Dear Allstat =
1. I would appreciate members' advice on two aspects of canonical correlation before I run this final stage of my data analysis.
2. The first is whether - though it has worked perfectly well in multiple regression - my mix of binary and numerical/ continuous independent variables is likely to create problems
of any kind in running the canonical correlation procedure or in assessing and interpreting its output (in SPSS14)? These 10 predictors comprise 4 single binary and 6 composite variables. Half of the 6 are binary and the remainder numerical/ continuous. All (4) of the DVs are numerical/ continuous.
3. My second query is how much value-added, if any, does my data analysis gain from using canonical correlation to complete the process- in this particular case? The four earlier stages have aimed to be progressively more sophisticated, involving in turn: bivariate correlations, and 3 variants of multiple regression (separate testing of each of the (8)research hypotheses, using subsets of the IVs; standard MR (using all 10 IVs, I DV) and standard MR using a standardised composite DV. The latter is a compound of the 4 (standardised) DVs.
4. Given that the 4 DVs (which will comprise in CC the Y variate) have already been combined to form the compound dependent variable in stage 3 of the MR, is the use of canonical correlation as the final stage of analysis likely to produce any extra benefits/ insights, or will it be useful merely as an additional test of the existing results? If it is capable of producing some value-added in the context indicated, what form is this likely to take?
With advance thanks =
Owen Murphy
|