Page MenuHomePhabricator

Mentor status dashboard: display mentor activity, responsiveness, and other stats
Open, MediumPublic

Description

User story & summary:

As an admin who cares about new editors receiving timely responses to their mentorship questions, I want to be able to monitor mentors and be fully informed about Mentor activity and responsiveness so I can make the right decision about whether to be proactive and reach out to a Mentor, reassign their mentees, possibly dementor them, or whether to allow a proposed automatic procedure (see T321509) to take over and cull them when a configured threshold is reached.

As a Growth team member interested in Mentorship, I need data about mentor activity, responsiveness, awayness, performance, abandonment, and exit from the program so I can measure and evaluate how Mentors are performing overall at an individual wiki and know how to best allocate team resources to optimize for mentee satisfaction and retention based on mentor response.

Background:

Currently, an admin has no easy access to information about mentor activity, as it affects what mentees are experiencing from their mentors. At task T321509, @Kudpung pointed out (at T321509#11881665) that a mentee was assigned a mentor with 1,465 edits and 508 days of inactivity. Some wikis (de, ru, ko) are coming up with bot-driven auto-dementorship to deal with this. As a partial solution, T321509 "Provide a system to automatically suspend or remove mentors from the list of mentors" was proposed. That proposal addresses the issue of ensuring we do not miss cases where mentors are inactive for long periods via a new automated dementorship process.

What the current manual system and the proposed automated system have in common is that they both lack the ability to provide admins with access to full information about mentor status and activity to support admins who wish to remain fully abreast of mentor status in order to support making administrative decisions about mentors when warranted. Under the current system, we don't know about Mentor1465 and their 508 days of inactivity, so admins don't do anything about them. Under the proposed system, we still don't know anything about Mentor-1465; the automated process will eventually strip their bit and reassign their mentees, and we will either learn about it after the fact (if logged or otherwise reported) or we will never know at all. The admin is left out of the picture, even if they might have been able to help (either the mentor, or the mentee) earlier in the process, had they been aware.

Regarding the automated system, in T321509#11882195, @KStoller-WMF wrote:

It is important to note that this type of automation can address inactivity, but it cannot reliably identify or remove mentors who are ineffective, as that would be much harder (and likely problematic) to attempt to address programmatically.

We need to support providing admins the information they need to reliably monitor and remove ineffective mentors in a way an automated system cannot. The way to support admins in this way is via a new dashboard tool that provides the missing information.

Proposal:

A new reporting tool should be created aimed at admins to provide them the ability to make manual decisions about mentors when appropriate. A Mentor Status Dashboard is proposed.

The Mentor Status Dashboard should be a table of all mentors with sortable columns:

  • MENTOR – mentor name
  • RTNDAT – self-declared away status/return date
  • LEDAT – date of last WP edit
  • N30 – number of edits in last 30 days
  • LQDAT – date of last mentee question
  • LADAT – date of last mentor response ('A' for answer, as in Q&A)

And some frequency stats t.b.d. to assess mentor activity, but perhaps things like:

  • Q30 – number of mentee questions in last 30 days to this mentor ('volume'; skip last 5 days)
  • A30 – number of mentor answers in last 30 days from this mentor ('responses'; skip last 5 days)
  • NA30 – number of mentee questions in last 30 days not answered ('dropped'; skip last 5 days?)
  • A30pctQ – mentor answers as a percent of all mentee questions in last 30 days, i.e., A30*100/Q30 ('responsiveness')
  • MTBQ30 – mean time between mentee questions in last 30 days ('mean Q frequency')
  • MTBA30 – mean time between mentor answers/responses ('mean A frequency'

and maybe also some stats to assess mentor burden, like:

  • A30pctE – mentor answers as a percent of all mentor edits in last 30 days (skip 5), i.e., A30*100/N30 ('burden')

This would provide admins with the data they need to watch what is going on, never miss a user like Mentor-1465 again, and decide what, if anything, to do about a mentor nearing (one of?) the dementorship threshold(s?).

Style

As for styling, some things to consider:

  • mentor name – set max-width and force-wrap or use ellipsis with tooltip to limit col 1 width and avoid horizontal scroll
  • liberal use of classes and a default style sheet, in order to enable individual admins to restyle the table to their liking using common.css.
  • perhaps an unobtrusive row background coloring of faded green, amber, and pink to highlight mentors at various levels of activity relative to a local community threshold might be useful; or just several row classes with no default bg-color so individual admins could pick their own scheme for whether and how to highlight.
  • transparency – who should be able to view it (everyone, imho)
  • dementor column – an optional convenience column with action and/or visibility restricted to admins which would remove the mentorship bit and initiate mentee reassignment
NOTE: Not shown are other stats of more interest to the Growth team itself, such as abandonment (no answer to questions, but mentor still active) and exit from the program. But stats like 'burden' (here defined as A30pctE, perhaps others) should be plotted against abandonment and exit as curves on a time-series graph, which might give an early warning about individual mentors who are approaching values that indicate they are at risk for leaving or not answering based on previous experience. If averaged over all mentors, it could show when a program at an individual wiki is at risk and needs resources applied to recruitment.

Event Timeline

Restricted Application added a subscriber: Aklapper. · View Herald Transcript
Mathglot updated the task description. (Show Details)
Mathglot updated the task description. (Show Details)

I'm not sure on the technical effort for this, but happy to discuss with the Growth team.

I don't think there's any need to get too technical to resolve this. All that's basically being asked is for the mentorship programme to exercise more due diligence when according the 'mentor' right to users. What we are often seeing at en.Wiki is either newbies mentoring newbies, or in the example I posted above, what we rather pejoratively and rightly or wrongly call 'hat-collectors'. That said, the local Wikis should be able to set their own thresholds, and IMO these should be quite high - at least as demanding as those for NPP's Guidelines for Granting.

I don't think there's any need to get too technical to resolve this. All that's basically being asked is for the mentorship programme to exercise more due diligence when according the 'mentor' right to users. What we are often seeing at en.Wiki is either newbies mentoring newbies, or in the example I posted above, what we rather pejoratively and rightly or wrongly call 'hat-collectors'. That said, the local Wikis should be able to set their own thresholds, and IMO these should be quite high - at least as demanding as those for NPP's Guidelines for Granting.

@Kudpung Admins can already configure higher minimum requirements for Mentors, such as increasing the threshold for account age and edit count, through the Mentorship configuration page.

That said, my understanding is that @Mathglot is advocating for something broader than minimum eligibility requirements alone, which seems reasonable. Reaching a certain edit count or tenure does not necessarily mean someone will be an effective, responsive, or engaged mentor.

I suspect calculating and maintaining more nuanced mentor metrics would require significant technical work, but I do see value in surfacing this kind of information. Data such as response rates, number of questions answered, or mentor activity levels could help admins evaluate the effectiveness of mentors over time. It could also provide useful feedback to mentors themselves. In fact, when the Mentor Dashboard was originally envisioned, there was interest in including “impact” metrics for mentors, such as the number of mentee contributions, questions answered, or thanks received from mentees.

Full transparency: the Growth team is balancing a large number of competing priorities, so I cannot promise we will be able to take this on. That said, I do plan to discuss it with engineers to better understand the level of effort involved. If we do not move forward with this work, it would not be because I disagree with the value of the idea. It would simply reflect the reality of limited resources and the need to prioritize projects that are likely to have the greatest impact relative to the effort required.

Hello @Mathglot,

Thank you for the proposal. My name is Martin, and I'm an engineer working on the mentorship features. Unfortunately, this would be a fairly involved task. It would mean expanding Special:ManageMentors with a number of extra columns, some of which are difficult to compute. Getting the dates of responses to mentee questions is a good example: while we can find a question easily enough (it's the latest edit tagged as a question), identifying its response is challenging, since the response carries no tag of its own.

We once considered adding an unanswered-questions list to the Mentor dashboard (T279002), which would have forced us to solve exactly these problems, but that work was never prioritized. Given the competing priorities @KStoller-WMF mentioned, I think it's highly unlikely we'd be able to take this on soon.

That said, I ran into the same problem when I was actively working with the Czech community, and put together a script to analyze mentor activity (https://github.com/urbanecm/growth-mentorActivityAudit). It's specific to Czech Wikipedia, but it might be a useful starting point for a similar bot on English Wikipedia.

As far as I can see, the simplest column to compute (the mentor's last edit) should already be available on Special:ManageMentors. Adding the date of the last question should be relatively straightforward too, in theory — though I'd need to double-check, and I'm not sure that value alone would be all that helpful. Either way, I'm sorry we can't take this on in full soon, but I hope the context and partial solutions above are useful. I'm happy to answer any questions you have.

Best,
Martin

@KStoller-WMF

Reaching a certain edit count or tenure does not necessarily mean someone will be an effective, responsive, or engaged mentor.

I certainly concur with that. Even the en.Wiki 's hugely successful 'Teahouse' when it was introduced is now suffering from to some extent from a plethora of less experienced 'hosts'. I do not believe however, that the significant technical work required to calculating and maintaining more nuanced mentor metrics would be conducive to the more immediate needs of basic onboarding of new editors, especially those who register with the sole intention of creating and posting a one-off, inappropriate new article.

Law of the Instrument (Maslow's Hammer): A famous cognitive bias states that if your only tool is a hammer, everything looks like a nail. Developers spend years mastering specific programming languages and architectures; consequently, their immediate mental shortcut is to use code to solve organizational or procedural inefficiencies or fulfill possible unnecessary wishes. While software can automate and optimize many tasks, treating every issue as a programming challenge can often lead to over-engineering or building unnecessary features that complicate the user's experienc - and this is precisely what has happened with the User Home Page and mentorship programe which despite its long period in development, has yet to have any measurable effect in these areas of en.wiki, the Foundation's flagship project.

The editing community, especially the contingent of experienced users who are directly concerned with patrolling and maintaining article quality, is possibly the better judge of 'the need to prioritize projects that are likely to have the greatest impact relative to the effort required.'. .