Report standardized beta coefficients in stepwise hierarchical regression

[Hanna Aileen]

Dear All,

I am running different regression analyses including a hierarchical regression with an interaction term. I am currently trying to find a way how to report the standardized beta coefficients in such regressions.

I used this command, but Stata seems to ignore the beta option for it showed neither the beta coefficient nor an error message:

stepwise, pe (.20) hierarchical: reg y x1 x2 (x3 x4) (x3*x4), beta

Is there a possibility to report the standardized beta coefficients in hierarchical regressions?
And if so, is there a way to use the estout command or something similar for those stepwise regressions?

Thanks in advance!

[Doug Hemken]

If this command runs, you could use

stdBeta

as a postestimation command, see

findit stdBeta

.

Can you provide a working example? When I try

stepwise, pe(.2) hier: regress mpg weight (displ price) (disp#price)

I get the error message that interactions are not allowed.

I think your best bet is to first standardize your variables, second form the interaction terms, then third perform your hierarchical regression.

[Hanna Aileen]

Thanks Doug Hemken! I will try it this way.
Unfortunately I was not able to install the package in Stata, the file could not be found/opened. Is there another way to install it?

True the interaction term like that

x1#x2

is not allowed - I formed the term manually before performing the regression and then it works.

Another problem I encountered is that when creating a regression table with the estab command like that:

eststo model3: quietly reg y1 x1 x2 x3 x4 c.x3#c.x4, beta
esttab model3, beta b(a3) star(* 0.05 ** 0.01 *** 0.001) r2(3) ar2(3) scalars(F) nogaps

there is a gap in the column for the beta coefficient of the interaction term. This error only happens in that combination. When running the regression command without estout

reg y1 x1 x2 x3 x4 c.x3#c.x4, beta

the beta of the interaction term is indicated. And when running the same command without beta the unstandardized coefficient of the interaction term is also given.

eststo model3: quietly reg y1 x1 x2 x3 x4 c.x3#c.x4
esttab model3, b(a3) star(* 0.05 ** 0.01 *** 0.001) r2(3) ar2(3) scalars(F) nogaps

Do you maybe know how to solve this problem?

[Doug Hemken]

While

regress, beta

reports coefficients, notice that these are standardized term by term, not variable by variable. This makes it problematic to interpret the standardized model as is, when there are interaction terms. I would guess this is why Jann has limited the usefulness of

esttab, beta

to models with only first-order terms.

The work-around is to first standardize your variables, then form interaction terms, then re-estimate your final model. (Term by term standardization reverses the first two steps.)

Although I previously suggested redoing the hierarchical procedure, you would NOT really want to do this: you would be testing substantively different first order coefficients. Just re-estimate the final model.

stdBeta

appears to be broken on the SSC archive website. You could instead

net from http://www.ssc.wisc.edu/~hemken/Stataworkshops

[Hanna Aileen]

Thanks a lot Doug Hemken , that reallly helped and it works now!