How to set the initial paramater

[hamid004]

I am trying to estimate a 2 level nested logit with 4 alternatives. I have seen that you set the initial parameters according to the work on biogeme using sweissmetro data. Is it compulsory I set the initial parameters if I don't intend to work with the or model my own work according to that in the biogeme

[timothyb0912]

Hi @hamid004 ---yes setting the initial parameters is compulsory.

Even when not using Biogeme data, I always estimate nested logit models after first estimating an MNL model,
which provides initial parameters.

I suggest starting from the coefficient values returned by an MNL without nesting and starting with scale parameters that are near 1.0 (say 1.5) for your nesting coefficients. I also suggest trying multiple starting values in different runs since the log-likelihood is not concave. Be sure to enter the transformed version of the scale parameters, as shown in the cells under "6b" in the example notebook.

[hamid004]

Thanks so much and I wish you a quick recovery. Having applied the method you suggested, I was able to estimate the nested logit but the results wasn't good . one of the nests has a very high co efficient like 20.076 while the other co efficient has a value less than 1 and all parameters aren't significant at all.
Also I would love to add other case specific variables to my model how can I add those.

Thanks