Research talk:Anonymous editor acquisition/Signup CTA experiment/Work log/2014-05-20
Add topicTuesday, May 20th
[edit]Note: I started this analysis on Tuesday, the 20th, but I'm finishing it on Wednesday the 21st. --Halfak (WMF) (talk) 23:22, 21 May 2014 (UTC)
OK. I need to know why SignupExpAccountCreationImpressions are tokened at such a low rate. My job today is to figure out what proportion of devices/browsers/OS's are not getting tokens.
First, we need to extract non-tokened events.
SELECT
userAgent AS user_agent
FROM SignupExpAccountCreationImpression_8539445
WHERE event_token IS NULL
AND timestamp > "2014051921"
ORDER BY RAND()
LIMIT 10000;
--Halfak (WMF) (talk) 23:22, 21 May 2014 (UTC)
I just finished running a script over the data that uses [1] to extract family fields from the user-agents. I had a few errors crop up from inside the parser (LAME!) so I just set those events to report "other" for user_agent details. This only happened 5 out of 10k times, so I'm not too worried. --Halfak (WMF) (talk) 23:40, 21 May 2014 (UTC)
OK... This would be better if I had a sample of token'd and non-token'd events.
SELECT
userAgent AS user_agent,
event_token IS NOT NULL as tokened
FROM SignupExpAccountCreationImpression_8539445
WHERE timestamp > "2014051921"
ORDER BY RAND()
LIMIT 10000;