R help - Dec 2012 - Relative strength of regression predictors (relaimpo vs. relimp)

Hello!

I am trying test my observed data against the null-hypothesis that
different items from a psychological questionnaire contribute equally to
the metric dependent variable that measures problems (sum score of a
questionnaire). That is, I am interested in relative strength of the
predictors.

Predictor items of the questionnaire are on a scale from 0-3, and
technically ordinal, although most people treat them as metric.

I predict Y in a linear regression by these items (that differ from each
other drastically in mean), controlling for age and sex. N=6000.

m1 <- lm(totWAS ~ 1234x+s5+q6789x+s10+s11+s12+s13+s14+q1516x+
sex+age, data=D)

Now I want to find out about relative strengths. Standardized coefficients
differ in the profile from results obtained by the relaimpo and the relimp
packages, the largest differences exist for the variable age (which is the
only variable I standardized; the standardized beta is very large compared
to the other risk factors, but the relative importance packages show that
is nearly has any importance).

Questions:
(1) What can lead to differences between relative importance package
results and standardized beta weights - mostly different means and standard
errors of predictors?
(2) What kind of procedure would you recommend - am I on the right track
with the relimp and relaimpo packages?
(3) Is there a way to only use standardized betas? I would prefer that
because it would enable me to use standardized confidence intervals to
reason that x1 has a meaningfully larger influence on y than x2 (if the CIs
do not overlap).

Thank you!
Torvon

	[[alternative HTML version deleted]]