[Limdep Nlogit List] Line search does not improve the function.

[Lixian Qian]

Dear professors,

I am running a Nested Logit Model in NLogit v5.0.

I find the location of alternative specific constants (ASCs) in the utility function can impact the model convergence.

When I specify the utility function with 3 alternatives as:
U(A) = a1*X1 + a2*X2  /
U(B) = a1*X1 + a2*X2 + asc_b + age_b*Age + sex_b *Gender /
U(C) = a1*X1 + a2*X2 + asc_c + age_c*Age + sex_c *Gender
The model estimation shows that  Line search at iteration   61 does not improve fn. Exiting optimization.

However, when I make a minor change on the utility functions to
U(A) = a1*X1 + a2*X2  /
U(B) = asc_b+ a1*X1 + a2*X2 + age_b*Age + sex_b *Gender /
U(C) = asc_c + a1*X1 + a2*X2 + age_c*Age + sex_c *Gender
The model can successfully converge, with the exact same LL function, all estimated coefficients as well as standard errors as in the previous case (though not converged as mentioned in output).

I don't know what is wrong with the first case.

Thank you and appreciate your advices.


Best,
Lixian