Gregory's Server@boud we chose our scope because we do not have unlimited resources
|
9 comments
 Assuming that "medium" is logarithmic, i.e. about 1000 or so users, there are a few hundreds of these - so you presumably intend to subsample these medium-sized instances, e.g. aim at 10-20 interviews. Splitting up into e.g. 5-10 medium and 5-10 small instances would increase the sample error and bias, as @jcolomb said, but also increase the chance of better studying all scales: big + medium + small. The risk of the big crushing the small seems like a significant governance question. @boud @darius @jcolomb I would also argue that a ‘medium and large’ runs a significant risk of skewing heavily white, and mostly male. Plus the fact that it’s folks from the US, with US money, on a network that for once isn’t as dominated by American interests and governance as almost everything that came before.  @mejofi @boud @jcolomb It just occurred to me that we might not be talking on similar terms here: for me the break point for "medium" is about 150 people, where it is unlikely for everyone to know everyone else. Because to me that is a place where governance mechanisms become necessary rather than just a good idea. I talk about this in https://runyourown.social Re: the American thing, it's a valid criticism and I am trying to be mindful of my biases. We want an international sample for sure.  @jcolomb @darius @boud We’re focusing on governance models, rather than eg software, so we're also choosing primary subjects who work at or beyond a minimum level of complexity. (An instance for 15 friends *should* have minor risks and governance challenges, though there are outliers—and huge instances are their own kind of beast.) By the time we begin work in 2024, we'll have "medium" more narrowly defined, but I expect wide variety in size. |
@darius @boud
no one has unlimited resources of course, but this is a sampling question. One has to make choices, and the question was about why you made the choice to reduce your analysis on medium and large size instances.
Indeed, If there are 10 000 small instances and 1 very huge one, why are you choosing to interview the huge one ?
It is impossible to prevent sample bias when planning interviews, but this one seems unnecessary ?