Anonymous editing was enabled on mobile web on 30th March 2015 to all users (previous it was available in an experimental mode of the site). Now we have almost a month's worth of data I thought it would be a good time to check the impact... it's a little disappointing to be honest... but it depends what we are optimising and consider the most important.
Quick summary: * All edits up 154% * Edits from logged in users down 152% * Errors up 600% * No noticeable impact on the new active editor graph [0] (editors that hit 5 edits in the month period) * First edits by logged in users down 176% (although this could arguably be said to be compensated by anonymous edits) * New account creation up by 192%
Follow ups * Aaron H, it would be great if you could report back with some findings on the quality of the edits during this same period. * Can anyone provide theories why registrations jumped so much? This might be related to the change or because of something else
Details on the queries I ran: In March for a 26 day period before the change: * 170,948 total edits [1] * 169,845 non-anonymous edits [6] ** by 40,658 distinct users [7] * 26,617 users completely their first ever edit [11] * 9,528 errors [8] * 219,012 accounts created on mobile [12]
For a similar 26 day period in April * 263,986 total edits [2] * 136,079 non-anonymous edits [4] ** by 26,823 distinct users [5] * 15,109 users completely their first ever edit [10] * 58,394 errors [9] * 419,976 accounts created on mobile [13]
[0] http://mobile-reportcard.wmflabs.org/ [1] select count(*) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150301000000 and timestamp < 20150327000000 [2] select count(*) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150401000000 and timestamp < 20150427000000 [3] https://phabricator.wikimedia.org/T97494 [4] select count(*) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150401000000 and timestamp < 20150427000000 and event_username is NOT NULL [5] select count(distinct event_username) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150401000000 and timestamp < 20150427000000 and event_username is NOT NULL [6] select count(*) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150301000000 and timestamp < 20150327000000 and event_username is NOT NULL [7] select count(distinct event_username) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150301000000 and timestamp < 20150327000000 and event_username is NOT NULL [8] select count(*) from MobileWebEditing_8599025 where event_action = 'error' and timestamp > 20150301000000 and timestamp < 20150327000000 [9] select count(*) from MobileWebEditing_8599025 where event_action = 'error' and timestamp > 20150401000000 and timestamp < 20150427000000 [10] select count(*) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150401000000 and timestamp < 20150427000000 and event_userEditCount = 0 [11] select count(*) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150301000000 and timestamp < 20150327000000 and event_userEditCount = 0 [12] select count(*) from ServerSideAccountCreation_5487345 where event_displayMobile = 1 and timestamp > 20150301000000 and timestamp < 20150327000000 [13] select count(*) from ServerSideAccountCreation_5487345 where event_displayMobile = 1 and timestamp > 20150401000000 and timestamp < 20150427000000
How can edits from logged-in users be down 152%? The maximum decrease should be 100% - and that would mean no edits from logged-in users. Same for first edits by logged-in users.
I seriously wonder about the number of "accounts" created on mobile - are we really adding 0.05% of all of the current existing accounts every single month just on mobile? Or does that include accounts that already existed on a WMF site?
Risker/Anne
On 28 April 2015 at 20:00, Jon Robson jrobson@wikimedia.org wrote:
Anonymous editing was enabled on mobile web on 30th March 2015 to all users (previous it was available in an experimental mode of the site). Now we have almost a month's worth of data I thought it would be a good time to check the impact... it's a little disappointing to be honest... but it depends what we are optimising and consider the most important.
Quick summary:
- All edits up 154%
- Edits from logged in users down 152%
- Errors up 600%
- No noticeable impact on the new active editor graph [0] (editors that hit
5 edits in the month period)
- First edits by logged in users down 176% (although this could arguably be
said to be compensated by anonymous edits)
- New account creation up by 192%
Follow ups
- Aaron H, it would be great if you could report back with some findings on
the quality of the edits during this same period.
- Can anyone provide theories why registrations jumped so much? This might
be related to the change or because of something else
Details on the queries I ran: In March for a 26 day period before the change:
- 170,948 total edits [1]
- 169,845 non-anonymous edits [6]
** by 40,658 distinct users [7]
- 26,617 users completely their first ever edit [11]
- 9,528 errors [8]
- 219,012 accounts created on mobile [12]
For a similar 26 day period in April
- 263,986 total edits [2]
- 136,079 non-anonymous edits [4]
** by 26,823 distinct users [5]
- 15,109 users completely their first ever edit [10]
- 58,394 errors [9]
- 419,976 accounts created on mobile [13]
[0] http://mobile-reportcard.wmflabs.org/ [1] select count(*) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150301000000 and timestamp < 20150327000000 [2] select count(*) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150401000000 and timestamp < 20150427000000 [3] https://phabricator.wikimedia.org/T97494 [4] select count(*) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150401000000 and timestamp < 20150427000000 and event_username is NOT NULL [5] select count(distinct event_username) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150401000000 and timestamp < 20150427000000 and event_username is NOT NULL [6] select count(*) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150301000000 and timestamp < 20150327000000 and event_username is NOT NULL [7] select count(distinct event_username) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150301000000 and timestamp < 20150327000000 and event_username is NOT NULL [8] select count(*) from MobileWebEditing_8599025 where event_action = 'error' and timestamp > 20150301000000 and timestamp < 20150327000000 [9] select count(*) from MobileWebEditing_8599025 where event_action = 'error' and timestamp > 20150401000000 and timestamp < 20150427000000 [10] select count(*) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150401000000 and timestamp < 20150427000000 and event_userEditCount = 0 [11] select count(*) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150301000000 and timestamp < 20150327000000 and event_userEditCount = 0 [12] select count(*) from ServerSideAccountCreation_5487345 where event_displayMobile = 1 and timestamp > 20150301000000 and timestamp < 20150327000000 [13] select count(*) from ServerSideAccountCreation_5487345 where event_displayMobile = 1 and timestamp > 20150401000000 and timestamp < 20150427000000 _______________________________________________ Wikitech-l mailing list Wikitech-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l
On 28 April 2015 at 20:00, Jon Robson jrobson@wikimedia.org wrote:
<snip>
Details on the queries I ran: In March for a 26 day period before the change:
- 170,948 total edits [1]
- 169,845 non-anonymous edits [6]
** by 40,658 distinct users [7]
- 26,617 users completely their first ever edit [11]
- 9,528 errors [8]
- 219,012 accounts created on mobile [12]
For a similar 26 day period in April
- 263,986 total edits [2]
- 136,079 non-anonymous edits [4]
** by 26,823 distinct users [5]
- 15,109 users completely their first ever edit [10]
- 58,394 errors [9]
- 419,976 accounts created on mobile [13]
Like Anne, I also find your math incoherent.
Relative change is generally expressed ( (new_amount) - (old_amount) ) / (old_amount)
Assuming the above numbers are the correct totals, then it would seem that:
* All edits up 54% * Edits from logged in users down 20% * Errors up 510% * First edits by logged in users down 43% * New account creation up by 92%
-Robert Rohde
Thanks Robert for the corrections on my math. I sent this late in a long long day as it has been on my to do list for some time so these numbers could be discussed. Sadly my brain was obviously not up to it's usual standards.
Glad you were able to make sense of the numbers and correct them :) On 28 Apr 2015 8:39 pm, "Robert Rohde" rarohde@gmail.com wrote:
On 28 April 2015 at 20:00, Jon Robson jrobson@wikimedia.org wrote:
<snip>
Details on the queries I ran: In March for a 26 day period before the change:
- 170,948 total edits [1]
- 169,845 non-anonymous edits [6]
** by 40,658 distinct users [7]
- 26,617 users completely their first ever edit [11]
- 9,528 errors [8]
- 219,012 accounts created on mobile [12]
For a similar 26 day period in April
- 263,986 total edits [2]
- 136,079 non-anonymous edits [4]
** by 26,823 distinct users [5]
- 15,109 users completely their first ever edit [10]
- 58,394 errors [9]
- 419,976 accounts created on mobile [13]
Like Anne, I also find your math incoherent.
Relative change is generally expressed ( (new_amount) - (old_amount) ) / (old_amount)
Assuming the above numbers are the correct totals, then it would seem that:
- All edits up 54%
- Edits from logged in users down 20%
- Errors up 510%
- First edits by logged in users down 43%
- New account creation up by 92%
-Robert Rohde _______________________________________________ Wikitech-l mailing list Wikitech-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l
In the same period during March where 219,012 accounts were created on mobile there were 228,723 accounts created on desktop [0] In the same period during April where 419,976 accounts were created on mobile there were 195,359 accounts created on desktop [1] The ServerSideAccountCreation event logging schema captures all accounts being created. I really have no idea what drove these accounts being created to surpass desktop. Could be mobile apps or mobile web - not sure. I'd encourage anyone interested who knows SQL to get access to the database and run some queries.
Any community members interested in helping out here? I'm very sad the increase in errors wasn't picked up sooner... :-/
[0] select count(*) from ServerSideAccountCreation_5487345 where event_displayMobile = 0 and timestamp > 20150301000000 and timestamp < 20150327000000 [1] select count(*) from ServerSideAccountCreation_5487345 where event_displayMobile = 0 and timestamp > 20150401000000 and timestamp < 20150427000000
On Tue, Apr 28, 2015 at 8:13 PM, Risker risker.wp@gmail.com wrote:
How can edits from logged-in users be down 152%? The maximum decrease should be 100% - and that would mean no edits from logged-in users. Same for first edits by logged-in users.
I seriously wonder about the number of "accounts" created on mobile - are we really adding 0.05% of all of the current existing accounts every single month just on mobile? Or does that include accounts that already existed on a WMF site?
Risker/Anne
On 28 April 2015 at 20:00, Jon Robson jrobson@wikimedia.org wrote:
Anonymous editing was enabled on mobile web on 30th March 2015 to all
users
(previous it was available in an experimental mode of the site). Now we have almost a month's worth of data I thought it would be a good time to check the impact... it's a little disappointing to be honest... but it depends what we are optimising and consider the most important.
Quick summary:
- All edits up 154%
- Edits from logged in users down 152%
- Errors up 600%
- No noticeable impact on the new active editor graph [0] (editors that
hit
5 edits in the month period)
- First edits by logged in users down 176% (although this could arguably
be
said to be compensated by anonymous edits)
- New account creation up by 192%
Follow ups
- Aaron H, it would be great if you could report back with some findings
on
the quality of the edits during this same period.
- Can anyone provide theories why registrations jumped so much? This
might
be related to the change or because of something else
Details on the queries I ran: In March for a 26 day period before the change:
- 170,948 total edits [1]
- 169,845 non-anonymous edits [6]
** by 40,658 distinct users [7]
- 26,617 users completely their first ever edit [11]
- 9,528 errors [8]
- 219,012 accounts created on mobile [12]
For a similar 26 day period in April
- 263,986 total edits [2]
- 136,079 non-anonymous edits [4]
** by 26,823 distinct users [5]
- 15,109 users completely their first ever edit [10]
- 58,394 errors [9]
- 419,976 accounts created on mobile [13]
[0] http://mobile-reportcard.wmflabs.org/ [1] select count(*) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150301000000 and timestamp < 20150327000000 [2] select count(*) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150401000000 and timestamp < 20150427000000 [3] https://phabricator.wikimedia.org/T97494 [4] select count(*) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150401000000 and timestamp < 20150427000000
and
event_username is NOT NULL [5] select count(distinct event_username) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150401000000 and
timestamp
< 20150427000000 and event_username is NOT NULL [6] select count(*) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150301000000 and timestamp < 20150327000000
and
event_username is NOT NULL [7] select count(distinct event_username) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150301000000 and
timestamp
< 20150327000000 and event_username is NOT NULL [8] select count(*) from MobileWebEditing_8599025 where event_action = 'error' and timestamp > 20150301000000 and timestamp < 20150327000000 [9] select count(*) from MobileWebEditing_8599025 where event_action = 'error' and timestamp > 20150401000000 and timestamp < 20150427000000 [10] select count(*) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150401000000 and timestamp < 20150427000000
and
event_userEditCount = 0 [11] select count(*) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150301000000 and timestamp < 20150327000000
and
event_userEditCount = 0 [12] select count(*) from ServerSideAccountCreation_5487345 where event_displayMobile = 1 and timestamp > 20150301000000 and timestamp < 20150327000000 [13] select count(*) from ServerSideAccountCreation_5487345 where event_displayMobile = 1 and timestamp > 20150401000000 and timestamp < 20150427000000 _______________________________________________ Wikitech-l mailing list Wikitech-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l
Wikitech-l mailing list Wikitech-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l
On Tue, Apr 28, 2015 at 10:31 PM, Jon Robson jrobson@wikimedia.org wrote:
<snip>
Any community members interested in helping out here? I'm very sad the
increase in errors wasn't picked up sooner... :-/
What does event_action = 'error' actually mean?
If the action is stopped by the AbuseFilter is that counted as an "error"?
-Robert
On Wed, Apr 29, 2015 at 8:19 AM, Robert Rohde rarohde@gmail.com wrote:
On Tue, Apr 28, 2015 at 10:31 PM, Jon Robson jrobson@wikimedia.org wrote:
<snip>
Any community members interested in helping out here? I'm very sad the
increase in errors wasn't picked up sooner... :-/
What does event_action = 'error' actually mean?
If the action is stopped by the AbuseFilter is that counted as an "error"?
It means at some point during the editing workflow the user hit an error that stopped them from finishing their edit. We do capture AbuseFilter hits in this process (but with some of these errors you can recover and complete the edit.
We also store the error associated, although a quick scan shows this is currently not very helpful.
In theory 'http' error should only happen when a user cannot get an edit token - I've updated the bug for those interested.
-Robert _______________________________________________ Wikitech-l mailing list Wikitech-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l
Hey folks,
As requested, I started a research project page to do some analysis around this. See https://meta.wikimedia.org/wiki/Research:Mobile_anonymous_apocalypse
It's just a stub now. I'll have to clear a few other projects off my plate in order to pick this one up. You should expect to see updates there in 2-3 weeks.
-Aaron
On Wed, Apr 29, 2015 at 12:10 PM, Jon Robson jdlrobson@gmail.com wrote:
On Wed, Apr 29, 2015 at 8:19 AM, Robert Rohde rarohde@gmail.com wrote:
On Tue, Apr 28, 2015 at 10:31 PM, Jon Robson jrobson@wikimedia.org
wrote:
<snip>
Any community members interested in helping out here? I'm very sad the
increase in errors wasn't picked up sooner... :-/
What does event_action = 'error' actually mean?
If the action is stopped by the AbuseFilter is that counted as an
"error"?
It means at some point during the editing workflow the user hit an error that stopped them from finishing their edit. We do capture AbuseFilter hits in this process (but with some of these errors you can recover and complete the edit.
We also store the error associated, although a quick scan shows this is currently not very helpful.
In theory 'http' error should only happen when a user cannot get an edit token - I've updated the bug for those interested.
-Robert _______________________________________________ Wikitech-l mailing list Wikitech-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l
-- Jon Robson
Wikitech-l mailing list Wikitech-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l
It would be helpful to see how much mobile acconts getting indf. blocked (and how much have all edits in main ns and file ns deleted)
Best
Date: Tue, 28 Apr 2015 23:13:32 -0400 From: risker.wp@gmail.com To: wikitech-l@lists.wikimedia.org Subject: Re: [Wikitech-l] Anonymous editing impact on mobile
How can edits from logged-in users be down 152%? The maximum decrease should be 100% - and that would mean no edits from logged-in users. Same for first edits by logged-in users.
I seriously wonder about the number of "accounts" created on mobile - are we really adding 0.05% of all of the current existing accounts every single month just on mobile? Or does that include accounts that already existed on a WMF site?
Risker/Anne
On 28 April 2015 at 20:00, Jon Robson jrobson@wikimedia.org wrote:
Anonymous editing was enabled on mobile web on 30th March 2015 to all users (previous it was available in an experimental mode of the site). Now we have almost a month's worth of data I thought it would be a good time to check the impact... it's a little disappointing to be honest... but it depends what we are optimising and consider the most important.
Quick summary:
- All edits up 154%
- Edits from logged in users down 152%
- Errors up 600%
- No noticeable impact on the new active editor graph [0] (editors that hit
5 edits in the month period)
- First edits by logged in users down 176% (although this could arguably be
said to be compensated by anonymous edits)
- New account creation up by 192%
Follow ups
- Aaron H, it would be great if you could report back with some findings on
the quality of the edits during this same period.
- Can anyone provide theories why registrations jumped so much? This might
be related to the change or because of something else
Details on the queries I ran: In March for a 26 day period before the change:
- 170,948 total edits [1]
- 169,845 non-anonymous edits [6]
** by 40,658 distinct users [7]
- 26,617 users completely their first ever edit [11]
- 9,528 errors [8]
- 219,012 accounts created on mobile [12]
For a similar 26 day period in April
- 263,986 total edits [2]
- 136,079 non-anonymous edits [4]
** by 26,823 distinct users [5]
- 15,109 users completely their first ever edit [10]
- 58,394 errors [9]
- 419,976 accounts created on mobile [13]
[0] http://mobile-reportcard.wmflabs.org/ [1] select count(*) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150301000000 and timestamp < 20150327000000 [2] select count(*) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150401000000 and timestamp < 20150427000000 [3] https://phabricator.wikimedia.org/T97494 [4] select count(*) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150401000000 and timestamp < 20150427000000 and event_username is NOT NULL [5] select count(distinct event_username) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150401000000 and timestamp < 20150427000000 and event_username is NOT NULL [6] select count(*) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150301000000 and timestamp < 20150327000000 and event_username is NOT NULL [7] select count(distinct event_username) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150301000000 and timestamp < 20150327000000 and event_username is NOT NULL [8] select count(*) from MobileWebEditing_8599025 where event_action = 'error' and timestamp > 20150301000000 and timestamp < 20150327000000 [9] select count(*) from MobileWebEditing_8599025 where event_action = 'error' and timestamp > 20150401000000 and timestamp < 20150427000000 [10] select count(*) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150401000000 and timestamp < 20150427000000 and event_userEditCount = 0 [11] select count(*) from MobileWebEditing_8599025 where event_action = 'success' and timestamp > 20150301000000 and timestamp < 20150327000000 and event_userEditCount = 0 [12] select count(*) from ServerSideAccountCreation_5487345 where event_displayMobile = 1 and timestamp > 20150301000000 and timestamp < 20150327000000 [13] select count(*) from ServerSideAccountCreation_5487345 where event_displayMobile = 1 and timestamp > 20150401000000 and timestamp < 20150427000000 _______________________________________________ Wikitech-l mailing list Wikitech-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l
Wikitech-l mailing list Wikitech-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l
wikitech-l@lists.wikimedia.org