ThaPredator
12. 8.550 | 1f1n1ty
13. 8.548 | MrKoolTrix
14. 8.462 | Xtrullor + HeliXiX
15. 8.436 | Adjeye
16. 8.310 | Spadezer
17. 8.264 | RetromanOMG
18. 8.156 | LucidShadowDreamer + SnowTeddy
19. 8.134 | Noisysundae
20. 8.128 | ConnorGrail
21. 8.094 | Demon-Wolf
22. 8.056 | KaixoMusic
23. 7.966 | 5TanLey
24. 7.960 | steelside
25. 7.890 | midimachine
26. 7.820 | TolanMusic
27. 7.662 | Techmo-X
28. 7.626 | jimmypig
29. 7.190 | Veronina

-----------------------------------------------------------

Congrats to FinnMK for the highest-scoring (and my personal favorite) track of Round 2! To be honest, I think this distribution itself is pretty cool. I'd like to direct your attention especially towards ranks 11-13, where ThaPredator and MrKoolTrix had such a close margin between them that the scores had to hit the thousandths place to identify a winner... and yet, 1f1n1ty STILL somehow managed to squeeze a score in between theirs. What a narrow matchup!

To understand what would happen if the scores were normalized, we need to first take a look at each judge's mean and standard deviation (for a better explanation of what this means, check out my post for Round 1). 

Judge | Mean | Standard Deviation

Bosa | 8.363 | 0.643
ChronoNomad | 8.490 | 0.594
ADR3-N | 8.479 | 0.608
samulis | 8.034 | 0.579
TaintedLogic | 8.457 | 0.829

Looking a bit low there, samulis! But that just goes to show how different people's scales can be. Let's see what happens if we normalize the scores to an overall mean of 8.0 and a standard deviation of 1.0 based on this data!

-----------------------------------------------------------

Rank. Score | Artist | Change in Position

01. 9.767 | FinnMK | -
02. 9.311 | bassfiddlejones + joshgawaldomusic | -
03. 9.157 | etherealwinds | -
04. 8.701 | johnfn + ethansight | -
05. 8.615 | Phonometrologist | +1
06. 8.533 | Metallica1136 | +3
07. 8.500 | JacobCadmus | -2
08. 8.479 | CloakedSoup | -1
09. 8.461 | garlagan | -1
10. 8.317 | ThaPredator | +1
11. 8.315 | McGorilla42 | -1
12. 8.287 | 1f1n1ty | -
13. 8.273 | MrKoolTrix | -
14. 8.158 | Adjeye | +1
15. 8.073 | Xtrullor + HeliXiX | -1
16. 7.876 | Spadezer | -
17. 7.810 | RetromanOMG | -
18. 7.647 | ConnorGrail | +2
19. 7.643 | LucidShadowDreamer + SnowTeddy | -1
20. 7.624 | Noisysundae | -1
21. 7.591 | KaixoMusic | +1
22. 7.569 | Demon-Wolf | -1
23. 7.452 | 5TanLey | -
24. 7.409 | steelside | -
25. 7.258 | midimachine | -
26. 7.171 | TolanMusic | -
27. 6.904 | Techmo-X | -
28. 6.892 | jimmypig | -
29. 6.207 | Veronina | -

-----------------------------------------------------------

In the end, none of the brackets ended up changing, which means that there is further evidence that all the winners were the right ones! But this is quite lucky, given that half of the musicians had rank changes post-normalization. Interestingly, the scores themselves changed a lot more this time around after normalization than they did for round one—probably because the standard of having a mean of 8.0 and a st.dev of 1.0 was calibrated towards the scores of the first round. In hindsight, I probably should've set a different normalization standard, since the judge stats changed as well. But that's a lesson for next round, I guess!

This time, I also did a more comprehensive analysis of which judges were able to predict the winners correctly. Again, no one actually got all the brackets right, but Bosa came insanely close, with only one incorrect bracket! (He preferred TolanMusic's quaint tunes over Demon-Wolf's glitch rock.) Of the 13 contestable brackets to guess from, Bosa got 12, Chrono got 11, TaintedLogic and samulis both got 10, and ADR3-N got 8. 

Since we had five judges, all brackets were able to have a majority vote, and in all cases (well, except a special one), the majority won. However, the most contested brackets—the ones that went 3 votes to 2—were between CloakedSoup & adjeye, Techmo-X & jimmypig, and RetromanOMG & LucidShadowDreamer + SnowTeddy. However, there was one special bracket. In this one, two judges voted for one side, two voted for the other, ...and samulis gave them both the same score. This "tie between five judges" was none other than ThaPredator vs MrKoolTrix, which also happened to be the closest bracket of the round. Go figure!

Anyways, that's all I have to say for this round. If you want to play around with the numbers yourself, be my guest! You can find the spreadsheet I used here. Thanks again to @gadogry for the statspost inspiration and to @ChronoNomad for organizing the whole shebang. Now get out there and make some more awesome music, y'all!

 


Comments

You must be logged in to comment on this post.


TaintedLogicTaintedLogic

2017-08-19 20:05:37

I appreciate your efforts yet again, LunacyEcho! :)


ChronoNomadChronoNomad

2017-08-19 22:00:59

Holy shi--

--ba inu, Batman! Those are some darn fine statistical statistics, Mr. Echo. Quite interesting yet again, and obviously a fair amount of work. Happily, just like our job as judges gets easier as the NGADM moves closer to the endgame, so too will this.

Huzzah!


McGorilla42McGorilla42

2017-08-20 01:50:46

You're actually the best. I love these stats and appreciate you taking the tim to share this with us! :)