The Cost of Collaboration for Code and Art: Evidence from Remixing

40
The Cost of Collaboration for Code and Art Evidence from a Remixing Community Benjamin Mako Hill 1,2 [email protected] Andrés Monroy-Hernández 2,3 [email protected] 1 Massachusetts Institute of Technology 2 Berkman Center, Harvard University 3 Microsoft Research February 26, 2013 revision: b047d49 (2013/02/26)

Transcript of The Cost of Collaboration for Code and Art: Evidence from Remixing

Page 1: The Cost of Collaboration for Code and Art: Evidence from Remixing

The Cost of Collaboration for Code and ArtEvidence from a Remixing Community

Benjamin Mako Hill1,2

[email protected]és Monroy-Hernández2,3

[email protected]

1 Massachusetts Institute of Technology2 Berkman Center, Harvard University

3 Microsoft Research

February 26, 2013

revision: b047d49 (2013/02/26)

Page 2: The Cost of Collaboration for Code and Art: Evidence from Remixing

Peer Production and Remixing

Benkler (2006); Raymond (1999); Lessig (2008); von Hippel (2005); Giles (2005);Wilkinson and Huberman (2007); Kittur and Kraut (2008), etc.

2 / 17

Page 3: The Cost of Collaboration for Code and Art: Evidence from Remixing

Remixing

The reworking and recombination of existing creative artifacts.Most commonly in reference to music, video, and interactive media.

É Widespread, and an important new communication modality (e.g.,Manovich 2005; Lessig 2009)

É Especially among use youth (Jenkins 2006; Palfrey and Gasser 2008)

3 / 17

Page 4: The Cost of Collaboration for Code and Art: Evidence from Remixing

Criticism of Peer Production

Others have suggested that peer productionand remixing are amateurish and poor quality

(e.g., Lanier, 2010; Keen, 2007).

4 / 17

Page 5: The Cost of Collaboration for Code and Art: Evidence from Remixing

“Could Hamlet have been written by a committee,or the Mona Lisa painted by a club? Could theNew Testament have been composed as aconference report? Creative ideas do not springfrom groups. They spring from individuals. Thedivine spark leaps from the finger of God to thefinger of Adam.”

– A. Whitney Griswold (1957)

5 / 17

Page 6: The Cost of Collaboration for Code and Art: Evidence from Remixing

Research Questions

RQ1: Are remixes, on average, higher quality

than single-authored works?

RQ2: Are code-intensive remixes, on average,

higher quality than media-intensive

remixes?

6 / 17

Page 7: The Cost of Collaboration for Code and Art: Evidence from Remixing

Research Questions

RQ1: Are remixes, on average, higher quality

than single-authored works?

RQ2: Are code-intensive remixes, on average,

higher quality than media-intensive

remixes?

6 / 17

Page 8: The Cost of Collaboration for Code and Art: Evidence from Remixing

Research Site

É A computer programmingenvironment

É An online community for

sharing projects

É Designed and built around

collaboration through remixing

7 / 17

Page 9: The Cost of Collaboration for Code and Art: Evidence from Remixing

Research Site

É A computer programmingenvironment

É An online community for

sharing projects

É Designed and built around

collaboration through remixing

7 / 17

Page 10: The Cost of Collaboration for Code and Art: Evidence from Remixing

Research Site

É A computer programmingenvironment

É An online community for

sharing projects

É Designed and built around

collaboration through remixing

7 / 17

Page 11: The Cost of Collaboration for Code and Art: Evidence from Remixing
Page 12: The Cost of Collaboration for Code and Art: Evidence from Remixing
Page 13: The Cost of Collaboration for Code and Art: Evidence from Remixing
Page 14: The Cost of Collaboration for Code and Art: Evidence from Remixing
Page 15: The Cost of Collaboration for Code and Art: Evidence from Remixing
Page 16: The Cost of Collaboration for Code and Art: Evidence from Remixing
Page 17: The Cost of Collaboration for Code and Art: Evidence from Remixing

Data and Measures

1,271,085 projects shared before April 1, 2011.

Outcome:

É Loveits

Independent Variables:

É Collaborativeness (Is project a remix?).

É Control blocks (similar to lines of code)

É Media elements including image and sound files.

Additional controls:

É Number of views

É Creator tenure, gender and age

10 / 17

Page 18: The Cost of Collaboration for Code and Art: Evidence from Remixing

Data and Measures

1,271,085 projects shared before April 1, 2011.

Outcome:

É Loveits

Independent Variables:

É Collaborativeness (Is project a remix?).

É Control blocks (similar to lines of code)

É Media elements including image and sound files.

Additional controls:

É Number of views

É Creator tenure, gender and age

10 / 17

Page 19: The Cost of Collaboration for Code and Art: Evidence from Remixing

Data and Measures

1,271,085 projects shared before April 1, 2011.

Outcome:

É Loveits

Independent Variables:

É Collaborativeness (Is project a remix?).

É Control blocks (similar to lines of code)

É Media elements including image and sound files.

Additional controls:

É Number of views

É Creator tenure, gender and age

10 / 17

Page 20: The Cost of Collaboration for Code and Art: Evidence from Remixing

Data and Measures

1,271,085 projects shared before April 1, 2011.

Outcome:

É Loveits

Independent Variables:

É Collaborativeness (Is project a remix?).

É Control blocks (similar to lines of code)

É Media elements including image and sound files.

Additional controls:

É Number of views

É Creator tenure, gender and age

10 / 17

Page 21: The Cost of Collaboration for Code and Art: Evidence from Remixing

Methods

Negative binomial regression model on a count of loveits.

loveits = β+β log blocks +β log media+βisremix +

β log blocks× isremix +β log media× isremix +

βage+βjoined +βfemale+β log prevloveits +β log views

11 / 17

Page 22: The Cost of Collaboration for Code and Art: Evidence from Remixing

Results: Plots of Prototypical Projects

Blocks Media Elements

0.10

0.15

0.20

0.25

0.30

0.35

50 100 150 200 10 20 30 40

Love

its

De Novo

Remix

12 / 17

Page 23: The Cost of Collaboration for Code and Art: Evidence from Remixing

Results: Plots of Prototypical Projects

Blocks Media Elements

0.10

0.15

0.20

0.25

0.30

0.35

50 100 150 200 10 20 30 40

Love

its

De Novo

Remix

RQ1: Remixes are rated lower.

12 / 17

Page 24: The Cost of Collaboration for Code and Art: Evidence from Remixing

Results: Plots of Prototypical Projects

Blocks Media Elements

0.10

0.15

0.20

0.25

0.30

0.35

50 100 150 200 10 20 30 40

Love

its

De Novo

Remix

RQ1: Remixes are rated lower.

RQ2: Code-intense remixes are rated higher.Media-intense remixes are rated lower.

12 / 17

Page 25: The Cost of Collaboration for Code and Art: Evidence from Remixing

Example: Code Intense Project

Original Remix

Remix received about the same loveits than theantecedent.

13 / 17

Page 26: The Cost of Collaboration for Code and Art: Evidence from Remixing

Example: Code Intense Project

Original Remix

Remix received about the same loveits than theantecedent.

13 / 17

Page 27: The Cost of Collaboration for Code and Art: Evidence from Remixing

Example: Media Intense Project

Original Remix

Remix received less loveits than antecedent.

14 / 17

Page 28: The Cost of Collaboration for Code and Art: Evidence from Remixing

Example: Media Intense Project

Original Remix

Remix received less loveits than antecedent.

14 / 17

Page 29: The Cost of Collaboration for Code and Art: Evidence from Remixing

Limitations

É Ratings are a single and imperfect measure of

quality

É Code & Media≈ Functional & Artistic

É Remixes and antecedents projects may

compete for loveits

É Generalizability beyond Scratch, beyond kids,

beyond remixing.

15 / 17

Page 30: The Cost of Collaboration for Code and Art: Evidence from Remixing

Limitations

É Ratings are a single and imperfect measure of

quality

É Code & Media≈ Functional & Artistic

É Remixes and antecedents projects may

compete for loveits

É Generalizability beyond Scratch, beyond kids,

beyond remixing.

15 / 17

Page 31: The Cost of Collaboration for Code and Art: Evidence from Remixing

Limitations

É Ratings are a single and imperfect measure of

quality

É Code & Media≈ Functional & Artistic

É Remixes and antecedents projects may

compete for loveits

É Generalizability beyond Scratch, beyond kids,

beyond remixing.

15 / 17

Page 32: The Cost of Collaboration for Code and Art: Evidence from Remixing

Limitations

É Ratings are a single and imperfect measure of

quality

É Code & Media≈ Functional & Artistic

É Remixes and antecedents projects may

compete for loveits

É Generalizability beyond Scratch, beyond kids,

beyond remixing.

15 / 17

Page 33: The Cost of Collaboration for Code and Art: Evidence from Remixing

Takeaways and Final Thoughts

Remixes are rated lower than de novo projects byScratch users.

É More code = smaller gap

É More media = wider gap

16 / 17

Page 34: The Cost of Collaboration for Code and Art: Evidence from Remixing

Takeaways and Final Thoughts

Remixes are rated lower than de novo projects byScratch users.

É More code = smaller gap

É More media = wider gap

16 / 17

Page 35: The Cost of Collaboration for Code and Art: Evidence from Remixing

References

Benkler, Y. (2006, May). The Wealth of Networks: How Social Production Transforms Markets andFreedom. Yale University Press.

Giles, J. (2005, December). Internet encyclopaedias go head to head. Nature 438(7070), 900–901.

Griswold, A. W. (1957, June). Baccalaureate address.

Keen, A. (2007, June). The Cult of the Amateur: How Today’s Internet is Killing Our Culture (3rdPrinting ed.). Crown Business.

Kittur, A. and R. E. Kraut (2008). Harnessing the wisdom of crowds in wikipedia: quality throughcoordination. San Diego, CA, USA, pp. 37–46. ACM.

Lanier, J. (2010, January). You Are Not a Gadget: A Manifesto (1 ed.). Knopf.

Lessig, L. (2008, October). Remix: Making Art and Commerce Thrive in the Hybrid Economy.Penguin Press HC.

Raymond, E. S. (1999). The cathedral and the bazaar: Musings on Linux and open source by anaccidental revolutionary. Sebastopol, CA: O’Reilly and Associates.

von Hippel, E. (2005). Democratizing innovation. Cambridge, Massachusetts: The MIT Press.

Wilkinson, D. M. and B. A. Huberman (2007). Cooperation and quality in wikipedia. In Proceedingsof the 2007 international symposium on Wikis, Montreal, Quebec, Canada, pp. 157–164. ACM.

17 / 17

Page 36: The Cost of Collaboration for Code and Art: Evidence from Remixing

Example andExample andData AppendixData Appendix

Page 37: The Cost of Collaboration for Code and Art: Evidence from Remixing

Empirical Model

loveits = β+β log blocks +β log media+βisremix +

β log blocks× isremix +β log media× isremix +

βage +βjoined +βfemale +β log prevloveits +β log views

[RQ1] Are remixes, on average, rated as higher qualitythan de novo projects?

[RQ1] Are code-intensive remixes, on average, rated ashigher quality than media-intensive remixes?

19 / 17

Page 38: The Cost of Collaboration for Code and Art: Evidence from Remixing

Empirical Model

loveits = β+β log blocks +β log media+βisremix +

β log blocks× isremix +β log media× isremix +

βage +βjoined +βfemale +β log prevloveits +β log views

[RQ1] Are remixes, on average, rated as higher qualitythan de novo projects?

[RQ1] Are code-intensive remixes, on average, rated ashigher quality than media-intensive remixes?

19 / 17

Page 39: The Cost of Collaboration for Code and Art: Evidence from Remixing

Empirical Model

loveits = β+β log blocks +β log media+βisremix +

β log blocks× isremix +β log media× isremix +

βage +βjoined +βfemale +β log prevloveits +β log views

[RQ1] Are remixes, on average, rated as higher qualitythan de novo projects?

[RQ1] Are code-intensive remixes, on average, rated ashigher quality than media-intensive remixes?

19 / 17

Page 40: The Cost of Collaboration for Code and Art: Evidence from Remixing

Regression Results

(Intercept) −3.359∗ −4.828∗

log blocks 0.166∗ 0.076∗

log media 0.384∗ 0.086∗

log isremix 0.279∗ −0.091∗

female 0.000 0.284∗

age −0.006∗ −0.002∗

joined −0.020∗ −0.008∗

prevloveits 0.568∗ 0.076∗

log blocks×isremix 0.059∗ 0.021∗

log media×isremix −0.305∗ −0.103∗

views 1.226∗

N 1239470 1239470log L -901476 -707714∗∗ indicates significance at p< .01

20 / 17