Visualizing Editor Trends

What are the results we'll be looking at today?

• The monthly decline in active editors in en may be attributed to the old timers. We see this across other languages too.
• Only the articles in the (beginning - 2007) cohorts continue to see active edit activity.
• Retention rates are much higher in languages like de.
• zh (continues to show uptrend on many fronts. Active Editors, retention etc).

Monthly Editor Activity Graphs

Definitions

Active Editor

• An 'active editor' is a registered (and signed in) person (not known as a bot) who makes 5 or more edits in a month in mainspace on countable pages. [1]
• SQL query

Editor Cohort

• Editors are grouped by the month of their edit.
• Eg: Editors who made their first edit in Jan 07 belong to the Jan 07 cohort.
• SQL query

The cohorts are stacked along the Y axis.

Time runs on the X axis

Filter

• Filter the cohorts by the age of the cohorts in every month. 

Filter by the age of a cohort in a month. 

Now we have selected the cohort of age 1 month in every month. Eg: Cohort Jan 07 is 1 month old in month Jan 07, 2 months old in Feb 07 etc

Eg : The contribution of only cohorts of age 1 month in every month.

No of active editors per month - stacked cohorts

Lets filter to see the active editors from every month who belong to 1 month old cohorts

Percentage of active editors per month - staked cohorts

Lets filter like before.

What have we seen?

• Newly joining active editors + active editors in their second month ~ 40% of active editors in a month. ~60% active editors are older experienced editors. 
• The general decline can be attributed to the fall in the number of older editors.
• Even monthly declines can be attributed to them.
• The older cohorts can trigger an increase or decrease in the total number of active editors in a month.
• This is the case with other languages too. (de, fr, it. etc)

Monthly activity graphs - de,fr,zh

de

fr

zh

Editor Longevity

The cohorts are on the Y axis. Every row represents represents the activity of a cohort over time.

Filter by %

Filter by the % value

Setting the retention levels at >5%, we see a dramatic slide starting in Dec 05

Editor Longevity

• Atleast 5% editor retention levels starts falling in Dec 05, much before the peak in March 07.
• The retention steadily fell and has reached a steady state these days.
• On avg these days only 5% active editors continue after  6-8 months of joining.

What about other languages?

• de has a fall in retention too, but nothing as dramatic the case in en.
• zh has an increased retention of late.

de

ru

zh

Filtered at >=5% levels

Editor Retention Rate Change

Editor retention rates - en. We can pick declines & spikes in monthly activity by looking at the vertical streaks of green & red. The spike are caused when a lot of older cohorts suddenly get active & the declines are similarly when their activity falls.

Retention rate graph

• Retention in the first month these days is only 20%. (Only 20% survive to the 2nd month.)
• Even in the second month there is only a 50% retention rate. 
• By the third month we only have about 10% of the active editors who joined in that cohort.
• We lose editors even in the second month of the cohort.

Article Edit Activity

Active Article

• An active article is defined as an articles which receives 5 >= edits/month.
• SQL Query

Article Cohort

• Articles are grouped by their month of creation.
• Eg: Articles created in Jan 07 belong to the Jan 07 cohort.
• SQL query

The cohorts are on the Y-axis. A row shows the cohorts activity over time.

Article cohort longevity - de

Article Longevity

I don't have the graph for en, the query kept failing :-(

Article cohort longevity for de filtered at >=5% levels. We see a dramatic fall for articles created after 2007.

What have we seen?

• Only the articles created till 2007 see continued editing.
• The articles in the subsequent cohorts see very little edit activity.
• Do we have a problem in surfacing editable articles?
• Newly created articles rapidly lose all editing activity.
• We have two buckets of editing activity, the cohorts till 2007 & the articles in the newly created cohorts.
• This is the case in all the major wiki I have looked at. (Looked at all except en - the query kept failing.)

Conclusions

Lets do more of data viz :-)

• A monthly decline in active editors is a drop in new comers & old timers, more old timers (~60%). We see this across languages too.
• So a decline in active editors can be strongly attributed to the decline in activity in the older editors.
• Only the articles in the (beginning - 2007) cohorts continue to see active edit activity.
• Exceptions to the peak are zh (continues to show uptrend) & ru (peaks in 2009).
• Retention rates are much higher in languages like de.

Ideas I'm working on

https://meta.wikimedia.org/wiki/Research:Editor_Behaviour_Analysis_%26_Graphs/Ideas

Relevant Links

• Article Graphs
• Editor Graphs
• The SQL Queries
• Ideas & Lines of work
• The project page
• The code on github

Questions

• Are we losing editors because they can't find content they can edit? Content they can contribute to?
• How do we categorize editors better? The categories on the pages they edit? (I'm experimenting here) 

Thank You

I have other interesting results too. When can we do this next :-)

The graphs look very different when we define "edits >=100 as an active editor".