Reproducibility of Results

A much too unfrequently asked question! A valid scientific experiment has to be reproducible. In Informatics, this norm is frequently violated, and even more often, nobody seems to care about reproducing some particular experiment. But that does not invalidate the norm.

For a good or even excellent thesis, this means: If your thesis goals include empirical and/or constructive elements, then your results need to be reproducible for others – these may be your advisors, and, even better, anybody who reads your thesis and wants to check your results or wants to examine your examples by himself or herself. So, if you write a program or a set of modules as a part of your thesis work, please do make sure that they run at least on our lab computers and that there is somebody of the staff who knows where the software is and how it can be used. Even better, write a little(!) simple(!) interface (which may or may not be graphical) that allows at least your experiments or test runs to be repeated, or, optimally, that gives everybody a chance to do their own experiments or test runs with their own test data. Frequently, bachelor or master candidates, when facing texts from the literature, complain that the reported results cannot be checked or easily varied; and then, the same candidates would often deliver their own results in exactly the same manner that they found so irritating about others’ work. That is inconsistent, isn’t it? And, what is worse: It is no sound science!