Hello R Help,
I have a question regarding Multinomial (Conditional) LOGIT models in R.
For my masters thesis I like to use Multinomial LOGIT models to analyse
consumer choice data.
After orientation on the R homepage and internet I found a method for
multinomial LOGIT. This model is as follows:
P_s(X_i) = exp(B_s * X_i) / som_j(B_j * X_i)
s = Class or Response (bought brand A, B, C or D)
X_i = Characteristics that describe characteristics of purchase incident i.
B_s = Parameter vector for Class s.
This model estimates for each class a B_s vector. (or actually one less
because B_0 = 1)
I use the function multinom from the nnet package.
Now I am also interested in Conditional Multinomial Logit model with the
following formulation.
P_s(X_is) = exp(B * X_is) / som_j(B * X_is)
s = Class or Response (bought brand A, B, C or D)
X_is = Characteristics that describe characteristics of purchase incident I
for Class s.
B = Parameter vector for all classes s (only one..).
This model estimates one B vector.
My question is: “Is it possible to estimate a Conditional Multinomial Logit
model as described above with R� And How??
Something that comes close is the implementation by John Hendrix in the
CATSPEC package. This allows for constraints on the response variable, but
it estimates a B_s vector for each Class and not one B vector. This is not
what I like to have and not according to the specification of a Conditional
LOGIT model as I found in literature.
I’m interested in commands on this. Is the information above correct because
it could well be that I am wrong since I am new to this kind of models.
Thanks in advance for any responses or suggestions.
Arne Jol
Student Informatics & Economics
Erasmus University Rotterdam
E-mail: arnejol at gmail.com
Tel: +31 618088152