Test of significance

[JB]

Hello,

Thank you very much for this new program for GSCA ! I have a question about testing the significance. I know that I can use the confidence intervals for that. However, if I want to do a Student's t, I can also do it with the formula: Estimate / SE with a df equal to the number of bootstrapping minus one. Nevertheless, it happens that I have a significant Student's t when the IC95 confidence interval contains zero! What should I choose in this case?!? Is the confidence interval more conservative?

Thank you for your help!

Jean-Baptiste

[heungsun.hwang]

Dear Jean-Baptiste,

Thank you for your inquiry. We rather prefer (bootstrap percentile) CI over a t-test because the t-test is a parametric test assuming normality of a parameter estimate. This parametric test is not consistent with GSCA (or PLS) which typically does not require a distributional assumption. No literature is available that shows that GSCA estimates are normally distributed.  

Best,

Heungsun