[Limdep Nlogit List] standard errors of rho in bivariate probit with selection

[rhythm Em]

Dear All,

I am using bivariate probit with sample selection (y1 is only observed when y2=1). My model looks like:

Bivariate ; lhs= y1, y2 ; rh1= X1, x2, x3, x4..; rh2= X1, x2, x3, x4... ; robust; selection $

rho of this model was significant upto adding 5 independent variables but when I added sixth one, standard error of rho suddenly bumped up. I tried with several other variables. Whatever I added as sixth variable, SE of rho surprisingly increased and making it non-significant. Was this type of problem reported before? I am in dilema of using univariate probit or bivariate with selection because of the rho issue, theory and data suggests to use bivariate with selection though.  What could be the alternative solution to my problem? 

I would appreciate your quick help. Thanks.

Sincerely,
Rhythm