Why Do People Give Up FLOSSing?

Courtney Miller, David Gray Widder Christian Kästner, Bogdan Vasilescu

!1

A Study of Contributor Disengagement in Open Source

OSS ’19, May 26th 2019, Montréal, Canada

F L O S S

Modern Software

!2

Many OSS Projects are Analogous to Critical Infrastructure

!3

Survival Rate Among FLOSS Projects is Very Low

Having Contributors in

Drivers Seat is Critical to

Project Survival

Bad

Good!4

41% of failed OSS projects cited developer issuesCoelho, J., & Valente, M. T. (2017, August)

Contributor Onboarding

!5

Contributor On-Boarding Has Been Thoroughly Researched

Established Contributor

Uh oh…

!6

The Body of Work on Established Contributors is Much Smaller

Predictive Factors

?

FLOSS FLOSS!7

We’re Interested in Exploring Factors to Predict Disengagement from OSS

Engaged Contributor

Disengaged Contributor

!8

+

General Methodology

Literature Review

Operationalization of Predictive Factors

Survival Model

Survey of Disengaged Contributors

All Public GH Contributors

Regular and Active Contributor

Publicly Displayed Contact Information

Random Sample of Population

Survey Sample!9

Used GHTorrent to Collect Survey Sample

All Public GH Contributors

Regular and Active Contributor

Publicly Displayed Contact Information

Random Sample of Population

Survey Sample!10

Used GHTorrent to Collect Survey Sample

!11

Identifying Regular and Active Contributors Who Disengaged

> 100 Commits > 100 Commits> 100 Commits < 5 Commits

Present

6 mo. 6 mo. 6 mo. 6 mo.

All Public GH Contributors

Regular and Active Contributor

Publicly Displayed Contact Information

Random Sample of Population

Survey Sample!12

Used GHTorrent to Collect Survey Sample

All Public GH Contributors

Regular and Active Contributor

Publicly Displayed Contact Information

Random Sample of Population

Survey Sample!13

Used GHTorrent to Collect Survey Sample

All Public GH Contributors

Regular and Active Contributor

Publicly Displayed Contact Information

Random Sample of Population

Survey Sample!14

Used GHTorrent to Collect Survey Sample

“Could you help us understand your reasons for reducing your contributions to GitHub projects?”

!15

Sent Out Simple One-Question Survey to Sample

21.5% Response Rate

!16

We Used Card Sorting Analysis to Parse Raw Survey Responses

70% 34% 25%

Survey Says Out of 151 Respondents…

!17

Macro Survey Results

Occupational Issues

Social Issues

Technical Issues

Breakdown of Survey Responses

!18

!19

+

General Methodology

Literature Review

Operationalization of Predictive Factors

Survival Model

Survey of Disengaged Contributors

Related Research

When contributor works …

… what contributor does.

!20

We Performed Literature Review

Office Hours Nights & Weekends

More Support Work Less Support Work

H1:

VS.

VS.

!21

We Were Interested In Stratifying Survey Responses

H2:

!22

+

General Methodology

Literature Review

Operationalization of Predictive Factors

Survival Model

Survey of Disengaged Contributors

07:00 19:00

Office Hours

Nights & WeekendsFLOSS

workingHours

!23

More Support Work Less Support Work

New Code!

supportWork

!24

CommitsPull Requests Issues

Etc.

Stratified Survey Results

!25

Stratified Survey Results

!26

Stratified Survey Results

!27

!28

TransitionFoundA Good Example: Brett Cannon

!29

TransitionFoundIdentifying Transitions

!30

+

General Methodology

Literature Review

Operationalization of Predictive Factors

Survival Model

Survey of Disengaged Contributors

But What Exactly Is A Survival Model?

!31

(Days)

Organizational Affiliation ListedNo Organizational Affiliation Listed

vs.

Activity WorkingHours SupportWork Project Popularity

Organizational Affiliation

Team Size

Transition Found

!32

Survival Model Factors

Survey Respondents with Transition Data

Control Group

Survival Model Participants

+

!33

Participants with Transition 2.48 Times More Likely to Disengage

!34

“I’ve moved to more of a product owner role in my position, so less time to develop.”(P50)

“I started employment with a company that didn’t really like open sourcing their projects.” (P25)

!35

VS.

“[It’s] quite simple. I was active because the project was sponsored by my employer, i.e., my job. I'm not active on GitHub

because I've moved to another project.”(P75)

Working Predominantly During Office Hours Increased Risk of

Disengagement by Factor of 2.20

Elective Participation

Mandatory Participation

!36

Operationalization of Predictive Factors Provided Promising Results

Data Driven Operationalization

Of Predictive Factors

Valuable Insights About Disengaged OSS

Contributors

To learn more about sustained participation of contributors in FLOSS…

!37

Friday, May 31st at 11:20, Sophie Qiu Going Farther Together:

The Impact of Social Capital on Sustained Participation in Open Source

!38

Questions?

Occupational reasons were most commonly cited in survey, specifically occupational transitions and novel transitionFound operationalization proved successful factor in survival model

Used operationalizations of potential disengagement factors to (1) categorize survey respondents; and (2) predict disengagement

Established contributors who do different kinds of work and perform said work at different times tend to cite different reasons for their disengagement.

Factors such as project popularity, when contributors work, and how much they work can also be used to predict established contributor disengagement from FLOSS

Key Takeaways

Courtney Elta Miller

Courtney.Miller17@ncf.edu

@CourtneyElta