[Limdep Nlogit List] RPL/ML McFadden/Train specification test

[Iain Pardoe]

I thought I'd share some simple code for computing the McFadden/Train
specification test (J. App. Econometrics, 2000) to determine if mixing
is needed for a multinomial logit model:

read;file="C:\Documents and Settings\ipardoe\My
Documents\limdep\misc\mt.txt";
nobs=27924;nvar=27;names=id,price,range,acc,speed,poll,size,big,lugg,cos
t,stav,
suv,spo,wag,tru,van,ev,comev,colev,cng,meth,colmet,nous1,ch,noev,nocng,n
ous2$

create ; rang1 = range/100
       ; acc1 = acc/10
       ; spee1 = speed/100
       ; size1 = size/10
       ; cost1 = cost/10 $

NLOGIT ; Lhs = ch
       ; Choices = A1,A2,B1,B2,C1,C2
       ; Rhs = price,rang1,acc1,spee1,poll,size1,big,lugg,cost1,stav,
               suv,spo,wag,tru,van,ev,comev,colev,cng,meth,colmet
       ; Prob = prmnl $

calc   ; list
       ; loglr = logl
       ; kr = kreg $

namelist ; z = price,rang1,acc1,spee1,poll,size1,big,lugg,cost1,stav $
matrix ; wmeans = Gsum(z,prmnl,id) $ ? limit 10 vars in matrix with this
many groups
create ; zpric = .5*(price-wmeans(id,1))^2
       ; zrang = .5*(rang1-wmeans(id,2))^2
       ; zacc  = .5*(acc1-wmeans(id,3))^2
       ; zspee = .5*(spee1-wmeans(id,4))^2
       ; zpoll = .5*(poll-wmeans(id,5))^2
       ; zsize = .5*(size1-wmeans(id,6))^2
       ; zbig  = .5*(big-wmeans(id,7))^2
       ; zlugg = .5*(lugg-wmeans(id,8))^2
       ; zcost = .5*(cost1-wmeans(id,9))^2
       ; zstav = .5*(stav-wmeans(id,10))^2 $

namelist ; z = ev,cng $
matrix ; wmeans = Gsum(z,prmnl,id) $
create ; zev   = .5*(ev-wmeans(id,1))^2
       ; zcng  = .5*(cng-wmeans(id,2))^2 $

NLOGIT ; Lhs = ch
       ; Choices = A1,A2,B1,B2,C1,C2
       ; Rhs = price,rang1,acc1,spee1,poll,size1,big,lugg,cost1,stav,
               suv,spo,wag,tru,van,ev,comev,colev,cng,meth,colmet,
 
zpric,zrang,zacc,zspee,zpoll,zsize,zbig,zlugg,zcost,zstav,zev,zcng $

calc   ; list 
       ; loglu = logl
       ; ku = kreg
       ; lrtest = 2*(loglu-loglr)
       ; df = ku - kr
       ; pvalue = 1 - Chi(lrtest,df) $

This replicates the results in table 3 of the McFadden/Train paper
(although not exactly for some reason).  The data are available from the
J. App. Econometrics data archive.  There's probably a cleaner way of
calculating the artificial z-variables using matrices but I'm still
pretty new to limdep so this is the best I can do just now ...

Best wishes, Iain

Iain Pardoe <ipardoe at lcbmail.uoregon.edu>
Assistant Professor of Decision Sciences
Charles H. Lundquist College of Business
1208 University of Oregon
Eugene, OR 97403-1208, USA
Ph: 541-346-3250, Fax: 541-346-3341
http://lcb1.uoregon.edu/ipardoe