How to get effect sizes (for direct effects and interactions) in xtreg, re

[Michael Reuter]

Dear Statalisters,

this may be a dumb question, but I could not find the answer anywhere so far: How do I determine effect sizes in a random-effects panel model (xtreg, re)?

I am interested in two things:

1) What is the effect size of the direct effect of a given variable? Is it simply the coefficient in the model? Are these coefficients already standardized so I can interpret them as "one standard deviation change in independent variable is associated with a <coeffcient> standard deviations change in the dependent variable?

2) What is the best way to interpret effect sizes in an interaction? I calculated marginal effects of one independent variable at different levels of the other independent variable -- but how do I interpret them?

Best,
Michael

[Jeff Wooldridge]

Michael: Estimating a linear equation using the RE estimator (which is a particular generalized least squares estimator) has no impact on how we interpret the coefficients in the equation. The independent variables won't be standardized unless you do that first. For the purposes of interpretation, It's just like estimating a linear model by OLS. If you have interactions then you have to be careful in interpreting the main effects unless you've centered the variables first (or you use the margins command). Using standardized independent variables ensures they are properly centered.