11-28-2014 12:50 AM
It would be very grateful if someone can help me..
I'm trying to use multiple linear regression with a priori testing hypotheses about autocorrelation and constant variance of residuals, but my model residuals show clear pattern, and strong positive autocorrelation.. (data are not time-series)
I know that OLS estimators even in this case will be still unbiased, but std. errors and confidence intervals with t and F statistics (and p-value) will be misleading...
My question is whether the problems mentioned above, are also problems for the PLS regression or Correlated Component Regression, or to be more precise, whether PLS or CCR can handle this problem and give me a reliable standard errors for (cross validated) R^2?