About this paper
The ABC of Software Engineering Research by Klaas-Jan Stol and Brian Fitzgerald, published October 2018. See link for full citation.
Notes
There are too many ways in which terms describing research methods in software engineering get used, and these authors have a solution. The reason, at least according to the introductory discussion in this paper, is in part a case of discipline envy. This is the idea that we don’t quite know what software engineering is, but we know what those people over there do, and we like that, so we’ll co-opt it.
You could argue that the entire idea of software engineering is discipline envy. A collection of computing experts from academia and industry didn’t quite know how to formalise the problems faced by software makers in the 1960s, but they did know what engineers do, and they liked that. In fact, it’s not clear that they (or at least we, their intellectual descendents) truly understood what engineers do, but nonetheless we gave it a jolly good go. In 1968, in the town of Garmisch in Germany, a discipline was born.
Now, it’s interesting that discipline envy has turned up so early in this discussion, because the contribution in this paper is a framework borrowed from social science researchers. To understand the applicability of cross-discipline seeding, we have to ask how strong the analogy “software engineering is just like X” seems to be (as well as identify how well the proposed idea worked in field X). So, is software engineering like a social science?
Here, the authors carve the field in two. They distinguish “solution-seeking” research, in which we identify what we ought to do about a problem, from “knowledge-seeking” research, in which we identify what people do do about the problem. The bad news about ditching the solution-seeking half of the discipline is that we just lost the engineering from software engineering, the bit where we use scientific results to propose novel solutions to problems.
The good news is that knowledge-seeking software engineering research does look quite a bit like a social science. People, in some context, do things, and we can try to understand that. Indeed that is the origin of the ABC initialism: Actors (the people), Behaviour (the things) and Context.
Well, we can understand bits of it at a time. Like good consultants, the authors introduce a quadrant diagram. On one axis, the “generalisability” of a research method, from highly universal contexts to deeply specific contexts. On the other, the “obtrusiveness” of the method, reflecting whether the researchers are passive observers or active interferers.
As the Labrary stands at the intersection, we approve of the idea that two different things lie on a continuum, rather than being an either/or choice. This framework makes the point that while a particular research technique or strategy is situated somewhere in the general/obtrusive map, others are available elsewhere. The reaction to a highly-controlled lab experiment should not be to declare that the result is not generalisable, but to understand what else could be done to explore generalisations of its results.
The discussions of where particular research strategies fit in the map are interesting, though some of the analogies drawn are fairly tenuous. The authors show where on the map the maximum applicability of a method for each of the key properties lies: universally contextual, unobtrusive research generalises over Actors best, while highly-obtrusive methods allow more precise measures of particular Behaviours and more focus gives a more realistic Context. It would be really beneficial to see a similar framework for “solution-seeking” literature, so we can evaluate the applicability of techniques developed in software engineering research to “practical” problems.
