As seen on

Favicon for http://digiphile.tumblr.com

digiphile.tumblr.com

If A Data Point Has No Context, Does It Have Any Meaning?

A lesson from Erin Simpson

Kelsey D Atherton3 Likes4,975 Views

Favicon for https://storify.com

Favicon for https://twitter.com

twitter.com

546 views

Favicon for http://digiphile.tumblr.com

digiphile.tumblr.com

Favicon for http://www.tumblr.com

[cracks knuckles] MT @FiveThirtyEight: We're mapping kidnapping rates in Nigeria 1982 to 2014. http://53eig.ht/1oof5cw pic.twitter.com/FQjZQDHJme
EM Simpson @charlie_simpson
·
Tue, May 13 2014 19:35:13
Reply Retweet Favorite
//twitter.com/charlie_simpson/status/466300650016763904
— EM Simpson (@charlie_simpson)Tue, May 13 2014 19:35:13
EM Simpson @charlie_simpson
Nope MT @armarvin: .@charlie_simpson @FiveThirtyEight Do you think the #GDELT is sound enough to make calls on #Nigeria kidnapping trends?
Tue, May 13 2014 19:52:26
Reply Retweet Favorite
//twitter.com/charlie_simpson/status/466304983017594880
— EM Simpson (@charlie_simpson)Tue, May 13 2014 19:52:26
EM Simpson @charlie_simpson
All trend analysis using #GDELT has to take into account the exponential increase in news stories which generate the data. @FiveThirtyEight
Tue, May 13 2014 19:54:48
Reply Retweet Favorite
//twitter.com/charlie_simpson/status/466305581083414529
— EM Simpson (@charlie_simpson)Tue, May 13 2014 19:54:48
EM Simpson @charlie_simpson
#GDELT isn't designed for tracking discrete events like "kidnappings" or "suicide bombings" bc it's based on news reports. @FiveThirtyEight
Tue, May 13 2014 20:01:46
Reply Retweet Favorite
//twitter.com/charlie_simpson/status/466307333648576513
— EM Simpson (@charlie_simpson)Tue, May 13 2014 20:01:46
EM Simpson @charlie_simpson
The beauty of automated event data is that you don't hand-code things. The power-and the limitations-come from automation @FiveThirtyEight
Tue, May 13 2014 20:02:41
Reply Retweet Favorite
//twitter.com/charlie_simpson/status/466307563697352704
— EM Simpson (@charlie_simpson)Tue, May 13 2014 20:02:41
EM Simpson @charlie_simpson
So if #GDELT says there were 649 kidnappings in Nigeria in 4 months, WHAT IT'S REALLY SAYING is there were 649 news stories abt kidnappings.
Tue, May 13 2014 20:04:50
Reply Retweet Favorite
//twitter.com/charlie_simpson/status/466308105416884225
— EM Simpson (@charlie_simpson)Tue, May 13 2014 20:04:50
John Beieler @johnb30
Apropos of nothing, @jay_yonamine & Schrodt have written things about using event data. http://jayyonamine.com/wp-content/uploads/2012/07/YonamineSchrodt_A_Guide_to_Event_Data.pdf … and http://jayyonamine.com/wp-content/uploads/2012/06/Working-with-Event-Data-A-Guide-to-Aggregation-Choices.pdf …
@charlie_simpson
·
Tue, May 13 2014 20:05:52
Reply Retweet Favorite
//twitter.com/johnb30/status/466308363970568192
— John Beieler (@johnb30)Tue, May 13 2014 20:05:52
EM Simpson @charlie_simpson
News coverage varies widely over time and space. That's really important when making comparisons across time and space. @FiveThirtyEight
Tue, May 13 2014 20:06:49
Reply Retweet Favorite
//twitter.com/charlie_simpson/status/466308603079847936
— EM Simpson (@charlie_simpson)Tue, May 13 2014 20:06:49
EM Simpson @charlie_simpson
As basic validation of #GDELT for this problem set, I'd like to see the following for kidnappings in Nigeria...
Tue, May 13 2014 20:09:10
Reply Retweet Favorite
//twitter.com/charlie_simpson/status/466309194002731009
— EM Simpson (@charlie_simpson)Tue, May 13 2014 20:09:10
EM Simpson @charlie_simpson
1) Total number of stories coded for Nigeria over time (what is the shape of that curve)?
Tue, May 13 2014 20:09:37
Reply Retweet Favorite
//twitter.com/charlie_simpson/status/466309310126231552
— EM Simpson (@charlie_simpson)Tue, May 13 2014 20:09:37
EM Simpson @charlie_simpson
2) What are the total number of events generated for Nigeria over time? (What is the shape of that curve?)
Tue, May 13 2014 20:10:16
Reply Retweet Favorite
//twitter.com/charlie_simpson/status/466309471242027008
— EM Simpson (@charlie_simpson)Tue, May 13 2014 20:10:16
EM Simpson @charlie_simpson
3) How does the number of kidnappings compare to the number of coded events? Same shape? Key differences?
Tue, May 13 2014 20:10:48
Reply Retweet Favorite
//twitter.com/charlie_simpson/status/466309607380369408
— EM Simpson (@charlie_simpson)Tue, May 13 2014 20:10:48
EM Simpson @charlie_simpson
4) How many overall events are coded with a specific geolocation? How many get coded to a centroid? (And where is the centroid?)
Tue, May 13 2014 20:11:40
Reply Retweet Favorite
//twitter.com/charlie_simpson/status/466309825719070722
— EM Simpson (@charlie_simpson)Tue, May 13 2014 20:11:40
EM Simpson @charlie_simpson
5) How many kidnapping events are coded with a specific geolocation? Does that change over time?
Tue, May 13 2014 20:12:11
Reply Retweet Favorite
//twitter.com/charlie_simpson/status/466309955184623616
— EM Simpson (@charlie_simpson)Tue, May 13 2014 20:12:11
EM Simpson @charlie_simpson
6) How does this information track with other open source reporting? HRW, UN, WB local NGO crime reporting? Can we corroborate trends?
Tue, May 13 2014 20:12:50
Reply Retweet Favorite
//twitter.com/charlie_simpson/status/466310118460514304
— EM Simpson (@charlie_simpson)Tue, May 13 2014 20:12:50
EM Simpson @charlie_simpson
In conclusion: VALIDATE YOUR FREAKING DATA. It's not true just because it's on a goddamn map. #GDELT @FiveThirtyEight
Tue, May 13 2014 20:13:45
Reply Retweet Favorite
//twitter.com/charlie_simpson/status/466310350837911554
— EM Simpson (@charlie_simpson)Tue, May 13 2014 20:13:45
EM Simpson @charlie_simpson
Learn the data generating process. Learn the coding rules. Match it against some real world reporting. THEN publish #GDELT @FiveThirtyEight
Tue, May 13 2014 20:14:45
Reply Retweet Favorite
//twitter.com/charlie_simpson/status/466310600600322048
— EM Simpson (@charlie_simpson)Tue, May 13 2014 20:14:45
EM Simpson @charlie_simpson
And never, EVER use #GDELT for reporting of discrete events. That's not what it's for. Not kidnappings, not murders, not suicide bombings.
Tue, May 13 2014 20:15:48
Reply Retweet Favorite
//twitter.com/charlie_simpson/status/466310866225217536
— EM Simpson (@charlie_simpson)Tue, May 13 2014 20:15:48
EM Simpson @charlie_simpson
Event data, and #GDELT in particular, have unique quirks. But all data have to be understood on their own terms. It's not magic, people.
Tue, May 13 2014 20:17:46
Reply Retweet Favorite
//twitter.com/charlie_simpson/status/466311358024527872
— EM Simpson (@charlie_simpson)Tue, May 13 2014 20:17:46
EM Simpson @charlie_simpson
I'll buy a beer for whoever storifys that rant...
Tue, May 13 2014 20:37:36
Reply Retweet Favorite
//twitter.com/charlie_simpson/status/466316350315851778
— EM Simpson (@charlie_simpson)Tue, May 13 2014 20:37:36
Hi
Ward Harkavy @WHarkavy
@charlie_simpson yes! Thank God somebody who knows what the fuck she's doing is pointing this out.
@charlie_simpson
·
Tue, May 13 2014 20:41:46
Reply Retweet Favorite
//twitter.com/WHarkavy/status/466317399416119298
— Ward Harkavy (@WHarkavy)Tue, May 13 2014 20:41:46
Brian Root @brian_root
Go read @charlie_simpson's timeline to understand why the @FiveThirtyEight map of Nigerian kidnapping data is so problematic. Great rant.
@charlie_simpson
·
Tue, May 13 2014 20:48:40
Reply Retweet Favorite
//twitter.com/brian_root/status/466319137871237121
— Brian Root (@brian_root)Tue, May 13 2014 20:48:40
Zeynep Tufekci @zeynep
Another major data blunder by @FiveThirtyEight. Go read @charlie_simpson TL. GDELT data is what it is; it cannot do what you wish it does.
@charlie_simpson
·
Tue, May 13 2014 20:53:04
Reply Retweet Favorite
//twitter.com/zeynep/status/466320244903256065
— Zeynep Tufekci (@zeynep)Tue, May 13 2014 20:53:04
Tony Boyles @AABoyles
I was going to blog about #GDELT and http://fivethirtyeight.com/datalab/mapping-kidnappings-in-nigeria/ …, but @charlie_simpson has twitter-ranted most everything I meant to say.
@charlie_simpson
·
Tue, May 13 2014 21:04:07
Reply Retweet Favorite
//twitter.com/AABoyles/status/466323022379765760
— Tony Boyles (@AABoyles)Tue, May 13 2014 21:04:07

3