Re: mulitivariate linear regression and Bonferroni

Jerry Dallal <gdallal@xxxxxxxxxxxxxxxxxxxxxxxx> wrote in
news:ppCdnfrJJMTe6IzZRVn-tA@xxxxxxxxxxx:

If an investigator were to come to me with a multiple regression
model and a question about which regression coefficients might be
statistically significant, without any prior thought about narrowing
down the field, I'd certainly demand an adjustment, as would most people
I work with.


I think you need more info before making such a conclusion. There are
better ways than a bonferroni adjustment that you can use to choose a
"good" model IN THE ABSENCE OF ADDITIONAL INFORMATION, such as Mallows Cp,
along with a scrutiny of residuals.

If you have an a priori reason to include an independent variable, though,
and your experimental design limits concerns over multicolinearity, and
you're not doing anything where you're running more than one model, a
Bonferroni correction hardly seems necessary.

I suspect that with 9 independent variables, the referee seems concerned
about inferences that are coming out of the multilinear regression, and he
wants you to defend inclusion of some of the independent variables. If the
OP can demonstrate that some thought went into the model, and that he
wasn't shooting with a shotgun to see what got hit by pellet's (the Cheney
approach to stats!), he'd have a fairly easy time steering the criticism.



--
Scott
Reverse name to reply
.

Quantcast