Standarised coefficients

[Sterimar R.]

Dear All,

I am trying to replicate some one results. The regression uses a simple OLS cross-country estimates. It is trying to explain income as function of institutions, geography and trade. It runs the following regression:

lnGDP1995_i = a + b RuleLaw_i + cDistanceEquator_i + d lnOpenness_i

where the regressors (and not the dependent variable) are "scaled in the sense that they represent deviations from the mean divided by the standard deviation".

Two questions:

(1) Is there a straight forward way to do this with stata? The option "reg y x ,beta" standardises the dependent variable too.
(2) Less stata related - but I have never seen this type of regression before: how do you interpret the coefficients? is it "one standard deviation change in RuleLaw has a b effect on (ln) Income?"

Thank you very much for any help!

Sterimar

[Clyde Schechter]

For (1), there is no simple option analogous to -beta- that you can specify on the regression command. But you can quickly generate the standardized predictors using the -egen- command's -std()- function.

(2) Your suggested interpretation of the coefficients is correct. This type of thing is typically done where there are multiple predictors that have arbitrary scales and one of the goals of the research is to try to make a formal scale-free comparison of the strengths of the different effects. (Though there is some controversy about whether such comparisons are valid anyway.)

[Dick Campbell]

The SPost software written by Scott Long and Jeremy Freese can be used to obtain coefficients of this type. Typing findit spost will give you a list of various versions of the package. If you have version 13 of Stata you should install the spost13 package. You will then have access to the listcoef command which will give you the standardization you want. Of course you can transform the variable directly as you indicated. The interpretation you suggest in your note is correct. You can find further discussion of the various kinds of standardized coefficients in Regression Models for Categorical Dependent Variables Using Stata (3rd ed) by Long and Freese. Your outcome variable is not categorical but their discussion applies.

[Richard Williams]

I highly recommend the Long & Freese book that Dick Campbell mentions. If that is overkill for your purposes, I describe the listcoef command and the different kinds of standardizations it does at

http://www3.nd.edu/~rwilliam/xsoc73994/L04.pdf

The handout focuses on logistic regression but the appendix talks specifically about OLS.

I am not that crazy about standardization in OLS regression but if you are going to do it I think X standardization usually makes as much or more sense as full standardization does.

[Sterimar R.]

Thank you everybody!

[Jorrit Gosens]

Another option is to run the regression on the original values. Then use tabstat to calculate the sd of all variables.
Multiplying coefficients from the original regression with sd for each variable will yield identical results