User:Pcb21/WP growth
|
|
| Contents |
Wikistats high-water marks
All the usual Alexa caveats apply here - in particular note that my and other users' fascination with the wikistats causes Alexa to get mentioned a lot on Wikipedia - causing Wikipedia users to dis-proportionately go to Alexa - thus (probably) ensuring the Alexa tool bar is more widespread on Wikipedia users browsers than the average Joe Internet User. (Counter-argument - Wikipedia is close to the open source philosophy - apparently 15% of traffic arrive with a moz browser - compared with only 1-2% of internet users as a whole. Thus fewer Alexa toolbars. I have a inkling that wikipedians are also more spyware aware.... and we did survive the last slashdotting remarkably well....) In any case this subsection records (high water mark) trends not absolutes so we should be ok!
http://www.alexa.com/data/details/traffic_details?q=wikipedia&url=http://www.wikipedia.org/
- Alexa daily traffic rank (guesstimates of date and number from graph)
- Late 2002 - 4000
- Mid March 2003 - 3500
- End March 2003 - 3300
- Late May 2003 - 2600
- Mid July 2003 - 2200
- Late August 2003 - 1100
- 1st September 2003 - 950
- 22nd September 2003 - 920
- 29th September 2003 - 870
- 10th November 2003 - 800
- 10th December 2003 - 775
- 15th December 2003 - 670
- 25th February 2004 - 593 - press release day
- 8th March 2004 - 568
- Alexa weekly traffic rank
- 5th February 2004 - 840
- 12th February 2004 - 770
- 24th February 2004 - 752
- 26th February 2004 - 735
- 16th March 2004 - 669
- Alexa three monthly traffic rank
- 5th February 2004 - 868
- 12th February 2004 - 860
- 24th February 2004 - 828
- 17th March 2004 - 781
By looking up the reach per million users of several sites it is possible to work out a correspondence between reach rank (which for wikipedia has historically been roughly equal to traffic rank (=reach rank * time spent at site) and the number of users.
Quantitative measures of article quality
(EN stat page (http://www.wikipedia.org/wikistats/EN/TablesWikipediaEN.htm)) We are very good at measuring the quantity of articles. Measures of quality are much harder to come by. Some surveyors even seem to get slightly outcast by the community for highlighting problems.
It is only really possible to look at quant measures in the post-rambot era (i.e. November 2002 onwards) because his prodigious efforts had a massive effect on db averages.
Some quant measures of quality:
- Mean article size has been rising by about 20bytes per month for the past year. (This is a linear rise, rather than exponential as for number of articles)
- Percentage of articles over 500 bytes has dropped from 76% to 75% and now 74%.
- Percentage of articles over 2000 bytes has increased from 25% to 26%.
- The two statistics essentially reflect the fact that all Rambot articles are between 500 and 2000 bytes long, and that the effect of Rambot articles is slowly diminishing over time, causing a greater spread in article size, though this does not indicate an increase in variability of quality, it is just an artefact of the rambot flood.
- Number of edits per article per month: Has stayed pretty steady.
- Nov 2002 0.3
- Dec 2002 0.6
- Jan 2003 0.4
- Feb 2003 0.3
- Mar 2003 0.3
- Apr 2003 0.3
- May 2003 0.4
- Jun 2003 0.4
- Jul 2003 0.3
- Aug 2003 0.4
- Sep 2003 0.3
- Oct 2003 0.3
- Nov 2003 0.5
- Dec 2003 0.4
- Jan 2004 0.3
- Feb 2004 0.3 (estimated 27th February)
Conclusions: All measures are more or less constant, indicating that whilst quantity of articles is increasing, quality is constant. This can be taken in a positive or negative way!
update
A few more measures that suggest article quality is falling, with a dramatic increase in short articles:
In the three months to end of February 2004, the number of articles grew by 22%. However the number of non-article "articles" (more precisely the number of articles that are counted in the official count, but have less than 200 characters (this sentence has more than 200 characters, by the way) grew by 50%. This indicates a significant increase in useless one or two line articles. The number of articles that are counted as non-articles even by the official count are not recorded, but presumably grew similarly.
In February 2004, the number of articles below 2kb (my threshold for stub display) increased as a proportion of total database for the first time since rambot (the previous 16 months had shown a decrease). The proportion crept back to 74.2%
- Further the number of articles less than 256 bytes shot up from 9.2% to 10.0% (this reflects the first point above)
Exponential growth?
If the number of articles is growing exponentially, then the time taken for the article count to double should be constant. Is this true? (First few months omitted because of rounding noise)
Time taken to double from end of:
- Sep 2001: 2 months
- Oct 2001: 3.5 months
- Nov 2001: 3 months
- Dec 2001: 5.5 months
- Jan 2002: 5.5 months
- Feb 2002: 7 months
- Mar 2002: 6 months (rambot effect starts)
- Apr 2002: 5 months
- May 2002: 5 months
- Jun 2002: 4 months
- Jul 2002: 3 months
- Aug 2002: 2 months
- Sep 2002: 3.5 months
- Oct 2002: 13 months (now having to double rambot articles)
- Nov 2002: 13 months
- Dec 2002: 13 months
- Jan 2003: 13 months
- Feb 2003: not yet doubled
as usual rambot hampers the lay-statisticians efforts
for later months it is worth trying time to taken to add 25% more articles as this gives more data than doubling:
- Oct 2002: 4 months
- Nov 2002: 4 months
- Dec 2002: 4.5 months
- Jan 2003: 5 months
- Feb 2003: 4 months
- Mar 2003: 4 months
- Apr 2003: 4 months
- May 2003: 4 months
- Jun 2003: 4 months
- Jul 2003: 4.5 months
- Aug 2003: 4 months
- Sep 2003: 4 months
- Oct 2003: 4 months
This is useful empirical evidence that post-rambot, Wikipedia #articles has been growing exponentially.