Data Envelopment Analysis Journal > Vol 2 > Issue 1

On the Use of DEA for Software Development Productivity Measurement

Mette Asmild, IFRO, University of Copenhagen, Denmark, Francisco Imperatore, Pan American Energy, Argentina,
Suggested Citation
Mette Asmild and Francisco Imperatore (2016), "On the Use of DEA for Software Development Productivity Measurement", Data Envelopment Analysis Journal: Vol. 2: No. 1, pp 81-111.

Published: 26 Oct 2016
© 2016 M. Asmild and F. Imperatore
Productivity measurement and analysis,  Semiparametric and nonparametric estimation,  Product Development
Data Envelopment Analysis (DEA)productivityefficiencysoftware development

Article Help


Download article
In this article:
1. Introduction
2. Literature Review
3. General Model Specification Considerations
4. Empirical Study
5. Discussion and Lessons Learned
Appendix: List of variables from some past papers


This paper first, based on a thorough review of the relevant literature, provides a general discussion of some of the many aspects one should consider when using Data Envelopment Analysis (DEA) to measure the efficiency of software development. It then contains a case study which investigates a few of the important aspects. Finally it provides recommendations and directions for future research in the area.

Rather than simply presenting a "final" set of results, the empirical study takes a step-by-step approach to considering, and thus illustrating, some of the issues to be aware of when doing empirical analysis, many of which are not limited to efficiency assessment of software development. Thus this paper, besides providing a thorough literature review and resulting guidelines specifically for the analysis of software development productivity, also aims at providing a practical "how to" guide for the empirical analysis that can be used more generally.

In terms of actual empirical results, we find a clear indication of variable returns to scale, of differences between new and evolutions developments and some indications of differences between programming languages. We also include potentially relevant variables such as requirements volatility, reliability (number of defects) and discuss many other relevant variables, for which data were not available here.