Omitted variables that are needed

[Janik Kaden]

Hello,
I am working on a meta regression. For testing for publication bias I am conducting the FAT-PET-PEESE procedure. For that I generated (among others) the variables SE and SEinv, which is standard error and inverted standard error. When I regress, SEinv is omitted due to collinearity. But for the procedure I need also SEinv. Is therer a way to supress the omission (not only the display option "noomitted")?
I would be really grateful for help.
Bes regards
Janik

[Maarten Buis]

How did you create SEinv?

[Janik Kaden]

I created an extra column in my excel file that is just one divided by the respective SE value. But also when I generate it in stata with

gen SEinv = 1/SE

it does not work

[Felix Scholl]

When a variable is perfectly collinear with another variable or together with multiple variables (and I assume this is the case, because otherwise Stata would not omit the variable), there is simply no way to include the variable in a model. How could you possible estimate the certeris paribus of a variable if that varialbe is a linear of other variables?

[Maarten Buis]

Felix Scholl The inverse is a non-linear transformation, so normally you don't run into that problem, and you can add both. It may be that all standard errors are in a range, that the inverse is so close to linear that it becomes impossible. However, since we don't have the data there is no way for us to determine that. Janik could look at the Statalist FAQ (top of the page in the black bar) to see how he can give us an extract of the data.

[Janik Kaden]

I rearranged some variables and out of sudden it worked out now. I still don't know what the problem was, but thanks anyway!