Regress - are the regression coefficient standardized?

Hi

I am conducting a multiple linear regression analysis with the following regress command: [b,bint,r,rint,stats] = regress(y,X) Where b is the coefficient array. When I want to compare the different regression coefficients with each other in order to estimate the influence on the response in y I generally have to standardize the regression coefficients in order to make them comparable. My question is: Are the regression coefficients already standardized when applying the regress command OR do I have to do it manually?

Thank you very much.

[Oleg Komarov]

The second but you can normalize your regressors with zscore.

[Alwin]

Thanks, Oleg. I normalized the regressors manually using the same equation zscore applies, before I read your comment. Luckily the results conform each other. Another question I have is, if I still have to include the ones columns as a constant (e.g. [ones(size(x1))) when applying regress. I tried both with and without constant and the regression coefficients seem to be the same. So it doesn't matter if I include the constant column, right?

Answer 1

[Oleg Komarov]

No, regress doesn't normalize automatically the regression coefficients.

One workaround is to normalize your regressors with zscore

What about the intercept?

The standardization centers the data around 0 and the regression line should go through the origin. However, in small samples you may still get a significant alpha.

• big sample: do not include the ones (but double check anyways, alpha should be 0 or non significant).
• small sample: include the intercept.