Interaction Collinearity Multilevel Regression

[Sofia Statalover]

Hi,
we are investigating the relationship between job satisfaction (Y) and some main predictors, among which jchall, an index (1-4) that sums up how significant the job tasks are considered by the workers. We also created a dummy variable to account for missings in jchall, called miss_jchall, and we inputed jchall=0 if miss_jchall=1. We ran several multilevel regression models without any problems, but then when adding an interaction term between jchall and imprik (which is a measure of how important for respondents is to be rich), stata output shows omitted values for jchall, but just for the last values (3 and 4). The output says that it is due to collinearity. Could someone help us with that?

[Carlo Lazzaro]

Sofia:
welcome to this forum.
As per FAQ, please note that real family name are preferred on this forum.
That said:
1) creating a two-level categorical variable for missing values and plug it in as a predictor in the right-hand side of your regression equation is a poor approach, as it produces biased coefficients (see

https://us.sagepub.com/en-us/nam/missing-data/book9419:

9-11);
2) I'm not clear with your statement about "running several mutilevel regression". Did you use -mixed-?
3) without an example/excerpt of your dataset (that you could share via -dataex-, as per FAQ again) it seems that levels 3 and 4 of that predictor are perfectly collinear with some other predictor: therefore, Stata omits them.