Abstracts
Page maintained by Norman Fenton. Last updated: 13 July 2000
Abstracts should appear on the front page of a report. There are two
types of abstracts: descriptive and informative. A descriptive abstract
says what is in the report without providing any of the results. An informative abstract
says what the report contains, including summarising the main results. You should always
write informative abstracts. Compare the following two abstracts describing the same case
study:
Version A (descriptive)
This report describes a major case study to evaluate the
effectiveness of using a certain formal method during software development. <general
blurb about formal methods removed> We describe the background of the particular method
used and discuss the claims made in favour of these methods. We describe the experimental
set-up and the particular software under investigation. We present a range of results
indicating the circumstances under which formal methods may be effective. We explain the
measurements that were used, along with the rationale for using them. We compare the
results of the measurements at different life-cycle phases. We consider the different uses
of the system. Finally, we present a number of strong recommendations.
Version B (informative)
VDM is one of the best known formal methods used in
software development. We describe a case study to evaluate whether higher quality code
results from the use of VDM. The case study involved an air traffic control system
developed over three years. Some of the modules in the system were developed using VDM
(160 modules making approximately 400 KLOC) while the rest of the modules (300 making
approximately 700 KLOC) were developed informally. We found that, prior to release, the
fault density of formally developed modules was not significantly different to the
informally developed modules (4 faults per KLOC being typical). However, the fault density
in the 6 months post-release was significantly lower for formally developed modules (on
average 0.6 faults per KLOC compared to 1.4 faults per KLOC). We also found that more
faults were found during the early development phases in the formally developed modules.
This favourable evidence to support formal methods is countered by the following
observations: i) the formally developed modules generally took 25% longer to complete than
similar sized informal modules. ii) the formally developed modules were those concerned
with the critical functions and were developed by more experienced and better qualified
staff with a strong mathematical background; iii). the non-formally developed modules
included all of the interface code so faults discovered in the first 6 months post-release
were inevitably more likely to be in this part of the system. Despite these reservations
we believe that the post release fault-density for the formally developed modules was very
low. We therefore recommend that companies should consider using formal methods such as
VDM for the most critical components, providing that they have well trained staff with a
very good mathematical background.
Obviously Version A is shorter, but only because the general blurb
about formal methods has been removed. Version A actually tells the reader nothing about
the case study. This writer is challenging the reader to read through the entire report in
order to find out the basic results. Version B, on the other hand tells us all the key
information about the case study without including anything superfluous. Version B does
not tell us that the paper contains a discussion about formal methods - it does not need
to since we know that a paper investigating formal methods must contain some such
discussion. Version B is better in every respect. Even if we do not have time to read the
paper (and most readers never get further than the abstract) it tells us what we really
need to know. It even makes us more likely to read the paper because it will identify and
target key readers.
Since informative abstracts are so obviously superior to descriptive
ones why do the majority of scientific writers still insist on providing descriptive
abstracts that infuriate us and insult our intelligence? The answer is simple: laziness
(although in some cases it may be due to the fact that the author really has nothing to
say!). Descriptive abstracts are often written before the work has even been
carried. In other words the abstract is merely a plan for the author. Now plans are fine
and necessary in order to complete a piece of work. But if you were delivering any product
you would not use your original project plan as a replacement for the product description.
So never use a descriptive abstract.
Return to top of page
Return to Good Writing Guide index
Return To Norman Fenton's home page
|