XTIVREG, variance explained

[socialscientist]

I am currently working on an R&R and one reviewer requested that I provide the variance explained by each covariate. Any suggestions would be very helpful. Thanks.

[Paul T Seed]

To get a worthwhile answer (or any at all), you need to say more about what you are doling. What is R&R? (presumably not "Rest & Relaxation") What code have you used so far? What output did you get? The Statalist FAQs explain more.

For a simple multiple linear regression model, the variance explained by a particular predictor is the decrease in the error variance when the predictor is added to the model.
This,however, is an ambiguous definition, as the variance explained depends on the other variables in the model at the time. For -tivreg-, you have to calculate error variance from the residual sum of squares and residual degrees of freedom. Both are returned by -xtivreg- :

e(rss) residual sum of squares
e(df r) residual degrees of freedom

[Jeff Wooldridge]

Tell your reviewer to join 21st century data analysis. More seriously, Paul's suggestion seems the best that can be made. Anything you do is pretty meaningless without uncorrelated covariates. If you are using random effects there is the extra complication of the error variance having two components.

[Valeria Frigerio]

I am trying to do something similar, after estimating a Bond-Blundell dynamic panel model (xtdpdsys), in which I have my dependent variable regressed over a set of individual characteristics and some macroeconomic variables which vary over time, for example the Euro/US$ exchange rate, and sometimes between groups of individuals (e.g.: gdp growth in the country of residence).

What i would like to measure, among other things, is the relative importance of individual and macro factors, decomposing explained variance, in order to try to assess whether my dependent variable is affected more by the business cycle or by individual aspects.

Any suggestions would be welcome, thanks.

[Joseph Luchman]

Valeria,

The technique that comes closest to determining relative importance of regressors in xtpdsys is most likely Luo and Azen's (2013) technique using dominance analysis in linear mixed models. There is an implementation of their method available for domin (SSC) with the wrapper program mixdom (included with domin upon ssc install-ing).

- joe

Luo, W., & Azen, R. (2013). Determining predictor importance in hierarchical linear models using dominance analysis. Journal of Educational and Behavioral Statistics, 38(1), 3-31.