Week4 Twitter data analyses

[Ether]

This thread is for discussion of Week4 Twitter data analyses posted here:

http://www.chiefdelphi.com/media/papers/2985

I will be adding additional analyses as time permits. Special requests will be considered also.

[mmaunu]

Let me start by saying thanks for all that you do for this community...I have grown tremendously over the last several years reading your posts.

What is uCCWM? CCWM without foul points?

[Navid Shafa]

I really like EPA. Certainly more useful for cross event comparisons than raw OPR/CCWM. Wish we could tie in average event score or something of the like to EPA. Being able to view OPR separate from fouls and EPA certainly helps remove outliers from the data.

Thanks for the data as usual. :)

[Ether]

What is uCCWM? CCWM without foul points?

Yes. It's CCWM without foul points (the "u" stands for Unpenalized). Sorry about that. I forgot to explain it.



uCCWM = ( final - foul ) - ( op_final - op_foul )

"regular" CCWM = ( final - op_final )

also:



EPA = ( final - foul ) - uAvg

DPR = op_final

uDPR = ( op_final - op_foul )

[Ether]

I really like EPA. Certainly more useful for cross event comparisons than raw OPR/CCWM. Wish we could tie in average event score or something of the like to EPA. Being able to view OPR separate from fouls and EPA certainly helps remove outliers from the data.

Thanks for the data as usual. :)

I just added a spreadsheet showing the average alliance scores (win, lose, all, unpenalized all, fouls) for each event.

Look for Average Alliance Scores at each Event near the bottom of the list of attachments:

http://www.chiefdelphi.com/media/papers/2985

Not sure what the best way would be to integrate these with EPA.

[Ether]

I'd like to get some feedback on which of the attachments you'd like to see again for Week5. I need to pare it back a bit.

http://www.chiefdelphi.com/media/papers/2985

[Basel A]

Posting here because it's a very similar subject.. On request of some twitter people, I used the @frcfms data to determine the % of matches determined by fouls (any change in result, including from or to ties) over the past three years, broken down into years and categories such as quals, semifinals, or 3rd match of an elimination series. Those statistics are available here: http://goo.gl/mKuTXS.

[brennonbrimhall]

Posting here because it's a very similar subject.. On request of some twitter people, I used the @frcfms data to determine the % of matches determined by fouls (any change in result, including from or to ties) over the past three years, broken down into years and categories such as quals, semifinals, or 3rd match of an elimination series. Those statistics are available here: http://goo.gl/mKuTXS.

Any chance we could get the number of data points (FMS tweets) you used? Stuff like that's really important for generating standard errors for confidence intervals and seeing just how much of a change there's been for this year. As Walter Lewin (https://www.google.com/search?q=walter+lewin&oq=walter+lewin) is fond of saying, "any measurement is useless without a knowledge of the uncertainties."

I will be adding additional analyses as time permits. Special requests will be considered also.


How easy would it be to post your dataset in a format that Excel tolerates? Doing so gets you out of 2-3. ;)
Is it possible to see the percent of matches that have been decided by fouls, broken down by week and by match type (qualification/elimination)?
Using assist scores, would it be possible to determine the average number of cycles an alliance generates? Again, is it possible to see averages per week, by match type?

[Ether]

How easy would it be to post your dataset in a format that Excel tolerates? Doing so gets you out of 2-3.

Thanks to Brandon, the raw Twitter data is available in CSV format (http://www.chiefdelphi.com/forums/showpost.php?p=1350317&postcount=11) this year. Knock yourself out. CSV files open directly with Excel if you have the file associations set to tell Windoze that Excel is the proper app to use to open them.

[brennonbrimhall]

Thanks to Brandon, the raw Twitter data is available in CSV format (http://www.chiefdelphi.com/forums/showpost.php?p=1350317&postcount=11) this year. Knock yourself out. CSV files open directly with Excel if you have the file associations set to tell Windoze that Excel is the proper app to use to open them.


Sweet! Thanks for pointing me in the right direction.

EDIT: It doesn't seem like assist points are part of the FMS feed. Is there another resource out there I could mine for analysis?

[jlmcmchl]

Sweet! Thanks for pointing me in the right direction.

EDIT: It doesn't seem like assist points are part of the FMS feed. Is there another resource out there I could mine for analysis?

The only true way to know assist points by match is to recalculate them after every match based on USFIRST's ranking page. Otherwise, you need a new data source from FIRST, i believe. The scoring/index.lasso page on USFIRST (You can find it's link and related links via a google search) looks promising, but I haven't found any match information from there.

[Ether]

Twitter data1 has final score, foul points awarded, autonomous, and "TeleOp"2 for each match

Match Results has final score for each match

Team Standings has Assist, autonomous, Truss&Catch, and "TeleOp"3 for each Team (total of all alliance scores for alliances on which the team played).



1The usual Twitter data caveats apply.

2Twitter "TeleOp" is equal to Assist + T&C + Goals

3 Team Standings "TeleOp" is equal to Goals + Fouls

[brennonbrimhall]

Is it possible to see the percent of matches that have been decided by fouls, broken down by week and by match type (qualification/elimination)?


I thought I'd post my analysis -- I was looking to see if the proportion of matches that ended in fouls was increasing, decreasing, or staying the same. I ran a Chi-Square GOF test on the expected number of matches decided by fouls against the number of matches played that week times the total average rate of matches decided by fouls.

My null hypothesis was that each match had the same proportion of matches decided by fouls. With the percentage of matches that are decided by fouls being 23.34%, my x^2 value ended up being 46.4. I came up with a p-value of 5.06945E-07. So clearly, weeks and match type produce different numbers of matches decided by fouls.


MDBF Played Expected Difference X^2 Std. Resid.
1E: 29 159 37.11062625 -8.110626253 1.772598979 -1.331389868
1Q: 228 783 182.7523293 45.24767072 11.20287612 3.347069781
2E: 61 205 47.84703385 13.15296615 3.615699965 1.901499399
2Q: 312 1039 242.5027716 69.49722845 19.91674046 4.462817547
3E: 48 264 61.61764359 -13.61764359 3.009531137 -1.734800028
3Q: 293 1237 288.7160042 4.283995754 0.06356634 0.25212366
4E: 41 245 57.18304045 -16.18304045 4.579868371 -2.140062702
4Q: 253 1184 276.3457955 -23.34579549 1.972261479 -1.404372272
5E: 64 283 66.05224673 -2.052246727 0.063763412 -0.252514182
5Q: 340 1493 348.4664465 -8.466446515 0.205703353 -0.453545316


But, there's not really a solid trend. I would have liked to see the number of matches decided by fouls go down by week, but it's not that clear cut. It will be interesting to see Week 6 results and how they compare.

[brennonbrimhall]

The only true way to know assist points by match is to recalculate them after every match based on USFIRST's ranking page. Otherwise, you need a new data source from FIRST, i believe. The scoring/index.lasso page on USFIRST (You can find it's link and related links via a google search) looks promising, but I haven't found any match information from there.



Twitter data1 has final score, foul points awarded, autonomous, and "TeleOp"2 for each match

Match Results has final score for each match

Team Standings has Assist, autonomous, Truss&Catch, and "TeleOp"3 for each Team (total of all alliance scores for alliances on which the team played).



1The usual Twitter data caveats apply.

2Twitter "TeleOp" is equal to Assist + T&C + Goals

3 Team Standings "TeleOp" is equal to Goals + Fouls



That's too bad that FIRST doesn't give that to us on a per-match basis. Thanks for the clarification.

[Ether]

That's too bad that FIRST doesn't give that to us on a per-match basis.

Join the chorus. Talk it up. Find out if there's a mentor on your team who might know someone within FIRST who would have the interest and authority to make that happen.

http://www.chiefdelphi.com/forums/showpost.php?p=1359258&postcount=4

http://www.chiefdelphi.com/forums/showthread.php?p=1144161#post1144161

http://www.chiefdelphi.com/forums/showpost.php?p=1358391&postcount=15

http://www.chiefdelphi.com/forums/showpost.php?p=1276104&postcount=20

http://www.chiefdelphi.com/forums/showpost.php?p=1147996&postcount=17

http://www.chiefdelphi.com/forums/showpost.php?p=1356165&postcount=3

http://www.chiefdelphi.com/forums/showpost.php?p=1356381&postcount=9

[Basel A]

Any chance we could get the number of data points (FMS tweets) you used? Stuff like that's really important for generating standard errors for confidence intervals and seeing just how much of a change there's been for this year. As Walter Lewin (https://www.google.com/search?q=walter+lewin&oq=walter+lewin) is fond of saying, "any measurement is useless without a knowledge of the uncertainties."

I believe the smallest was around 150 (total # of finals matches in 2012). Not large by statistical standards, but it's not like I'm publishing my results in the Midwestern Journal of FRC Statistics.

[Ether]

This thread is for discussion of Week4 Twitter data analyses posted here:

http://www.chiefdelphi.com/media/papers/2985


Week5 Twitter data analyses are posted here:

http://www.chiefdelphi.com/media/papers/2991