The Sky Is Not the Limit: Multitasking Across GitHub Projects
[COMPLETE ARTICLE READ IN FULL AT SOURCE, the author copy at cs.ucdavis.edu, including the full regression table and the threats-to-validity section] Vasilescu, B., Blincoe, K., Xuan, Q., Casalnuovo, C., Damian, D., Devanbu, P., & Filkov, V. (2016). 'The Sky Is Not the Limit: Multitasking Across GitHub Projects.' In Proceedings of ICSE '16. doi:10.1145/2884781.2884875. First author Bogdan Vasilescu, then UC Davis, confirmed on the paper's byline. THE KNOWLEDGE-WORK OUTPUT EVIDENCE, CORRELATIONAL AND MORE INTERESTING THAN EXPECTED. DESIGN: commit data on 1,255 prolific developers across 58,092 public GitHub repositories, prolific defined as at least five years between earliest and latest commit and at least 500 commits across at least ten non-fork repositories; the model runs on 78,552 week-level observations from 1,193 developers after outlier removal, a linear mixed-effects model with random effects for developer; THE DEPENDENT VARIABLE IS log(lines of code added) per week; R-squared marginal 0.35, conditional 0.55. FOUR COEFFICIENTS MATTER AND THEY DO NOT ALL POINT THE SAME WAY: SFocus, week-level spread across projects, -0.300 (0.004), 12 per cent of variance, 'Developers keeping fewer projects as their focus, rather than spreading themselves thin, are associated with more LOC added per week', WHICH SUPPORTS THE PREMISE AT THE WEEK LEVEL; SSwitch, repetitiveness of day-to-day switching, +0.225 (0.004), 10 per cent, 'a more repetitive day-to-day work pattern is associated with more outputs produced', WHICH SUPPORTS THE PREMISE AT THE DAY LEVEL; AvgProjectsPerDay, +0.046 (0.004), 1 per cent, 'More within-day multitasking is associated with more LOC added per week', WHICH RUNS AGAINST THE PREMISE AT THE HOUR LEVEL; and Projects, +0.263 (0.006), developers on more projects add more code. THEIR OWN CONCLUSION IS AN INVERTED U, VERBATIM: 'We find that up to a certain point, switching between projects is associated with productivity increases; but beyond that, it appears that overall output of developers declines. Yet, the survey responses indicate that developers seem to believe it's always better to work on ever-more projects, regardless of how many they already work on!' And on the day-level result: 'Psychology research suggests that repetitive work is associated with lower levels of arousal and decreased productivity, with switching tasks being one source of stimulation to counteract this effect. Our result suggests that switching between projects on consecutive days is more "expensive" than it is stimulating.' THREE LIMITS FROM THEIR OWN THREATS-TO-VALIDITY SECTION, AND THE THIRD IS THE ONE TO PRINT: 'All developers in our study are highly prolific contributors, and our survey respondents were self-selected. Thus, our results may not generalize beyond this research setting'; 'We considered contributions as commits only'; and 'Finally, measuring productivity as LOC added per week, or any other count of user interaction with the artifact, is certainly somewhat naive and incomplete. However, it is neither unprecedented nor unjustified by prior research.' TWO DEVELOPER VOICES FROM THEIR SURVEY, THE OTHER SIDE IN THEIR OWN WORDS: forty per cent agreed that working on multiple projects let them resolve more issues, including 'I can switch to another project when I get stuck and need some time to work out a good solution' [P96], and on quality, 'I find it easier to produce quality code if I break up long stretches working on a single codebase with time spent on a completely different problem' [P7]. THE INDUSTRY REPLICATION, read as ABSTRACT ONLY AT SOURCE: Kohl, K., Vasilescu, B., & Prikladnicki, R. (2020), 'Multitasking Across Industry Projects: A Replication Study', arXiv:2006.12636v1, 22 June 2020, first author Karina Kohl, GitLab data from one industry company with interviews replacing the original survey, verbatim: 'industry developers multitask as much as OSS developers focusing more (on fewer projects), and working more repetitively from one day to the next is associated with higher productivity, but there is no effect for higher multitasking.' Its dose-like observation that a short context change of less than three minutes did not affect results is a PERCEPTION rather than a measurement in its second clause.
[COMPLETE ARTICLE READ IN FULL AT SOURCE, the author copy at cs.ucdavis.edu, including the full regression table and the threats-to-validity section] Vasilescu, B., Blincoe, K., Xuan, Q., Casalnuovo, C., Damian, D., Devanbu, P., & Filkov, V. (2016). 'The Sky Is Not the Limit: Multitasking Across GitHub Projects.' In Proceedings of ICSE '16. doi:10.1145/2884781.2884875. First author Bogdan Vasilescu, then UC Davis, confirmed on the paper's byline. THE KNOWLEDGE-WORK OUTPUT EVIDENCE, CORRELATIONAL AND MORE INTERESTING THAN EXPECTED. DESIGN: commit data on 1,255 prolific developers across 58,092 public GitHub repositories, prolific defined as at least five years between earliest and latest commit and at least 500 commits across at least ten non-fork repositories; the model runs on 78,552 week-level observations from 1,193 developers after outlier removal, a linear mixed-effects model with random effects for developer; THE DEPENDENT VARIABLE IS log(lines of code added) per week; R-squared marginal 0.35, conditional 0.55. FOUR COEFFICIENTS MATTER AND THEY DO NOT ALL POINT THE SAME WAY: SFocus, week-level spread across projects, -0.300 (0.004), 12 per cent of variance, 'Developers keeping fewer projects as their focus, rather than spreading themselves thin, are associated with more LOC added per week', WHICH SUPPORTS THE PREMISE AT THE WEEK LEVEL; SSwitch, repetitiveness of day-to-day switching, +0.225 (0.004), 10 per cent, 'a more repetitive day-to-day work pattern is associated with more outputs produced', WHICH SUPPORTS THE PREMISE AT THE DAY LEVEL; AvgProjectsPerDay, +0.046 (0.004), 1 per cent, 'More within-day multitasking is associated with more LOC added per week', WHICH RUNS AGAINST THE PREMISE AT THE HOUR LEVEL; and Projects, +0.263 (0.006), developers on more projects add more code. THEIR OWN CONCLUSION IS AN INVERTED U, VERBATIM: 'We find that up to a certain point, switching between projects is associated with productivity increases; but beyond that, it appears that overall output of developers declines. Yet, the survey responses indicate that developers seem to believe it's always better to work on ever-more projects, regardless of how many they already work on!' And on the day-level result: 'Psychology research suggests that repetitive work is associated with lower levels of arousal and decreased productivity, with switching tasks being one source of stimulation to counteract this effect. Our result suggests that switching between projects on consecutive days is more "expensive" than it is stimulating.' THREE LIMITS FROM THEIR OWN THREATS-TO-VALIDITY SECTION, AND THE THIRD IS THE ONE TO PRINT: 'All developers in our study are highly prolific contributors, and our survey respondents were self-selected. Thus, our results may not generalize beyond this research setting'; 'We considered contributions as commits only'; and 'Finally, measuring productivity as LOC added per week, or any other count of user interaction with the artifact, is certainly somewhat naive and incomplete. However, it is neither unprecedented nor unjustified by prior research.' TWO DEVELOPER VOICES FROM THEIR SURVEY, THE OTHER SIDE IN THEIR OWN WORDS: forty per cent agreed that working on multiple projects let them resolve more issues, including 'I can switch to another project when I get stuck and need some time to work out a good solution' [P96], and on quality, 'I find it easier to produce quality code if I break up long stretches working on a single codebase with time spent on a completely different problem' [P7]. THE INDUSTRY REPLICATION, read as ABSTRACT ONLY AT SOURCE: Kohl, K., Vasilescu, B., & Prikladnicki, R. (2020), 'Multitasking Across Industry Projects: A Replication Study', arXiv:2006.12636v1, 22 June 2020, first author Karina Kohl, GitLab data from one industry company with interviews replacing the original survey, verbatim: 'industry developers multitask as much as OSS developers focusing more (on fewer projects), and working more repetitively from one day to the next is associated with higher productivity, but there is no effect for higher multitasking.' Its dose-like observation that a short context change of less than three minutes did not affect results is a PERCEPTION rather than a measurement in its second clause.
Written from the cOMPLETE ARTICLE READ IN FULL AT SOURCE, the author copy at cs.ucdavis.edu, including the full regression table and the threats-to-validity section.
What our sources record
Recorded when this source was checked, figures included, with page references and quotation marks removed.
Vasilescu, B., Blincoe, K., Xuan, Q., Casalnuovo, C., Damian, D., Devanbu, P., & Filkov, V. (2016). The Sky Is Not the Limit: Multitasking Across GitHub Projects. Proceedings of ICSE '16. doi:10.1145/2884781.2884875