Hi all.
Thanks for the debate and for sharing your figures and insights. I would like to offer
some comments on this (below).
----- Mensaje original -----
De: Yaroslav M. Blanter <putevod(a)mccme.ru>
Para: Research into Wikimedia content and communities
<wiki-research-l(a)lists.wikimedia.org>
CC:
Enviado: Miércoles 2 de Mayo de 2012 15:53
Asunto: Re: [Wiki-research-l] long in tooth: highly active editors are 1/3 fewer
The very active are in the vast majority of
cases still active - most
of the names near the top of this list
http://en.wikipedia.org/wiki/Wikipedia:EDITS [7] are blue linked which
means they have edited recently. Earlier this year the number of
editor whod made over 100,000 edits on En Wiki grew to over 150 and on
my projections there will be over 200 by the end of the year.
Now, I wanted to do it sometime, but your mail motivated me to do it today.
I counted the number of inactive users per number of contributions, taking
numbers from the first 7000 in the list. Placeholders are counted as inactive,
and this is a clear drawback, but there are too few of them to change the trend,
and some of them may be inactive as well.
The results first.
Range (numbers) Range (edits) #inactive % inactive
1-200 over 93828 32 16
201-400 67561-93655 33 16.5
401-600 52024-67556 38 19
601-800 43587-51942 39 19.5
801-1000 37805-43432 51 20.5
1001-1200 33271-37791 61 30.5
1201-1400 30256-33260 54 27
1401-1600 27593-30250 50 25
1601-1800 25364-27571 60 30
1801-2000 23682-25360 80 40
2001-2500 19699-23574 174 34.8
2501-3000 17089-19697 167 33.4
3001-3500 14777-17086 191 38.2
3501-4000 13049-14777 199 39.8
4001-4500 11674-13048 225 45
4501-5000 10495-11673 195 39
5001-5500 9570-10495 211 42.2
5501-6000 8699-9569 224 44.8
6001-6500 8011-8697 239 47.8
6501-7000 7379-8011 242v 48.4
These are very interesting figures, but only for EN Wikipedia. I concur with Gerard in
that we also need to compare figures with other languages, specially outside the group of
large Wikipedias.
The generational relay is a well-known effect in open communities (for instance, it has
also been studied in open source projects). However, the size of the community and the
size of the group of core contributions does matter. Losing 3 persons in a group of ~500
can be probably assumed by the rest of the group, whereas losing the same 3 in a group of
20 is a very different story.
Furthermore, the duration of idle periods (between two consecutive edits) is also
important. I am conducting a systematic analysis of this factor (that is, no sampling),
against other relevant metrics (lifetime, number of edits or date of the first edit). It
is not unfrequent for "casual editors" (< 100 edits) to have idle periods of
more than 2 or even more than 4 years. But this idle period is usually shorter for core
editors (longest periods usually between 3-6 months).
I mention this because, according to one of the comments on
http://en.wikipedia.org/wiki/Wikipedia_talk:List_of_Wikipedians_by_number_o…
the meaning of "inactive" top editors in this list is (verbatim): "editors
with more than 30 days since the last edit". I find this definition of "inactive
editor" at least questionable under the light of these results about idle periods.
The first conclusion is that editors with over 35K
edits are much less likely to
leave, increasingly unlikely as the # of edits goes up. This is clearly
statistically significant.
The second conclusion is that there is major loss of editors with about 20K
edits. I am not sure how statistically significant this is.
Since the table is clustered by rank, rather than by number of edits, I would report
instead about "top-2000" or "top-2500", since absolute figures in the
table are actually meaningful just relative to the performance of other editors. I would
also try to normalize edits by lifetime, to compensate the fact that editors with longer
lifetime had better chances to make more edits (which may hide fast-raising trends). But
the, admittedly, that would be a different table for a different purpose...
I obviously did not try to correlate this with the
lifetime, but if we take 10K
edits per year as an example, 2 years would be the most probable lifetime.
Richard Rohde reported slightly higher numbers.
So, yes, indeed, the editors leave after a couple of years, and they do not get
replaced.
In any case, I believe this is the key question to answer. Trying to characterize editors
who stopped their activity, either temporarily or permanently, is only one half of the
picture. The other half is learning what was the path that core editors followed till they
got there, and why now we have fewer people following that path.
Why is this interesting for the whole Wikipedia community? Just for the fun of counting
edits? For the sake of competition? No. It is important because very active editors are
supposed to have much more experience in the project, and that experience, that knowledge
about the editing process, about how to interact with other community members, and how to
build valuable content is a crucial asset for Wikipedia. Thus, I think that the focus
should also include senior members outside the list of top editors, but with a long-time
experience (e.g. +5 years). Let me recall that the vast majority of authors who have
participated in FAs had a total lifetime of more than 3 years (+1,000 days) in Wikipedia,
for all big languages (note: also for most of the middle-size Wikipedias).
Last, but not least, there is another important connection with maintenance activity.
Editors with special accounts (e.g. sysops) may become idle for several days in article
editing, but they continue to perform administrative duties systematically. As a result,
the trends in the number of new admins and RFAs, and number of administrative changes
performed over time should also complement this picture (since many, many admins were not
among the most prolific editors when they were appointed).
Best,
Felipe.
Cheers
Yaroslav
_______________________________________________
Wiki-research-l mailing list
Wiki-research-l(a)lists.wikimedia.org
https://lists.wikimedia.org/mailman/listinfo/wiki-research-l