[Limdep Nlogit List] Help with LCM
Alec Miners
Alec.Miners at lshtm.ac.uk
Wed Jan 20 00:17:01 AEDT 2016
Dear All
I am analysing a labelled DCE, using both dummy and effects coded models, using a LCM. There are 5 alternatives, and each respondent completed 12 DCE questions
The two models give exactly the same log-l'hoods (ie the fit is the same), when i exclude any class parameters (eg. age) from the LCM line. ie using the syntax below. Clearly the coefficients are different. Fine up to this point using the code below
NLOGIT
;checkdata
;choices = CHAT,ONLINEGR,ONLINE1T, EMAILORT, NONE
;lhs= choice1
;rhs = ONLINEGR,ONLINE1T, CHAT, NONE, EF2FGROU, EPHONE1T, ESESSION, ESESSIO0, EMINS15T,EMINS30T, ENURSE, EPEER
;LCM
;pds=12
;pts=3
$
NLOGIT
;checkdata
;choices = CHAT,ONLINEGR,ONLINE1T, EMAILORT, NONE
;lhs= choice1
;rhs = ONLINEGR,ONLINE1T, CHAT, NONE, DF2FGROU, DPHONE1T, DSESSION, DSESSIO0, DMINS15T,DMINS30T, DNURSE, DPEER
;LCM
;pds=12
;pts=3
$
However, when I start including statements such as (in both the effects and dummy models)
;LCM = d_young (a 0/1 variable indicating whether someone is young or not), the model fits and class proportions begin to differ but only when pds > 2.
I am confused as to why this would be the case??
All help much appreciated
Bw Alec
Alec Miners, PhD
Senior Lecturer in Health Economics
Dept Health Services Research & Policy
London School of Hygiene and Tropical Medicine
Room 134
15-17 Tavistock Place
London
WC1H 9SY
Tel 0207 927 2069
Fax 0207 580 8183
More information about the Limdep
mailing list