R. Rask et al., SIMULATION AND COMPARISON OF ALBRECHT FUNCTION POINT AND DEMARCO FUNCTION BANG METRICS IN A CASE ENVIRONMENT, IEEE transactions on software engineering, 19(7), 1993, pp. 661-671
Software size estimates provide a basis for software cost estimation d
uring software development. Hence, it is important to measure the syst
em size reliably as early as possible, i.e., during the requirements s
pecification. Two best known specification level metrics are Albrecht'
s Function Points and DeMarco's Function Bang. One problem in using th
ese metrics has been that there are only few tools that can calculate
them during the specification phase. We have built one such tool. Anot
her problem has been that no research data is available how these metr
ics correlate with one another. The paper compares these two metrics b
y a simulation study in which automatically generated randomized dataf
low diagrams (DFD's) were used as a statistical sample to count automa
tically function points and function bang in a built CASE environment.
These value counts were correlated statistically using correlation co
efficients and regression analysis. The simulation study permits suffi
cient variation in the base material to cover most types of system spe
cifications. Moreover, it allows sufficient sampling sizes to make sta
tistical analysis of data. The obtained results show that in certain c
ases there is a relatively good statistical correlation between these
metrics. No overall general correlation exists, however. The paper doe
s not show which one of the two metrics fares better as a size metric.
Yet, our study suggests to use in many cases Function Bang metric, be
cause its automatic calculation is simpler and depends less on judgeme
nt. Moreover, the study demonstrates that correlations depend upon a s
ystem type. This implies that in software projects one must be careful
with size estimates while using these metrics. In order to to know wh
en one needs to calibrate the size estimate we need to develop algorit
hms which help to detect logical system types and make adjustments acc
ordingly. The results also point out the need of empirical research in
which we can better derive the connection between specification level
metrics and the number of lines of code.