I'm no scientist. However, I do have a fair amount of common sense when it comes to experiments and evaluating your findings. One of the things I find quite frustrating about some of the articles on authorship I've been reading is the lack of information pertaining to people's basic assumptions before beginning their study. The key example I'm thinking of here is Kenneth Muir in his Shakespeare as Collaborator (1960).
Muir's general thesis in the section I'm concerning myself with is to show elements of collaboration in Edward III. To do so, he splits the play into two sections, his Part 'A' (the scenes thought to be Shakespeare's) and Part 'B' (everything else). He then goes on to show, rather impressively, that there are some big differences between the two sections, such as the frequency of 'new' words introduced.
What's the problem? Well, it's the lack of reasoning as to how he created the division of the two parts in the first place. For all we know, his divisions are entirely arbritrary. Worse, those divisions could have been created by the exact criteria which he then goes on to test, creating a circularity of argument which looks impressive but is anything but. Similarly, without knowing the grounds on which Parts 'A' and 'B' are distinguished, the tests lose their validity.
This is not to say that Muir is, necessarily, wrong. The ongoing cumulative weight of evidence supports a split close to the one he made. However, if you are going to prove something as contested as the authorship of an anonymous play, we as readers need to know where the experiment is starting from. The problem is that almost all articles since Muir take these divisions as read. If an experiment is going to be conducted in a laboratory environment, for lack of a better phrase, then this kind of received assumption needs to be interrogated.
In this sense, I'm far more interested in the work of MacDonald P. Jackson with LION. In articles I've read so far by Jackson, he begins with a thesis and then tests it in the widest possible sense; for example, in a 2006 Shakespeare Quarterly article testing for the author of Scene 8 of Arden of Faversham, he tests the scene against the entire corpus of English drama for a twenty year period around the estimated date of the play's authorship. This particular article tested the scene against 132 Elizabethan plays. Granted, it is only relatively recent technology that makes this kind of search practical, but it allows Jackson to provide cold, hard statistical data that can be clinically interrogated.
It strikes me that, in the study of authorship, one needs to commit to being either artistic or scientific. If you're going to be artistic, go with it; use aesthetic judgements, rely on instinct and the feel of a piece. If you're going to be scientific, follow the basic principles; define your search limits, explain your control defaults, provide objective data. Too often, however, sloppy science is used to lend credence to what is essentially a subjective judgement. The essential point I want to make here is that valid aesthetic judgements are undermined by badly-applied science. Sloppy science is obvious, and immediately invites interrogation and scepticism. It is all-too-easy to fight someone on scientific grounds, and in the ensuing debate the gut instincts lose their value.
This is all, of course, easier said than done. There are few articles on authorship which don't contain a certain measure of subjective interpretation. The difference is how you articulate that interpretation; whether you try to pass it off as being as scientific as your data, or you admit your own subjectivity. Passion and science can work together, but passion must not alter the science, otherwise one's experiment is compromised. The best scientists, it seems to me, are dispassionate about their method, but passionate about their results.