omitted because of estimability

[Martin Rabey]

I have run a stepwise logistic regression model:

stepwise, pr(0.1): logit recovered Cop i.new_pain_interference bmi etc etc.

and recieved the following error message:

note: 0b.new_pain_interference omitted because of estimability.

however the variable that the note above mentions appears in the final model:


-----------------------------------------------------------------------------------------
recovered | Coefficient Std. err. z P>|z| [95% conf. interval]
------------------------+----------------------------------------------------------------
sebt_normalised | .0172445 .0103059 1.67 0.094 -.0029547 .0374437
1.new_pain_interference | 1.502251 .7328757 2.05 0.040 .0658414 2.938661

Stata's documentation for stepwise states: "Stata’s solution of dropping the offending variables and observations is, in general, appropriate."

I am therefore confused that stata says that it has omited the variable, and that stepwise appears to drop offending variables, but the variable still appears to be in the model.

Any guidance welcomed.

Thank you

[Carlo Lazzaro]

Martin:
welcome to this forum.
Stata omitted the reference category of your categorical variable and kept level 1 as a predictor.
You can find something similar in -lasso- entry (page 158):

regress mpg 0bn.foreign 3bn.rep78 5bn.rep78 weight turn gear_ratio price length

[Martin Rabey]

Thank you Carlo - that explains it for me. Much appreciated.