ghost
commented
6 years ago This is to say, what are the implications of deploying novel methods in theory-generating paradigms for which they were not developed? How do we think about the implicit assumptions embedded in those paradigms (ie., re: the social world, the analyst, etc), and transform those assumptions for use with new methods, if they need to be transformed at all? (I would argue that this might be necessary to do)
I'm finding it difficult to rectify the use of unsupervised machine learning techniques in massive data sets with Glaser & Strauss' Grounded Theory (1967). In the first, I was somewhat confused with the conflation between GT and Tavory and Timmermans' Abductive Analysis (2012, 2014) in the introduction (23, n. 2)- Tavory and Timmermans explicitly developed AA so as to address the relative theoretical poverty of Grounded Theory by "challenging presumptions or pre-existing theory, and lead the socialanalyst to abductively generate new theory by imagining what would be socially required for those patterns to exist" - while the footnote is correct, this doesn't seem to develop "grounded theory", which seeks to bring no a priori theoretical assumptions to the analysand as the text would suggest. Confusingly, this is rectified later in the paper, with a meaningful difference being made between the two approaches (25).
More substantively, I find myself concerned with the application of computational technique to theoretical methods developed with ethnographic approaches in mind. The authors are absolutely correct in pointing out that "Machine memory augments human analytical limits by holding a massive array of language features simultaneously in mind so we can associate them in reliable constant comparison" (25); these algorithms become better grounded theorists than we could ever be. This notwithstanding, I feel that Grounded Theory was developed with these human analytical limits in mind- the absorption and deployment of spatial and social context, the limits of what the ethnographic lens finds interesting and why, etc. I fear that "perfect" Grounded Theorists, especially with no a priori assumptions guiding the inquiry, could amount to data mining for spurious connections, if not properly constrained. Abductive Analysis might be better suited to ML techniques, as it is guided by existing theory and seeks counterfactual data in refining these theories.