Research talk:Autoconfirmed article creation trial/Work log/2017-08-23

From Meta, a Wikimedia project coordination wiki
Jump to navigation Jump to search

Thursday, August 24, 2017[edit]

Today I've been working on the page creation datasets, documenting our methods, and continuing gathering and analyzing data.

Distribution of patroller workload[edit]

We saw yesterday that the number of active patrollers per day has been slowly increasing over time. We also saw that the number of patrol actions has not behaved in the same way, and that the number of created articles per day is slowly decreasing. One key consequence of the decrease in number of created articles and increase in number of active patrollers is that on average they have to do less work each to keep up with the influx of articles. However, it is common in online communities to see an uneven distribution of participation (ref the 1-9-90 rule of thumb). If the increase in active patrollers largely comes through ones that do very little work, it means that the core patrollers have to continue working just as hard.

H11 seeks to investigate this. We calculate the number of patrol actions done by each patroller every day, and then calculate the Gini coefficient of each day's patrollers. It looks like this:

Patroller gini coefficient 2012-2017.png

If the coefficient is low, it means the work is more evenly distributed, whereas a higher coefficient means that a few patrollers are doing most of the work. We can see in the graph that the coefficient fluctuates a lot, and that there is no clear trend in the data. It is therefore not clear that the increase in number of active patrollers had led to a decrease in the workload of core patrollers.

An alternative way to measure this is to calculate the proportion of the total number of patrol actions that are done by the 25% most active patrollers. That plot looks exactly like the Gini coefficient, except the Y-axis is perhaps more easily interpreted:

Patroller prop actions top quartile 2012-2017.png

This graph suggests that a fairly small group of patrollers are doing most of the work. For example, the daily average number of active patrollers in 2017 is around 40, meaning that 10 patrollers are usually doing 70–80% of the work (some days a bit more, some days a bit less). As discussed previously, this suggests that the workload of the core contributors is not decreasing significantly from the increase in number of active patrollers.