I have obtained responses on around 48 items measuring employer attractiveness. The goal of my analysis is to cluster respondents according to these employer attractiveness dimensions. I plan to reduce the dimensionality with PCA and also confirm the structure with CFA before getting to cluster analysis. 48 items were measured in forced 4-point Likert scale, but the data I obtained for analysis was transformed by another research to binomial level - important or not important (0 or 1 values). I was wondering what effect that can have on PCA and cluster analysis that these data are binomial. Can these analysis techniques be executed on binomial data?
Using a "common sense" approach the trasformation from 4 level variables into dicotomic variables have clearly reduced the richness of information expressed in each variable, so I would expect more difficulty in extracting conclusions from statistical analyses on such dataset than if you would have analyzed original data. Considering the topic you have addressed, PCA/Factor analysis and Cluster analysis for binary data, I would give a look to: * Homogeneity analysis: (see http://www.r-bloggers.com/finding-patterns-amongst-binary-variables-with-the-homals-package/), a sort of factor analysis that can be used to analyze binary data. * The mona function in R cluster package: a cluster analysis tailored for binary data (see Cluster analysis of boolean vectors in R)