[PPT] - A S ocial Capit al Perspect ive of Participant Contribution in Open PowerPoint Presentation

SLIDE 1

A “ S

cial Capit al Perspect ive” of Participant

Contribution in Open Source Communities –The Case of Linux

Presentation for Singapore Management University

July 14, 2011

Myong Rae (Ray) Chang, Ph.D.

SLIDE 2

Motivation Research Model Results Analyses Implications Limitations & Future Study Method Theoretical Challenges

Thread-level Dynamic Process Network Measures Content Analysis PLS LINUX & OSSD

Agenda

OSS Development Process: Network Capital  Contribution

Four Dimensions of Network Capital

2

SLIDE 3

The Linux Kernel

Linus Tovalds, Jan. 2011

“ Not j ust Android. What I’ ve f ound t hat has been most f un f or me has been when people are using Linux in ways t hat I don’t use it or in ways t hat I never int ended it t o be used, people using it in embedded areas, and wit h cellphones like Android but also all t he crazy people using it in print ers and TVs.”

Desktop/Laptop: , Server/Mainframe: , Supercomputer:
Smartphone (Android): [rough averages from multiple sources, worldwide, 2010 ]

3% 65% 92% 25%

3

SLIDE 4

Linux Open Source Community

The largest OSS development community
Being evolved dynamically at every second
A “virtual” workplace open to any participants from any

place in the world at any time (but only “ hackers” survive)

“ Open” to any contribution on “ voluntary” basis (note:

many are now paid workers from Linux-related companies)

All “peer-reviewed” process: from ideas to codes & feedbacks
Driven by crowd wisdom, not by dedicated plans or profits
Highly technical and rigorous discussions: it should work!
Administered by “ Maintainers” of numerous subsystems

4

SLIDE 5

http:/ / www.gossamer- threads.com/ lists/ linux/ kernel/ 655933

 A hist ory of int ellect ual knowledge-exchanges on a single specific t opic

Threaded Discussion

5

SLIDE 6

A: [RFC] CPU controllers? B: Re: [RFC] CPU controllers? C: Re: [RFC] CPU controllers? D: Re: [RFC] CPU controllers? B: Re: [RFC] CPU controllers?

Node: each part icipant Link or tie: relat ion of in-reply-t o (wit h t he at t ribut e of message mult iplicit y) A B C D

Note: Although each message is “ broadcast” to all subscribers of the mailing list,

nly respondents to the message would take special meaning w.r.t . the topic;

i.e., the network is a construction based on “reciprocity” relationship, a characteristics

f online relationship.

init iat or t ie st rengt h (mult i-link)

A Message-exchange “Network” of a Thread = a knowledge exchange network built on reply-to relations formed by the messages within a thread

A Network Representation

6

SLIDE 7

Thread# # 6559 55933 Thread# # 91 913809 Thread# # 376114 14

Patterns of Thread Network

A “ signat ure” of how t hey have communicat ed/ collaborat ed

n a specific t ask

7

SLIDE 8

Netwo work rk S Size % M Messa ssages (Tim ime) e)

Evolution of A Thread Network

Thread# # 91 913809 seed eed 8

SLIDE 9

0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 110% 10 20 30 40 50 60 70 80 90 100 % Messages Relative Change of Network Measure Density

Avg. Geodesic Distance

Degree Centralization Betweenness Centralization

Dynamics of Network Measures

9

SLIDE 10

Netwo work rk S Size Growi wing Stable le Mature re % M Messa ssages

Evolution of A Thread Network

We are int erest ed in “ early-st age” net work building t o influence t he t hread performance in t he lat er st ages

10

SLIDE 11

Research Questions

 Then, what types of network capital will be associated with

participant contribution in terms of quantity & quality?



Will the early-stage accumulation of network capital affect participant contribution in the later stage?

 During the lifecycle of OSSD, how can we help to elicit

“ more” and “ better” participant contribution?

11

SLIDE 12

Theoretical Background

 Social Capital Perspectives in Online Communities

S

t ruct ural propert ies (posit ion, st ruct ure)

 individual’s knowledge sharing behavior

[Cross & Cummings 2004; Nerkar & Paruchuri, 2005]

Role of relat ional propert ies  learning and knowledge t ransfer

[Hansen 1999; Uzzi & Lancast er 2003]

Various t ypes of social capit al(st ruct ural, cognit ive, relat ional)

 knowledge sharing of individuals and groups

[Wasko & Faj ar 2005 in elect ronic net works of pract ice; Kuk 2006 in an OS S communit y]

 Theoretical Challenges

Fragment ed view vs. a “ comprehensive” framework: a holist ic

model int egrat ing various dimensions of social capit al

Extant isolated element-level approach (actor, dyadic link, ego-centric

[Borgatti & Foster 2003]) vs. underst anding of act ors’ “ collective

behavior” : a net work (e.g., t hread) level approach

“ S

t at ic” snapshot approach vs. “ Dynamic” aspect of social capit al

12

SLIDE 13

High Low

Structural Capital

Network centralization

Relational Capital

Network strength

Governance Capital

Administrator participation

Node Link Network

Four Dimensions of Network Capital

Dynamic Capital

Network growing speed

Time

T=6 hrs T=12 hrs

13

SLIDE 14

Network Centralization Network Strength Administrator Participation Network Growing Speed

H1a H1b H2a H2b H3b H3a H4b H4a

Contribution Quality Contribution Quantity

H5

positive negative

OSS Performance (2/3) Growing-stage Network Capital (1/3)

Research Model: PLS

Code in Initial Message Inhibiting Climate

14

SLIDE 15

Hypotheses 1a/1b

Network Centralization

H1a H1b

Contribution Quality Contribution Quantity

 Two Contrasting Views on Centralization

Discourages diverse views
Negative impact on creativity
Reduces autonomy of participants
Rapid diffusion of innovative knowledge
Easy access to experts with lower cost
Integration of diverse ideas
Benefits outweigh the costs from the lack of idea diversity
Hub-structure facilitates member contribution
Continuous review and feedback systems
Multiplicity of views and ideas more easily integrated

 In Thread-level Collaboration:

vs.

Structural Capital

vs.

15

SLIDE 16

Hypotheses 2a/2b

Network Strength

H2a H2b

Contribution Quality Contribution Quantity

 Tie Strength (link thickness)

Useful conduits for knowledge exchange (in many social network studies)
Essential for substantive contribution
A network is stronger if containing more strong ties.
Building a normative environment fostering collaboration and coordination
A sense of “ reciprocity” ensures continuing supportive exchanges and

generating in-depth discussions.

 Network-level Strength Relational Capital

vs.

16

SLIDE 17

Hypotheses 3a/3b

Admin Participation

H3a H3b

Contribution Quality Contribution Quantity

 In Traditional Organization Settings

Leader’s involvement is effective for affective and continuance contributions.
S

trong governance encourage self-concept-based motivation.

Admin’s tight control evokes a cathedral type of decision making structure.
Undermining participants’ autonomy and sense of ownership [von Krogh 2003].
Members do not participate when most work is conducted by a leader.

 In OSS Communities Governance Capital

vs.

17

SLIDE 18

Hypotheses 4a/4b

Network Growing Speed

H4a H4b

Contribution Quality Contribution Quantity

 Research on Interpersonal Communications

The quantity and quality of exchanged knowledge are highly associated with

the rate at which knowledge is delivered [Carlson & Zmud 1999].

Faster responses allow receivers to act upon in a timely manner.
Level of detail or extensiveness is often more important than response time.
Rapid responses might decrease the perceived value of knowledge exchanged.
S

low responses tend to be more rational and cognitive (faster ones be more emotional).

Developers place more weight on “ accuracy” : slow responses reduce uncertainty and

equivocality by providing more complete knowledge [Weiss et al. 2006].

 In Knowledge-intensive Online Communities Dynamic Capital

vs.

T=6 hrs T=12 hrs

18

SLIDE 19

Hypotheses 5 and Controls

Increasing # of proposed ideas

 higher probability of quality ideas

[Barki & Pinsonneault 2001]

 Quantity  Quality

Contribution Quality Contribution Quantity

H5 Code in Initial Message Inhibiting Climate

Initial message with codes proposes

more concrete ideas to call for more contribution [Roberts et al. 2006]

Inhibiting culture structurally prevent

members’ contribution [Bogozzi & Dholakia 2006]

 Controls 19

SLIDE 20

 RFC (Request For Comments) t hreads collect ed from archive

S

uggest ion/ discussion/ collaborat ion on new innovat ive ideas & feat ures

6,852 RFC t hreads during Jan. 2000 ~ Jun. 2008

(over 90%

f t hem had less t han 15 messages)
Used t hreads wit h enough(>25) messages  223 RFC threads
For each t hread, a mat ching set of a net work file and a t ext file of

messages were creat ed (only original cont ent s used aft er removing quot es, program codes, et c.) 223 RFC Threads LINUX Communit y 223 Net works (growing st age) LIWC 223 t ext files of whole messages Comput ing Measures Net work Measures OS S Cont ribut ion C/ C++ C/ C++

Method: Data Mining

20

SLIDE 21

Growing Stage* (Initial 1/ 3)

f a Thread

Operationalizing Network Capital

Sample Period Construct Item Definition

Network Centralization degree centralization betweenness centralization Network Strength multi-link ratio multi-message ratio Administrator Participation admin-node ratio admin-message ratio Network Growing Speed message entering rate Xi: degree/ betweenness centrality of node i X*: maximum degree/ betweenness centrality Centralization =

∑ ∑

− −

i i i i

X X Max X X ) ( ) (

* *

proportion of strong ties in a thread = # strong ties / # total links message overloading per link in a thread = # total messages / # total links proportion of administrators in a thread = # administrators / # total participants proportion of administrators’ messages = # messages of admins/ # total messages log (message occurring rate per unit time) = log (# total messages / total elapsed hours)

* Note: for capturing the growing-stage dynamics, during the sample period we measured each item three times with the increase of message volume and used averaged values in the analysis.

21

SLIDE 22

 Adopting Objective Content Analysis

An aut omat ed t ext analysis t ool, LIWC, t hat calculat es t he degree t o which

people use different cat egories of words

Int ernal dict ionary: 4,500 words and word st ems, 32 word cat egories

(average word-capt ure percent age: over 86% )

Cat egories: cognit ive, posit ive/ negat ive emot ions, insight , causat ion, et c.
Text cont ent  word count or percent age by cat egory
Broadly used in social sciences, linguist ics, healt h, informat ion science, et c.

 Contribution Quality: percent age of words in cat egory “ INSIGHT”

Words associat ed wit h “ learning” and “ understanding” [J. Pennebaker et al.]
Act ively t hinking about somet hing in a self-reflect ive or insight ful way
Generally a class of verbs called ont ological verbs and included words:

underst and, realize, know, meaning, et c.

Healt h and immune funct ion [Klein & Boals, 2001; Pet rie et al. 1998]
Experient ial learning [Abe 2009]

 Contribution Quantity: word count per message in a t hread

Quantifying Contribution

22

SLIDE 23

Stable and Mature Stage (Later 2/ 3)

f a Thread

Operationalizing Contribution & Controls

Sample Period

Construct

Item Definition

Contribution Quantity

word count per message

log (word count per message for a thread)

Growing Stage (Initial 1/ 3)

f a Thread

Contribution Qualtiy

Insight word percentage

percentage of words in a thread which belong to LIWC category “ Insight”

Code in Initial Message

ini-code

1 if program codes are found in the thread- initiating message 0 otherwise

Inhibiting Climate

Inhibit word percentage

percentage of words in a thread which belong to LIWC category “ Inhibit”

Dependent Variables Control Variables 23

SLIDE 24

Results: Structural Model

0.068+ (t=1.0842) 0.161* (t=2.4805) 0.053 (t=0.6850)

0.131*

(t=-2.1806)

0.191**

(t=-3.1556)

0.189**

(t=-2.7374)

0.021

(t=-0.3214) 0.121+ (t=1.7694) 0.001 (t=0.0179)

Network Centralization Network Strength Administrator Participation Network Growing Speed Contribution Quality Contribution Quantity

positive negative R2=.120 R2=.085

0.221**

(t=-3.7493)

0.033

(t=-0.5106)

Code in Initial Message

0.115 (t=1.4865)

Inhibiting Climate

0.030 (t=0.3780)

Notes: t-statistics are in parentheses. +p<0.1; *p<0.05; **p<0.01

24

SLIDE 25

Findings and Implications

 Supported Hypotheses

Centralized communicat ion st ruct ures can serve as effect ive conduit s

for qualit y discussions only when “ nat urally” est ablished by peer members; administ rat ors’ “ art ificial” int ervent ions/ cont rols would result in adverse

ut comes (H1, H3)
Administrators need t o keep t heir int ervent ion t o a minimum during

t he growing st age of a t hread; non-admin peer part icipant s wit h relevant expert ise should lead t he discussion (H1, H3)

For more cont ribut ion, part icipant s need fully ut ilize pre-established

communicat ion relat ionships (H2) and take time in responses t o provide accurat e and det ailed informat ion (H4)

25

Centralization Strength Administrator Growing Speed Contribution Quality Contribution Quantity Code Inhibit

SLIDE 26

Findings and Implications

 Unsupported Hypotheses

H1a: Mixed effect s from t wo opposing forces (diversit y vs. learning)
H2b: S

t rong t ies may produce many redundancies, prevent ing t he generat ion of innovat ive idea

H4b: Cont ribut ion qualit y does not aut omat ically occur wit h t he

progression of t ime

H5: Voluminous discussions are not essent ial for qualit at ive

cont ribut ions

26

Network Centralization Network Strength Administrator Participation Network Growing Speed Contribution Quality Contribution Quantity Code Inhibit

SLIDE 27

Limitations and Future Research

 Limitations

In st aging t he OS

S D lifecycle, t he approach using real time-clock needs t o be employed along wit h message volume approach

Automated text analysis and obj ect ive assessment st ill need more

validat ion t o be used in IS cont ext

Limited variance explanation suggest s t he exist ence of unobserved

fact ors affect ing t he members’ cont ribut ion

 Future Research

Abundant research challenges exist t o advance t he underst anding of t he

evolutionary aspects of OSSD collaboration in net work/ t hread level

Exploring t he changes in the structural formations wit h t he new ent ry
f part icipant s; e.g., wit h whom do t he new ent rant s t end t o int eract ?

Among t he new ent rant s, who cont ribut es t he most in t erms of net work capit al and part icipant cont ribut ion and why?

Network simulation models for t he dynamic process of online