Need your help for instrumental variable estimation on panel data with binary outcome

[Rodrigue Kaki]

Dear usesrs
I am doing research on the impact of mentoring programme on employment. The outcome variable (called employment) is binary (1=got an employment, 0=otherwise). The independent variable is “Status” (0=Control group; 1=Treatment group). And I have 2 controls variables (sex, and age). I have panel data.
First, I run the impact of offering the programme (ITT) on the outcome using a mixed effect logit model on panel data.
To avoid endogeneity, I have chosen an instrumental variable (called Participation) (with 0=Do not really participate in the programme; and 1=Participate in the programme). I would like to run an instrumental variable mixed effects model on panel data. I would like to know if the command “xtivreg” is appropriate. I mean the following command with stata 16.1:
Xtivreg Status sex age (status=Participation),
I am facing problem with the good command as the outcome is a binary variable and also, I am in face of panel data. Need your help please!!!

[Fei Wang]

It's ok to use -xtivreg-, but I'm a bit confused about the endogenous variable and instrument variable. The independent variable in the ITT equation (random treatment assignment -- "Status"?) should be the IV for the self-selection into the program ("Participation"?). If I understand correctly, the code should be

xtivreg employment sex age (Participation = Status), options

[Rodrigue Kaki]

Thanks Fei Wang for the code. My concern now is the code. How to set the options to ensure that I am taking into account the mixed effect?

[Fei Wang]

A conservative setting would be

xtivreg employment age i.period (Participation = Status), fe vce(cluster Status)

[Rodrigue Kaki]

The new code is not clear for me. Could you please provide more explanations?
Thanks

[Fei Wang]

Option -fe- means controlling for individual fixed effects, -vce(cluster Status)- means standard errors being clustered at the treatment level, and -i.period- means period dummies.

[Rodrigue Kaki]

Is it possible to control for both individual fixed and random effects in the same model?

[Fei Wang]

Is it possible to control for both individual fixed and random effects in the same model?

No, you can't. They are competing estimation approaches.

[Rodrigue Kaki]

Thanks Fei Wrang