You are on page 1of 2

Editor: Norman F. Schneidewind, Naval Postgraduate School, Code ASISs. Monterey.

CA 93943, fax (408) 656-3407, e-mail 0442p@vml .cc.nps.navy.mil

New Software-Quality Metrics Methodology


Standard fills measurement need
Normm F. Schnridewind, Naval Postgrudirate School and 1061 Working Group Chair
Last December, the I E E E Stan- needed is a theoretical underpinning identifying, implementing, analyzing,
dards Board approved the Standard for the validation and application of and validating software-quality met-
for a Software-Quality Metrics Meth- metrics. coupled with large-scale ex- rics. This methodology applies t o all
odology, 1061, the first IEEE-issued periments o n the statistical relation- software at all phases of any software
standard that deals with quality metrics. ship between quality factors such as life-cycle structure. As stated above,
As a process standard, 1061 does time-to-failure (an example of a prod- the 1061 standard does not prescribe
not mandate specific metrics for use. uct characteristic of interest to cus- specific metrics.
One reason for this approach is that tomers) and early indicators of quality
industry needs a methodology for im- such as complexity (a product charac- Standard’s audience. This standard
plementing a metrics plan that is inde- teristic of interest t o developers). is intended for those associated with
pendent of particular metrics that Quality factors and indicators are the acquisition, development, use,
might be used. both metrics. However. one point that support. maintenance, or audit of soft-
Selling this approach to the metrics 1061 makes is that often the metric ware, and is aimed particularly at
community has not been easy because those measuring or assessing the qual-
a significant part of that community ity of software.
feels compelled to “measure some- The standard can be used by
thing” and to “put measurement into
practice.” The 1061 Working Group To achieve high software acquisition or project managers, t o
subscribes to the objective of putting identify and prioritize the quality re-
“measurement into practice” as long quality in a system, the quirements for a system;
as there is a p l a n t o support it: 1061 software’s attributes must system developers, to identify spe-
provides the plan. cific traits that should be built into the
be clearly defined; software to meet the quality require-
Philosophy behind 1061. The phi- otherwise, assessment of ments:
losophy of 1061 is that an organization quality is left to intuition. quality assurance organizations
can use whichever metrics it deems and system developers, t o evaluate
most appropriate for its applications whether quality requirements are be-
as long as the methodology is fol- ing met:
lowed and the metrics are validated system maintainers, t o assist in
(see below). (that is. quality factor) of greatest in- change management during product
Another reason for this approach is terest to the customer cannot be mea- evolution: and
that the 1061 Working Group did not sured in an early phase of the soft- users, to assist in specifying the re-
reach a consensus about which metrics ware-development process: instead. a quirements of the requested system.
should be mandated for use (the pro- metric such as complexity can be used
visions of a standard are mandatory. as an indicator of quality during de- Why software-quality metrics? Soft-
not optional). Consistent with this ap- velopment. ware quality is the degree to which
proach was the working group’s char- Theories of measurement and vali- software possesses a desired combina-
ter, as provided in the IEEE Stan- dation have emerged to provide a ra- tion of attributes. T o achieve high
dards Board approval of the project tionalization for metric selection and software quality in a system, this com-
authorization request, which called application.'^' This development. cou- bination of attributes must be clearly
for development of a standard meth- pled with the identification of com- defined: otherwise, assessment of
odology. posite metrics that show a significant quality is left to intuition.
However, despite its reservations relationship with quality factors,’ led For this standard, defining software
about endorsing specific metrics. the the working group to believe that a quality for a system is equivalent t o
working group provided definitions set of quality metrics, anchored in defining a list of software-quality at-
and descriptions of several popular both theory and practice. could be tributes required for that system. An
metrics in appendixes (see below). published as a standard in the next appropriate set of metrics must be
which are included in the 1061 docu- several years. identified to measure the software-
ment but are not part of the standard. quality attributes.
Many empirical studies conducted Scope of the standard. This stan- The purpose of software-quality
t o validate various metrics have not dard provides a methodology for es- metrics is to make assessments
yielded consistent results. What is tablishing quality requirements and throughout the software life cycle as

April 1993 105


to whether the software-quality re- When it is possible to measure fac- specified percentage of the applica-
quirements are being met. Using met- tor values at the desired point in the tions of the metric.
rics reduces subjectivity in software- life cycle, these quality factors are
quality assessment by providing a used to evaluate software quality. At Appendixes. Several appendixes in
quantitative basis for making deci- some points in the life cycle, certain the standard provide definitions and
sions. However, such usage does not quality factors (for example, reliabili- examples of commonly used quality
eliminate the need for human jlgg- ty) are not available; they are ob- factors and metrics. Unique features
ment in software evaluations. tained after delivery or late in the of this standard are appendixes that
Organizations and projec’s that use project. In these cases, other metrics contain numerical results from using
these metrics can expect tc gijoy the
1 are used early in a project to predict popular metrics and comprehensive
beneficial effect of more vi le soft-
I quality factors. case studies - one in the mission-crit-
ware quality. The history of applying metrics in- ical area (aerospace and defense soft-
dicates that metrics were seldom vali- ware development) and another in the
Methodology of metrics. The soft- dated (that is, it was not demonstrated commercial arena (off-the-shelf com-
ware-quality metrics methodology is a through statistical analysis that the mercial software) - that illustrate the
systematic approach to establishing metrics actually measured software application of the methodology in a
quality requirements and identifying, characteristics as they purported to). step:by-step fashion.
implementing, analyzing, and validat- However, validating metrics before Appendix A contains examples of
ing process and product software- they are used to evaluate software factors, subfactors, and metrics, and
quality metrics for a software system. quality is important. Otherwise, met- the relationships among them; Appen-
It spans the entire software life cycle rics might be misapplied (that is, met- dix B, sample metrics descriptions and
and comprises five steps. These steps rics 4ght be used that have little or computations; Appendix C, examples
must be applied iteratively because in- no :tionship to the desired quality of methodology usage (for instance,
sights gained from applying a step can cha, I xistics). mission-critical o r commercial envi-
show the need for further evaluation C*!i8rIityfactors may be affected by ronment); and Appendix D, annotat-
of the results of prior steps. The steps mu!’ ,’I * variables. A par’icular met- ed bibliography, references, and stan-
are: ric, I i’ :efore, might not : fficiently
1 dards.
reprc i nt any single factc.1 because it
(1) Establish software-quality re- ignores other variables. To go further. Copies of Standard
quirements. A list of quality factors is 1061 can be obtained by contacting
selected, prioritized, and quantified at Validity criteria. T o be considered the I E E E Standards Office, 445 Hoes
the outset of system development or valid, a metric must demonstrate a Lane, PO Box 1331, Piscataway, NJ
system change. These requirements high degree of association with the 08855-1331, phone (908) 562-3800.
are to be used to guide and control quality factors it represents. This is Please d o not request copies from the
development of the system and, on equivalent to accurately portraying author.
delivery of the system, to assess the quality condition(s) of a product
whether the system meets the quality or process. A metric may be valid with Acknowledgments
requirements specified in the contract. respect to certain validity criteria and
(2) Identify software-quality metrics. invalid with respect to other criteria. This standard was developed by a broad-
The software-quality metrics frame- The following criteria are used: based committee of representatives from
work is applied in selecting relevant industry, government, and academe. I thank
metrics. Correlation. The variation in the the members of the working and balloting
(3) Implement the software-quality quality factor values for a product or groups and the many individuals who con-
metrics. Tools are procured or devel- process explained by the variation in tributed comments toward development of
the standard.
oped, data is collected, and metrics the corresponding metric values must
are applied at each phase of the soft- exceed a specified threshold.
ware life cycle. Tracking. A change (increase or de- References
(4) Analyze the software-quality crease) in a quality factor value for a
metrics results. These results are ana- product or process must be accompa- 1. H. Zuse, Software Complexity: Measures
lyzed and reported to help control de- nied by a corresponding change (in- and Methods, Walter de Gruyter, 1991.
velopment and assess the final product. crease or decrease, as appropriate) in
(5) Validate the software-quality a metric value. 2. N.E. Fenton, Software Metrics: A Rigor-
metrics. Predictive metrics results are Consistency. If quality factor values ous Approach, Chapman and Hall, 1991.
compared with quality factor results are rank-ordered for products o r pro-
3. N.F. Schneidewind, “Methodology for
to determine whether the predictive cesses, the corresponding metric val- Validating Software Metrics,” IEEE
metrics accurately measure their asso- ues must have the same ordering. Trans. Software Eng., Vol. 18,No. 5 , May
ciated factors. Predictability. If a metric is used to 1992,pp. 410-422.
predict a quality factor for a product
Why metrics validation? The pur- or process, it must predict within a 4. N.F. Schneidewind, “Minimizing Risks
pose of metrics validation is to identi- specified accuracy. in ApplyingMetricson MultipleProjects,”
fy product and process metrics that Discriminative power. A metric Proc. Third Int’l Symp. Software Reli-
can predict specified quality factors, must be able to discriminate between ability Eng., IEEE CS Press, Los Alami-
the quantitative representations of high- and low-quality products o r pro- tos, Calif., Order No. 2975,1992,pp. 173-
182.
quality requirements. If metrics are to cesses.
be useful, they must indicate accurate- Reliability. A metric must demon- 5. J.C. Munson andT.M. Khoshgoftarr,“The
ly whether quality requirements have strate the above correlation, tracking, Detection of Fault-Prone Programs,”
been achieved o r are likely to be consistency, predictability, and dis- IEEE Trans. Software Eng., Vol. 18, No.
achieved in the future. criminative power properties for a 5, May 1992,pp. 423-433.

106 COMPUTER

7-

You might also like