Interpretation of standardized coefficients

[Abhilasha Sahay]

I know that we have previously discussed how the use of standardized coefficients may obfuscate results rather than shedding more light. I completely agree!

However, I am curious to really understand how one can use these coefficients, if one had to. Consider we run an OLS on standardized y and x, which yields the standardized coefficient beta.

std(y) = 0.191 std(x)

One may interpret this as 1SD change in X is associated with 0.191SD change in y. Consider the s.d of the unstandardized variables are sd(Y) = 27.37 and sd(X) = 0.094. In this case, would it be reasonable to infer that 1SD change in X (sd=0.094) is associated with 1.8% change in Y (0.094*0.191)?

Thanks in advance!

[Richard Williams]

If you really want to play around with standardized coefficients I suggest you do

findit spost13_ado

install it, and then use the listcoef command. It can standardize y, standardize x, or both, and will do it for commands besides regress, like logit. You seem to want to do x-standardization, which I think is often superior to the more common xy standardization.

The best thing to do is read Long & Freese's book, which also describes a bunch of other great commands:

https://www.stata.com/bookstore/regr...ent-variables/

But, if you just want a quick discussion of listcoef, see

https://www3.nd.edu/~rwilliam/stats3/L04.pdf

[Abhilasha Sahay]

Thank you Dr.Williams for your prompt response!
I am going through the references on listcoef and I realized that most of this material focuses on binary dependent variable. I guess my question is concerning the interpretation of the coefficient, i.e. how can we convert std to % change. The method I described seems to be a back of the envelope calculation, which some people have used and I wanted to vet that from experts like yourself.

Look forward to your thoughts!