Quantitative Community Management

Asheesh Laroia


Executive Director, OpenHatch

Overview

About me

  • 2000: DeCSS
  • 2001: Read GNU Manifesto
  • 2001: Seth David Schoen
  • 2006: Met him
  • 2007: Concluded the community is too small

  • 2009: Founded OpenHatch

Topic: Who are we,
as a community?

FLOSS survey, 2001

Rishab Aiyer Ghosh

Rüdiger Glott

Bernhard Krieger

Gregorio Robles


International Institute of Infonomics, Maastricht


Gender stats


1.1% women
in FLOSS survey


1.6% women
in separate FLOSS-US survey

Survey methodology


“Rather than selecting out a small, well-controlled sample...

we allowed respondents to decide for themselves whether they should be considered “developers”..."
"Our goal has been to analyze the entire... community."


Topic: What are our projects like, on the whole?

"Who Writes Linux?" report

  • Yearly from the Linux Fondation,
    these numbers re: 2.6.30

  • Changes per hour: 6

  • # of lines: 11 million

  • # of developers: 1,150

  • # of companies: 240

All SourceForge Projects (n=145,850)


“Mature” and “Production” SourceForge Projects (n=29,821)


SF.net Projects Downloaded >=99 times (90th %ile) 

Scratch projects 1+ year after publication (n=249,428)


Google Code Projects (n=195,834)


Active Google Code Projects (n=74,398)


Github public projects (developers are “watchers”) (n=265,088)

Radical flamebait questions

"Does Ghosh's survey find fewer women because it mostly surveyed people who start projects?"


"Are the men in FLOSS and the women generally using separate hosting services?"


"Are women under-represented because, as a group, they were less likely to fill out the survey?"
Reflections: What are we measuring, and why?

  • Academic factoids

  • Not actionable

  • Being measured by people who don't have an interest in the results.

Radical flamebait conclusion


"Opt-in surveys are hopelessly broken,
unless you know, very clearly,
who has responded and who did not."

- Benjamin Mako Hill

Radical counter-flamebait:

± 50% is good
enough for activists




Radical counter-flamebait:

± 50% is good
enough for activists



But do we know it's +/- 50%?

Radical counter-flamebait:

± 50% is good
enough for activists



How do we measure progress?

Going forward,
let's try to
be useful.

2008 Wikipedia survey


  • For 1 week, a link on top of every page
    (I don't remember seeing it...)

  • Goals of survey: Answer...

        Why do people start+stop editing?

        Do people know WMF is a non-profit?

        What are Wikipedia editors' demographics?
        
  • Collaboration between WMF and UNU-MERIT

Basic demographics

Age (overall)

  • 25% younger than 18
  • 50% younger than 22


Gender

  • Readers: 31% female, 69% male
  • Editors: 13% female, 87% male


Language

  • 26% Russian
  • 25% English

Wikipedia Editor Survey, 2011

  • "The first ever semi-annual survey of
    Wikipedia editors"

  • "conducted on Wikipedia and presented
    to logged-in users"

  • Results: 8.5% female.

  • Is it getting worse?

  • Will we ever know?

comScore vs.
UNU-MERIT


UNU-MERIT: 26% Russian

comScore: 2.5% Russian


Pew Survey, 2010

 Goal: understand Internet use
and adoption in the United States

Method: Call random USians over 18

Results: % of US (not % of WP)

Afterward: Publish everything

Pew's Wikipedia demographics

Age

  • 18-29: 62%
  • 30-49: 52%
  • 50-64: 49%
  • 65+: 33%

Gender

  • Male: 56%
  • Female: 50%

Pew vs. UNU-MERIT


Gender (UNU-MERIT)

  • Readers: 31% female, 69% male


Gender (Pew)

  • Readers: 47% female, 53% male

Other discrepancies

  • Age, marital status, education level, ...

Data recovery

  • Adjust response data to match Pew demographics, using logistic "propensity score" to model non-random selection.

  • Female editors: 12.7% => 16.1%

  • US female editors: 17.8% => 22.7%

  • Credit: Benj. Mako Hill and Aaron Shaw
    (Search: [hill shaw gender wikipedia pew])

What they say
vs.
What they do

  • Wikipedia editor survey 2011:
    • 70% say receiving a Barnstar
      makes them more likely to edit.

  • Shaw & Hill, 2012 (Shaw dissertation)
    • Measure edit rate changes over
      5 weeks pre and post
    • Net -1.72 edits per week change
    • "Movers": +3
    • "Non-movers": -5
    • Search: [shaw shaw interactional
      account dissertation]

Topic: wikiHow demographics and more


  • Inspired (and shocked) by Wikipedia
    Editor Survey results

  • Wondered if they had the same
    lack of gender diversity

  • Ran a survey!

Survey methodology

  • Over three weeks, find active users

  • Send them a talk page message

  • ~50% response rate; N=126

  • Sent by the wikiHow community manager

wikiHow demographics

  • 56% of respondents were female.

  • 52% are 15 or younger.
    24% are 16-25.

  • The older the contributor, the
    more likely to be male.

How to increase
data quality

  • Ask readers to fill out the same survey.

  • Adjust editor response rate by
    readers' response/non-response
    proportions.

Questions about
wikiHow data


  • 50% of survey respondents under 15?

  • Or 50% of age respondents under 15?

  • Was gender mandatory to fill in?

  • Which editing levels were more/less
    likely to respond?

    Questions about
    wikiHow data


    • 19/123 did not fill out age

    • Gender was required
      • (did people refuse to answer
        because of that?)

    • Which editing levels were more/less
      likely to respond?
      • We may never know.

      Topic: Why do Thunderbird contributors give back?

      Topic: Behavioral studies

      GNOME Women's Outreach Project

      (or, "The first great FLOSS behavioral study")

      GNOME Women's Outreach Project

      GSOC 2006: 181 applicants


      Women's Summer Outreach Program,
      Started by Hanna Wallach and Chris Ball:
      100 applicants

      Structure: Separate funding,
      same model as GSoC:
      mentored coding internship


      Conclusion: Targeted outreach changes the behavior we see!

      GNOME Women's Outreach Project

      Open questions:

      • Do Women's Outreach Project participants stick around in GNOME similarly to other summer interns?

        Maybe more, maybe less?

      • Answer may lie in Kevin Carillo's Ph.D. thesis
        • but opt-in nature makes that hard

      A hypothetical
      behavioral study

      • Select 200 random users

      • Find out their demographic info

      • Watch their activity levels

      • (this is hypothetical for now)

      2010:
      Open Source
      Comes to Campus

      • ~30% of applicants were women

      • No gender-specific outreach

      • Great 2-day event...
      • ...but we did leave an impact?

      Tracking
      Open Source
      Comes to Campus

      • Compare Github activity against
        other CS students who did not
        attend event




      It worked in Boston


      Clones popping up:
      • PyStar Philly
      • RailsBridge Boston
      • Chicago Python Workshop
      • Columbus Python Workshop
      • Beginners & Friends Python Programming Workshop
        in Auckland, NZ (hi Tim McNamara!)


      Tying them together as
      OpenHatch Affiliated Events

      Limitations of
      $CITY Python Workshop for women + friends

      • Major urban areas, only?

      • Only applies if you can hijack an existing user group

      Changes to Open Source Comes to Campus


      • Work with existing CS club
        (ACM, Women in CS, etc.)

      • Use exit survey to improve
        event

      • Plans to check back in
        with attendees

      Open Source Comes to Campus survey notes

      • Gender as a text field
        has 100% response rate

      • Undergrads really don't
        know git (:

      Topic: Project-driven contributor metric tracking

      Meego community health

      • 2011: Dave Neary and Dawn Foster

      • Goal: Illuminate community activity:
        Bugzilla, mailing lists submissions, wiki edits

      • http://wiki.meego.com/Metrics/Dashboard

      • A thrilling ball of Tomcat, Pentaho, and MySQL

      Wikipedia bot messages
      (or, "Does niceness matter?")



      Huggle!

      N approx. 10,000

      Wikipedia bot messages


      "Changing the tone and language of the generic vandalism warning..."

      • increasing the personalization (active voice rather than passive, explicitly stating that the sender of the warning is also a volunteer editor, including an explicit invitation to contact them with questions)

      • decreasing the number of directives and links

      • and decreasing the length of the message;

      ...led to more users editing articles in the short term

      Wikipedia bot messages

      Being too "nice" can backfire:

      9.6% of editors who received the new version edited in the file namespace at all afterwards.

      For the default, 18.6% went on to make edits to files.

      Nice != Vague

      MediaWiki community health

      • "What are the areas with more activity?"

      • Are we expanding or shrinking?

      MediaWiki community health

      Measure everything

      Debian mentorship, 2009:
      "Four days"

      • Can we review new contributors'
        packages within four days?

        if so, they know what to expect.

      • Package review increased sharply at the start...

      • and then flatlined to its old amount.

      • Follow through is hard.

      Ubuntu Developer Advisory Team

      "This team in terms of UbuntuDevelopment, tries to fulfill the following tasks in the Ubuntu world:
      • Reach out to new contributors, thank them for their work and get feedback.

    • Reach out to people who might be ready to apply for upload rights and help them.

    • Reach out to contributors that went inactive and get feedback from them and offer help."

    • (Source: their homepage, last edited 2012-04-02)

      New Contributor Report

      • DAT asked open-ended questions; 63% response rate

      • 9 love launchpad; 9 dislike it

      • Reviews are "surprisingly painless"

      • Docs are troublesome: “overwhelmed at all the information” and by "contradictory information" that is "difficult to follow in a logical manner"

      • Contributing is "a surprisingly painless process"

      Ubuntu Developer Advisory Team

      The real magic is in Trello cards

      Data from Ultimate Debian Database

      General approach: Make people happy
      rather than tell them what to do

      Trello "demo"
      (whiteboard)



      • But does it work?

      FLOSS is metrics-poor

      • Mirrors make it hard to count Debian users.

      • Web app authors are privacy-sensitive.

      • Follow-through is hard for volunteers.
        • Four days, in Debian
        • do you read your web analytics?

      OpenHatch "greenhouse":
      Ubuntu DAT clone

      • First: Port to Debian

      • Then: Create a control group

      • Finally: Make generic

      • GSoC student:
        David Lu

      Six months of meta-organizing

      GSoC meta mentorship
      (pipe dream)

      • Question: What makes GSoC better?

      • Sub-question: what does a good GSoC mean?

      • More failed students!
      • Are students still active 3-6 months later?
      • Happy mentors.

      GSoC meta mentorship
      (pipe dream)

      • Theory:
        • mentors would benefit from being
          in touch with each other
        • mentors would benefit from being
          asked to report on status

      • Test: Create opt-in meta-mentorship

      • ENOSPC

      Thanks

      • Benjamin Mako Hill, for graphs (and FLOSSmole for the source data)
      • Ubuntu DAT for giving me access
      • Sarah Mei for slide piracy

      Other resources

      • FLOSS Mole
      • metrics-wg

      Stay in touch

      • asheesh@openhatch.org
      • http://lists.openhatch.org/events
      • http://www.rvl.io/paulproteus/lca/
      • Sponsor us


      Do something


      lca

      By paulproteus

      lca

      • 1,030