Hello all,
I'm running a model with generalized least squares (GLS) and I need to compare the relative importance of predictors, hence the standardized coefficients.
I tried obtaining these coefficients using the z-scores of my predictors (scaled variables), but when I run the model, the intercept is not equal to zero (so I presumed the coefficients are not standardized).
Why this procedure did not work for GLS?
Is there another way to get standardized coefficients for GLS?
Can I compare t-values in order to compare the relative importance of my predictors?
Thankyou,
Your procedure is fine. The intercept is not zero because it is an estimate of the popn intercept which would be 0 if all possible data was standardized. Best, D. Booth
I agree that the procedure is fine. What you need to consider when talking about "compare the relative importance of predictors" is to what extent "ceteris paribus makes sense. Possibly you used GLS because you feared that the predictors were correlated, if so can you change a predictor without simultaneously changing others?
Thankyou for your answers!
Kenneth Carling , thankyou for your response. I used GLS because I suspected my residuals had an phylogenetic autocorrelation structure, and I tested for multicolinearity among my predictors before running the model. So, I don't understand why changing one predictor would simultaneously change the others, could you clarify that?
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression