As I understand it, Grimmer and King have produced a computer assisted clustering scheme that lets a researcher quickly run data through (approximately) the entire universe of clustering models and pick the one that he/she thinks suits the data best. Such a just so justification makes me uncomfortable, mainly because I don't want to be accused of picking the classification scheme that's convenient for my interpretation. That seems like cheating to me, in some way. In the paper, Grimmer and King give two reasons as to why this is so, but I'm afraid I don't quite get it. I still hold on to the idea that a check against some "gold standard" human supervised classification scheme is necessary.
So I guess this is a clarification question in some way: what's the best way to pick a clustering method, and why do Grimmer and King think there's no objective method to appeal to?
As I understand it, Grimmer and King have produced a computer assisted clustering scheme that lets a researcher quickly run data through (approximately) the entire universe of clustering models and pick the one that he/she thinks suits the data best. Such a just so justification makes me uncomfortable, mainly because I don't want to be accused of picking the classification scheme that's convenient for my interpretation. That seems like cheating to me, in some way. In the paper, Grimmer and King give two reasons as to why this is so, but I'm afraid I don't quite get it. I still hold on to the idea that a check against some "gold standard" human supervised classification scheme is necessary.
So I guess this is a clarification question in some way: what's the best way to pick a clustering method, and why do Grimmer and King think there's no objective method to appeal to?