Page MenuHomePhabricator

What percentage of edit reverts on Indonesian Wikipedia are performed from a mobile device?
Closed, ResolvedPublic


To evaluate how we'll be working with Indonesian Wikipedia patrollers we want a clearer picture of whether they're primarily patrolling from a mobile device or a desktop one.

it is possible to through EditAttemptStep logs to differentiate between desktop and mobile devices, even if users force desktop mode on mobiles.

  • What percentage of edit reverts are performed from mobile and desktop devices?
  • What is this percentage when only considering users with more than 250 edits?

Event Timeline

I added the 250 edits question because I imagine that new editors are more likely to be on mobile but then drop off, and therefore not be engaged with the Automoderator project - we're really targeting experienced users, so that's a breakdown that I think would be more useful to us.

@Samwalton9-WMF here are the results

Proportion of reverts made by device (users with less than 250 edits)

desktop654 (72.9%)
mobile243 (27.1%)

Proportion of reverts made by device (users with more than 250 edits)

desktop5308 (73.5%)
mobile1915 (26.5%)

Approach and considerations

  • Two full months of data (August and September) from EditAttempStep logs were considered. The user agent field is cleaned out during data sanization, so we only have 90 days of data to work with. July would be partial as well as initial days' data would have already been sanitized.
  • As you already know, if a user forces desktop mode on a mobile, those edits would be logged as desktop edits. As a work around, the device family field was used capturing all major known mobile device families. I used an initial list from this analysis and expanded on that. So results aren't 100% accurate, but the closest we can get. I think the actual percentages for mobile might be a little higher, because if the list missed any device families, they would have been classified as desktop.
  • Also, we don't have data for all the reverts. During the last two months, there were ~11K reverts, however, we only EditAttemptStep data for ~8K reverts. The rest have null values for device family.

Analysis notebook