[Limdep Nlogit List] MXL model: changing reference alternative- inconsistent ASCs

[Thao Thai]

Dear Professor Greene and Nlogit users,

Thank you so much for answering my questions so far. I deeply appreciate
your help!

I have another issue that I would be grateful if you could help me with.

I have a stated preference data with 6 alternatives: HOS, COM, PRI, IND,
GOV, NON. When I changed the reference alternative (from HOS to COM), while
the ASCs results from CL models are consistent, the results from the mixed
logit models are not consistent in terms of signs. Please see the results
below, the ASC of HOS (compared to COM) and the ASC of COM (compared to
HOS) are both negative. I would expect the signs of ASCs should be opposite
given the two ASCs of HOS and COM are statistically significant in mixed
logit. The log likelihood values of two MXL models are different too.

I  even tried ASCs only models and models with ASCs specified as dummy or
effect coded variables. The ASC results of the mixed logit models are not
consistent when I changed the reference alternative.

I would be truly grateful if you could suggest what has gone wrong.

My coding for the mixed logit models is below. I have the seed number for
them.

*HOS as the reference alternative*
*CL *model: logL = -4001.83388
     COM|    -.10806         .15871     -.68  .4959     -.41913    .20300
     PRI|     .40590***      .14002     2.90  .0037      .13146    .68034
     IND|   -1.11685***      .15286    -7.31  .0000    -1.41645   -.81725
     GOV|    -.07470         .14043     -.53  .5948     -.34993    .20053
     NON|    -.26651*        .14521    -1.84  .0665     -.55112    .01810

*Mixed logit *model: logL = -3949.60375
        |Random parameters in utility
functions..............................
     COM|    -.62569***      .23365    -2.68  .0074    -1.08362   -.16775
     PRI|     .23861         .17884     1.33  .1821     -.11191    .58914
     IND|   -1.75203***      .22127    -7.92  .0000    -2.18572  -1.31835
     GOV|    -.22635         .19028    -1.19  .2342     -.59930    .14660
     NON|    -.86462***      .23917    -3.62  .0003    -1.33338   -.39585

*COM as the reference alternative*:
*CL* model: logL = -4001.83388
     HOS|     .10806         .15871      .68  .4959     -.20300    .41913
     PRI|     .51396***      .15789     3.26  .0011      .20451    .82342
     IND|   -1.00878***      .18703    -5.39  .0000    -1.37535   -.64222
     GOV|     .03337         .16450      .20  .8392     -.28904    .35577
     NON|    -.15844         .17278     -.92  .3591     -.49709    .18020
<Thao.T.Thai at monash.edu>

*Mixel logit* model: logL = -3945.57277

        |Random parameters in utility
functions..............................
     HOS|    -.46844**       .22207    -2.11  .0349     -.90369   -.03319
     PRI|     .29024         .19299     1.50  .1326     -.08801    .66849
     IND|   -1.60623***      .25165    -6.38  .0000    -2.09946  -1.11301
     GOV|    -.19455         .20915     -.93  .3523     -.60448    .21538
     NON|    -.75668***      .25507    -2.97  .0030    -1.25662   -.25675

*Coding *

CALC ; Ran(12345) $
sample;all$
reject;sprp=1$
Nlogit
;lhs = cho, cset, alti
;choices = H, C, P, I, G, N
;checkdata
;Export output
;export=both
;rpl
;fcn= com(n), pri(n), ind(n), gov(n), non(n)
;pts=200
;pds=set
;halton
;maxit=300
;model:
U(H) =  rlh1    * RL_H1  + rlh2  * RL_H2
     + fl      * FL_H
     + crh1    * CR_H1  + cr2  * CR_H2
     + loh1    * LO_H1
     + sa      * SA_H
/

U(C) =  com + rlc1    * RL_C1  + rlc2  * RL_C2
     + fl      * FL_C
     + crc1    * CR_C1  + cr2  * CR_C2
     + loc1    * LO_C1  + loc2  * LO_C2
     + sa     * SA_C

/
U(P) = pri
     + rlp1    * RL_P1
     + fl      * FL_P
     + crp1    * CR_P1  + cr2  * CR_P2
     + lop1    * LO_P1  + lop2  * LO_P2
     + sa      * SA_P

/
U(I) = ind
     + rli1    * RL_I1  + rli2  * RL_I2
     + fl      * FL_I
     + cri1    * CR_I1
     + loi1    * LO_I1
     + sa      * SA_I

/
U(G) = gov
     + rlg1    * RL_G1
     + fl      * FL_G
     + crg1    * CR_G1
     + log1    * LO_G1
     + sa      * SA_G

/
U(N) = non
     + rln1    * RL_N1
     + fl      * FL_N
     + crn1    * CR_N1
     + lon1    * LO_N1  + lon2   * LO_N2
     + sa      * SA_N

$

Thank you so much!
Best regards,
Thao