Well, as I said in #14, when you are creating latent variables in structural equation models, these variables have no inherent metric or scale and this fact leaves the model unidentified. The conventional way to identify the model is to fix the loading of the latent variable on one of its indicators at 1. But another way to identify the model is to fix the variance of the latent variable at 1, and in some contexts, or for some audiences, this approach may be the more understandable way, and it eliminates the question of which indicator variable to select to receive the unit loading, a selection that is usually completely arbitrary. Identifying the model by fixing the latent variable's variance to 1 has the advantage of treating all the indicators equally.
When all is said and done, the purpose of data analysis is to increase our understanding of the phenomena the data represent. So we should structure the analyses and present the results in the most informative, comprehensible way we can. Most of the time, standardization is counter-productive because it converts variables measured in units that people understand to variables measured in units that are obscure, even secret, and irreproducible. They make the results less understandable. That's why I think they should only be used when the variables are, themselves, already measured in obscure or arbitrary units--at least then they don't make anything worse.
When all is said and done, the purpose of data analysis is to increase our understanding of the phenomena the data represent. So we should structure the analyses and present the results in the most informative, comprehensible way we can. Most of the time, standardization is counter-productive because it converts variables measured in units that people understand to variables measured in units that are obscure, even secret, and irreproducible. They make the results less understandable. That's why I think they should only be used when the variables are, themselves, already measured in obscure or arbitrary units--at least then they don't make anything worse.
Comment