I want to conduct some multivariate analyses using the EMI-2 but with so many subscales my sample size is not big enough. How can I reduce the data?
Yes, this is a problem that is a quite serious practical limitation of the instrument that I have always recognised. It’s not appropriate to simply add up subscale scores to reduce the data (see above). One thing I once toyed with to reduce the data to manageable proportions was to test a smaller set of higher order models on the basis that if these models fit the data well enough, one could collapse scores on the first-order factors. If you are familiar with the paper describing the development of the EMI-2 (Markland and Ingledew, 1997), you will recall that we grouped conceptually-related subscales and tested five submodels, because of the sample size problem in testing the model as a whole. More recently, I tested a set of higher order models based on these groupings. Although the fit of these higher order models was generally good, I was not too happy with this approach for conceptual reasons. For example, the Interpersonal Motives grouping comprised social recognition, affiliation
