AlexandersStripsExperiment

Thanks to all who participated.

Four people did the exercise. Since the strips fall into equivalence sets we looked for agreements with Alexander's rankings +/- 4. And since there is a very large equivalence set in the lower rang of rankings, we considered only the top 15 ranked strips. The agreement with Alexander's results varied from 5 to 9 (out of 15)

The conclusions I've drawn are that: 1 using all 35 strips takes too long and is too confusing Therefore: use a subset of the strips, at least one from each of the equivalence sets that Alexander identifies, upto 12 or 15 strips

2 people need to be very clear about what they are looking for. One of the "subjects" last night, having been to told to order the strips by "coherence" gave high rankings to tjose strips that were "nice and random" Therefore: discuss the idea of coherence a little before the experiment (but avoid talking about comprehension, which is the punchline)

3 there seemed to be a stronger correlation between the rankings produced by the members of the group than between any one member and the results provided by Alexander. (without Alexander's original papers it's necessary to read between the lines: my guess is that his subjects wer psychology students on their lunch hour, not software engineers down the pub, this may be significant) Therefore:when looking at the results emphasise the similarity between the local results, while refering to Alexander's results in passing

A few people asked me what this was all in aid of. The intention is to run a "brown bag" session at AMS on the subject of "quality". It'll be fairly wide-ranging, but the core of it will be refactoring. It's my thesis that ISO9000 style quality as usually implemented is woefully innapropriate for software development for a bunch of reasons, but a large one is that it neglects a large part of the customer population for a software product.

ISO9000 defines "quality" as "meeting the customers' requirements". In software development this is usually taken to mean the end-users' functional requirements. I'd suggest that the developers who will work on your part of the product after you (with XP, that might happen tommorrow) are also customers, and have a set of non-functional requirements that must be met. Foremost amongst these is comprehensibility.

Alexander's experiment is about comprehensibility. I intend to use it to show that comprehensibility is in some sense objective. The question then is how to increase the comprehensibility of a body of code: the answer is refactoring, which can be a difficult practice to justify.

--KeithBraithwaite³host³³date³August 2, 2000³agent³Mozilla/4.73 en? (WinNT?; I)³AlexandersStripsExperiment