I've been saving this because I just did a lit review about cognitive apprenticeship in pair programming, but much of the research talked about pair programming generally.
https://twitter.com/JuttaEckstein/status/1454002930818949122
Also, if you've recently started following me, I just started my PhD studies because I want to improve scaffolds around pair programming. Follow for more pair programming and other random learning sciences/cognitive science content... To the research!
There is research that supports and refutes the usefulness of pair programming. More in the support than refute. The challenge in pair programming research is how do you measure success (research design & methodological rigor) and who is doing the pairing?
A lot of the research is done w/ college students or middle & high school students on class assignments. I've had trouble finding research done with professional developers let alone professionals using a mature codebase. The more challenging thing is how do you measure success?
Sun et al. (2019) in Search & Research on Economics of Pair Programming solve algebraically using defect removal time, product size, defect density, work time, number of developers, number of pairs, and developer salary.
Smith et al. (2018) in Long Term Effects of Pair Programming examine partnership participation, project and exam scores, withdraw rates (the participants were college students), time between courses, GPA, and gender.
Plonka et al. (2014) look at knowledge transfer in Knowledge transfer in pair programming: an in-depth analysis. They point out that Pandey et al. (2003) suggest that knowledge spread among team members can reduce project risk because of less reliance on individuals in the team.
Baheti (2002) in Assessing distributed pair programming measures productivity, software quality, and student feedback. Depending on the study outcomes can vary because there's no consensus on how to measure if pair programming is "successful".
I was discussing with my advisor how I plan to measure "success" when studying pair programming. One of the topics that came up was code or program quality. Some studies will create a test suite & measure time to complete based on getting all passing tests.
Pairing is a highly complex activity that does offer benefits, however quantifying those benefits to determine if pairing is worth it is very challenging. Because it depends. It depends on the environment, the team, the individuals, the organization, power dynamics, and so on...
Even in the original tweet, "pair must deliver more work than an individual" Which individuals? Which pairs? How are we quantifying work? Because I can definitely sink a bunch of hours into writing a pretty simple program 🤣😂 Did I deliver "more work"?
Also, my opinion, if the only dimension you're looking at is work output that's a pretty limited view. We do know that people working in pairs find work more enjoyable (generally) (Xinogalos et al., 2017; Celepkolu & Boyer, 2018). Happy people tend to stay where they're happy.
Anyway, I love this topic and could talk about it for a while. Can't wait to share what I find next. :)
• • •
Missing some Tweet in this thread? You can try to
force a refresh
