Cost estimation models (2)
     
Date   27 January 1997
Leader   Rob DeLine
     
Papers   Chris F. Kemerer.
An empirical validation of software cost estimation models.
Communications of the ACM 30:5 (May 1987), pp. 416-429.

Charles R. Symons.
Function point analysis: Difficulties and improvements.
IEEE Transactions on Software Engineering SE-14:1 (Jan. 1988), pp. 2-11.

Donald J. Reifer.
Asset-R: A function point sizing tool for scientific and real-time systems.
Journal of Systems Software 11:3 (March 1990), pp. 159-172.
     
Supplemental Readings   Daniel V. Ferens.
COCOMO.
Encyclopedia of Software Engineering, Volume 1, John J. Marciniak (editor-in-chief), 1994, pp. 103-110.

Allan J. Albrecht.
Function points analysis.
Enc. of Soft. Eng., Vol. 1, pp. 518-524.

Lawrence H. Putnam.
Rayleigh model.
Enc. of Soft. Eng., Vol. 2, pp. 992-1000.

Chris F. Kemerer.
Software cost estimation models.
Software Engineer's Reference Book, Chapter 28, John McDermid (editor), CRC Press, Inc.
     
Meeting Material   To help clear up some confusion about the cost estimation models, I produced a set of slides, available in Acrobat format. The slides briefly cover the basics of the COCOMO, SLIM, and function points models, as well as the three readings from this week.
     
Other Material   Please see the meeting of 27 January for a set of relevent web sites.
     
Notable Remarks   David Garlan felt that hope for overcoming the inaccuracies that plague current estimation models lies in making them specific to the problem domain and/or the system's architectural style. In particular, the productivity/project factors present in both the COCOMO and function points models reminded him of ATT's architectural review question list.
     
Post Mortem   The Kemerer study has the flaw that the choice of study projects (mostly data-processing COBOL systems) seems biased in favor of models in the function points family, which historically arose in the data-processing community, and against the SLOC models. Nonetheless, the data analysis Kemerer performs is worth showing to students, and he raises several interesting points. His summary of the models under study, however, are not substantial enough to be understood without suppliment.
   The Symons paper was the weakest of the lot. His criticisms of the function points model are straightforward and evident, and his method of analysizing his improvements is dubious. The high point of the paper is his attempt to leverage off an apparently typical exercise in the specification of data-processing systems -- the separation of the data into entity-types and transactions -- to make the calculation of function points less arbitrary.
   Reifer's paper is interesting for its attempt to bring domain specificity to the function points model. For those that balk that the function points approach, as sketched in the canonical papers, is too arbitrary, this paper can help allay some of that criticism.

[DeLine 02/20/97]