Hereby you receive a question about bias nodes in Artificial Neural Networks.
The lecture slides on Artificial Neural Networks indicate that (the weights of) bias nodes are analogous to the intercepts in logit models. In other words, bias nodes can be conceived as Alternative Specific Constants.
However, in what way do we specify bias nodes i.e. ASCs in the Learning-MNL model? I can imagine that, during research, ASCs are of behavioural interest since these constants depict the average utility of the unobserved error term. Yet, I find it difficult to add these bias nodes to the RUM-MNL part of the L-MNL model. Is it even possible (since bias nodes are associated with the MLP part rather than the RUM-MNL part of L-MNL models).
Dear Mr. Van Cranenburgh,
Hereby you receive a question about bias nodes in Artificial Neural Networks.
The lecture slides on Artificial Neural Networks indicate that (the weights of) bias nodes are analogous to the intercepts in logit models. In other words, bias nodes can be conceived as Alternative Specific Constants.
However, in what way do we specify bias nodes i.e. ASCs in the Learning-MNL model? I can imagine that, during research, ASCs are of behavioural interest since these constants depict the average utility of the unobserved error term. Yet, I find it difficult to add these bias nodes to the RUM-MNL part of the L-MNL model. Is it even possible (since bias nodes are associated with the MLP part rather than the RUM-MNL part of L-MNL models).
Thank you in advance for considering my question.
Best regards, Milan Moleman