Management Information Systems Quarterly
Are Two Heads Better than One for Software Development? The Productivity Paradox of Pair Programming
Abstract
Extreme programming is currently gaining popularity as an alternate software development methodology. Pair programming, a core practice of this methodology, involves two programmers working collaboratively to develop software. This study examined the efficacy of pair programming by comparing the performance effectiveness and affective responses of collaborating pairs with those of individual programmers treated as nominal pairs. In a controlled laboratory experiment involving student subjects, proxies for entry level programmers working on entry level tasks, two factors were manipulated: programming setting (collaborative pair versus individuals) and programming task complexity (high versus low). Participants who worked in the individual condition were randomly combined into nominal pairs. The performance and affective responses of the collaborating pairs were then compared with those of the best performers and the second best performers of each nominal pair. Results indicated that programming pairs performed at the level above the second best performers and at the level of the best performers in each nominal pair. This relationship was found to be consistent across both levels of task complexity. Consequently, there was no evidence of an “assembly bonus effect,” where the performance of a collaborating pair exceeds the performance of its best member working alone. While this finding may appear counterintuitive due to the general perception of two heads being better than one, it is consistent with the findings in small group research. When affective responses were considered, programming pairs reported higher levels of satisfaction than those of the best and second-best performing members in nominal pairs. They also showed higher levels of confidence in their performance compared to those of the second-best members. But the confidence levels of pairs were no different from those of the best performing members in nominal pairs. Theoretical and practical implications of these findings are presented.