R help - Jun 2011 - mlogit model that contains both individual-specific parameters and universal parameters

Hello, 

I am pretty new to mlogit, and still trying to figure out what models to use.I
have a data set of N individuals, each of which faces I alternatives. The
utility function of individual n, for choice i is: 

u(i,n) = alpha(i) * x1(i,n) + beta * x2(i,n) 

where alpha(i) is the individual specific parameter, and beta is shared among
all individuals. I don't really know how to set this up in mlogit. 

If I assumed that beta is individual-specific (beta(i)), then I can divide the
data set to many subsets, each of which corresponds to a particular individual
i, and run this model for each subset to estimate alpha(i) and beta(i). 
y ~ x1 + x2 
This can be done just fine. 

I have gone over tutorials by Train and by Heshner but I haven't found out
how to solve this problem yet. Any suggestions are welcome. Thank you so much
for your time!

	[[alternative HTML version deleted]]

On Wed, 22 Jun 2011, Quang Anh Duong wrote:
> Hello,?
>
> I am pretty new to mlogit, and still trying to figure out what models to 
> use.I have a data set of N individuals, each of which faces I 
> alternatives. The utility function of individual n, for choice i is:?
>
> u(i,n) = alpha(i) * x1(i,n) + beta * x2(i,n)?
>
> where alpha(i) is the individual specific parameter, and beta is shared 
> among all individuals. I don't really know how to set this up in 
> mlogit.?
I guess you mean that alpha(i) is the alternative-specific coefficient of 
the individual-specific regressor x1(n)? And x2(i,n) is an 
alternative-specific regressor with coefficient beta. If so, the model is
y ~ x2 | x1. (Possibly, you may want to exclude the alternative-specific 
intercepts.)

But see the extensive package vignettes for more details:

   vignette("mlogit", package = "mlogit")
   vignette("Exercises", package = "mlogit")

hth,
Z
> If I assumed that beta is individual-specific (beta(i)), then I can divide
the data set to many subsets, each of which corresponds to a particular
individual i, and run this model for each subset to estimate alpha(i) and
beta(i).?
> y ~ x1 + x2?
> This can be done just fine.?
>
> I have gone over tutorials by Train and by Heshner but I haven't found
out how to solve this problem yet. Any suggestions are welcome. Thank you so
much for your time!
>
> 	[[alternative HTML version deleted]]
>
>