First of all, I am not a statistician, but I have a working knowledge of statistics from college course work and vocational practice. I am working with experimental data, where I have 9 cases, 32 independent variables, 12 unrelated dependent variables. I have found in most cases that a subset of these 32 indy vars is highly linearly correlating with these a these dependent variables, but the process of finding them is long an drawn out.
Theoretical correlations aside, I want to optimize the subset of these independent variables that generates coefficient t-values > 2 and highest R^2. My current routine is to start with all 32 indi var vals, run the PLS regression with SYSTAT 13, copy the coefficient/error table to excel, where a VBA macro calculates and sorts the t-values. Then I repeat the above, but narrowing the variable list by culling the lowest t-value variable. I repeat this until I have t-values on all variables > 2.
If the R^2 is less than 3 nines, then I go back and find the step (in the calc-drop-lowest-tvalue-repeat routine) where the R^2 value when down, instead of up. I start from there, but instead of dropping the lowest t-value variable, I cull the one above it and run the regression, to see if better than 3 nines R^2 is obtainable.
My question is, does SYSTAT have a way to do this within the package itself or is there a canned SYSTAT 13 script that accomplishes this?