Re: mulitivariate linear regression and Bonferroni


Herman Rubin wrote:

Some rather UNWARRANTED strong statements in his follow-up to

In article <tpadnUzPmpc96IzZ4p2dnA@xxxxxxxxxxx>,
Jerry Dallal <gdallal@xxxxxxxxxxxxxxxxxxxxxxxx> wrote:
Data Matter wrote:

< snip >


I suggest that a decision approach be taken instead.

That is certainly one approach you or anyone can take.


The use of significance levels is a holdover from the dark
ages, and it has always been misused.

That statement is categorically WRONG! And you have nothing
of substance to back up your unilateral claim.


Those who understand
it know that "significance" does not mean "importance".

Or even "practical usefulness". I think all good data analysts
understand that. ALL of my students who passed my Data
Analysis courses understand that. So, what's your point, Herman?

Just because SOME folks don't understand that idea doesn't
make it a universal strawman for your to throw darts at, does it?


With 10 variables, with a sufficiently large sample on can
ignore most aspects of the loss-prior combination for
prior Bayes analysis. But is the sample large enough? In
this case, there is no semi-objective reasonable procedure.

Bayesian or non-Bayesian, each coefficient HAS the
unequivocal INTERPRETATION of the effect of THAT
independent variable IN THE PRESENCE of all the other
independent variables in the model, or more precisely, the
"partial correlation information between the particular
Xi and Y, given all the other Xj in the model"..

Once you know the MEANING of how that coefficient should
be interpreted, the rest are just different approaches to
extracting and making use of the PROPER information.

That was why I commented as I did in my follow-up. The OP doesn't
even know a multiple regression from a multivariate regression,
and every one leaped into his frying pan without even ASKING the
first question that should have been asked:

WHY did he suppress the constant term?


RF> Then, what you have is a UNIVARITE multiple regression, but not
RF> a MULTIVARIATE regression, which requires you to have TWO or
RF> more dependent variables.

RF> Why did you suppress the constant?

RF> If I were the reviewer, I would have commented as the above first,
RF> before asking what you were trying to do, and why do you do it
RF> this way.

-- Bob.

.

Relevant Pages

  • Re: multicollinearity in regression
    ... control variables but on which hypotheses are based. ... I could use Analysis of Covariance but 2 of the independent variables ... There are enough professional Quacks in the Multiple Regression ...
    (sci.stat.consult)
  • Re: Linearity of Independent varialbles in Multiple Regression.
    ... I wanted to know is there any way to check the linearity of Independent variables using sas. ... Estimating optimal transformations for multiple regression and correlation. ...
    (sci.stat.math)