chapter  6
Principal Component Analysis: Relationships Within a Single Set of Variables
Pages 76

The techniques we have discussed so far have had in common their assessment of relationships between two sets of variables: two or more predictors and one outcome variable in multiple regression; one predictor (group membership) variable and two or more outcome variables in Hotelling’s I2; two or more group membership variables and two or more outcome variables in Manova; and two or more predictors and two or more outcome variables in canonical correlation. Now we come to a set of techniques (principal component analysis and factor analysis) that “look inside” a single set of variables and attempt to assess the structure of the variables in this set independently of any relationship they may have to variables outside this set. Principal component analysis (PCA) and factor analysis (FA) may in this sense be considered as logical precursors to the statistical tools discussed in chapters 2 through 5. However, in another sense they represent a step beyond the concentration of earlier chapters on observable relationships among variables to a concern with relationships between observable variables and unobservable (latent) processes presumed to be generating the observations.