Getting exponentiated coefficients of a multilevel regression ("mixed" command)

[Agata Troost]

Hello, my question seems simple to me, but I can't find an answer.

I have a multilevel linear regression model and would like to have a table with exponentiated coefficients. Yet neither "svy, eform: mixed y x1 x2 || x3:" nor "mixed y x1 x2 || x3:, eform" works. After the latter I get an error "option eform not allowed r(198);" with no further explanation

I would be grateful for any help

[Clyde Schechter]

display exp(_b[x1]), exp(_b[x2]), exp(_b[x3])

[Agata Troost]

Thank you! But this is for separate coefficients no? Isn't there a way to just get the entire table with exponentiated coefficients, variance etc.? (I want to export it to a Word file with eststo and esttab, as I did with the non-exponentiated output)

[Maarten Buis]

What options are allowed is listed in the help file. eform is not in the list of allowed options after mixed. So that is in line with what the error message tells you.

Why do you want to see the exponentiated coefficients? What do you think these exponentiated coefficients mean? My suspicion is that you have a log transformed dependend/explained/left-hand-side/y-variable, and you want to move the coefficients back to the original metric. This does not work, because a logarithm is a non-linear transformation. You could interpret these in term of geometric means rather than arithmetic (normal) means, but does your audience know how to interpret those?

[Agata Troost]

Hi, I indeed have a log-transformed dependent variable and I was interpreting exponentiated coefficients as explained here: https://stats.idre.ucla.edu/other/mu...g-transformed/

I have interpreted my predictor of interest (which is not log-transformed, although in some of my models one of the other predictors is) step-by-step in the results section, so I hope the audience can follow that. But I don't interpret all predictors from my model, so I thought for people who want to do it themselves quickly it would be useful to provide the exponentiated coefficients in a table. I see now, however, that it can't be easily done in Stata - probably the fastest solution would be to exponentiate the coefficients with a command in Excel

[Maarten Buis]

The discussion in the link is correct. The problem I have, is that if I present my results I am pretty sure that my audience will ignore the geometric part of geometric mean because they don't understand it. At which point the results would be correct but misleading. So I suggest that you do not do so.

Instead you could look at https://blog.stata.com/2011/08/22/us...tell-a-friend/