between 1950 and 1970 occurred because of the replacement of assembler coding by high-level languages.

Nevertheless, as the panel attempts to point out in this report, increased collaboration between software engineers and statisticians holds much promise for resolving problems in software development. Some of the catalysts that are essential for this interaction to be productive, as well as some of the related research opportunities for software engineers and statisticians, are discussed below.


The panel strongly believes that the right model for statistical research in software development is collaborative in nature. It is essential to avoid solving the "wrong" problems. It is equally important that the problems identified in this report not be "solved" by statisticians in isolation. Statisticians need to attain a degree of credibility in software engineering, and such credibility will not be achieved by developing N new reliability models with high-power asymptotics. The ideal collaboration partners statisticians and software engineers in work aimed at improving a real software process or product.

This conclusion assumes not only that statisticians and software engineers have a mutual desire to work together to solve software engineering problems, but also that funding and reward mechanisms are in place to stimulate the technical collaboration. Up to now, such incentives have not been the norm in academic institutions, given that, for example, coauthored papers have been generally discounted by promotion evaluation committees. Moreover, at funding agencies, proposals for collaborative work have tended to "fall through the cracks" because of a lack of interdisciplinary expertise to evaluate their merits. The panel expects such barriers to be reduced in the coming years, but in the interim, industry can play a leadership role in nurturing collaborations between software engineers and statisticians and can reduce its own set of barriers (for instance, those related to proprietary and intellectual property interests).


As discussed above in this report, for statistical approaches to be useful, it is essential that high-quality data be available. Quality includes measuring the right things at the right time—specifically, adopted software metrics must be relevant for each of the important stages of the development life cycle, and the protocol of metrics for collecting data must be well defined and well executed. Without careful preparation that takes account of all of these data issues, it is unlikely that statistical methods will have any impact on a given software project under study. For this reason, it is crucial to have the software industry take a lead position in research on statistical software engineering.

Figure 8, a model for the interaction between researchers and the software development process, displays a high-level spiral view of the software development process offered by Dalal

The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement