ASC's in unlabeled choices

[miles]

Hi moderators,

I'm hoping to ask a question in relation to this post. We recently ran an unlabelled DCE study that had a "neither" option. The position of the neither option changed in each choice task (the position was predetermined for each choice task). We have the randomisation order as a separate variable that has been dummy coded - this being neith_1 (left), neith_2 (middle), and neith_3 (right).

We have used the following code to analyse the randomisation variable interaction with the asc of the neither option. The results are also presented below.

* We are hoping to ask for guidance if the nlogit syntax presented below is correct?
* If so, how should we interpret the results of the analysis?

Nlogit
;lhs = decision,cset,choice_s
;choices = 1,2,3
;checkdata
;model:
U(1) = mod_2*mod_2 + mod_3*mod_3 + mod_4*mod_4 + mod_5*mod_5 + tim_2*tim_2 + tim_3*tim_3 + con_2*con_2 + con_3*con_3 + int_2*int_2 /
U(2) = mod_2*mod_2 + mod_3*mod_3 + mod_4*mod_4 + mod_5*mod_5 + tim_2*tim_2 + tim_3*tim_3 + con_2*con_2 + con_3*con_3 + int_2*int_2 /
U(3) = asc + neith_1*neith_1 + neith_2*neith_2
$


| Standard Prob. 95% Confidence
DECISION| Coefficient Error z |z|>Z* Interval
--------+--------------------------------------------------------------------
MOD_2| 1.08453*** .10146 10.69 .0000 .88568 1.28339
MOD_3| .48504*** .07538 6.43 .0000 .33730 .63278
MOD_4| .82944*** .08082 10.26 .0000 .67103 .98785
MOD_5| 1.58897*** .11677 13.61 .0000 1.36010 1.81783
TIM_2| .54109*** .05997 9.02 .0000 .42356 .65862
TIM_3| .65376*** .06936 9.43 .0000 .51781 .78971
CON_2| .14013*** .05296 2.65 .0081 .03633 .24394
CON_3| .09698* .05103 1.90 .0574 -.00305 .19700
INT_2| .84318*** .05272 15.99 .0000 .73985 .94651
ASC| .16414 .14473 1.13 .2567 -.11952 .44780
NEITH_1| .37972*** .13346 2.85 .0044 .11815 .64128
NEITH_2| .49398*** .12300 4.02 .0001 .25290 .73505
--------+--------------------------------------------------------------------
Note: ***, **, * ==> Significance at 1%, 5%, 10% level.
-----------------------------------------------------------------------------

[Michiel Bliemer]

Your results state that Neither is more preferred if it appears on the Left or in the Middle, compared to being shown in the Right. Coefficients Neith_1 and Neith_2 do not seem to be statistically different from each other looking at their values and standard errors (but you can do a statistical test), so you could consider combining them into a single coefficient for Left/Middle.

Note that showing Neither in a different position also means that you are showing the other alternatives in a different position, which may also affect their constants. So you could include their positions also in their respective utility functions:

U(1) = alt1_1*alt1_1 + alt1_2*alt1_2 + mod_2*mod_2 + mod_3*mod_3 + mod_4*mod_4 + mod_5*mod_5 + tim_2*tim_2 + tim_3*tim_3 + con_2*con_2 + con_3*con_3 + int_2*int_2 /
U(2) = alt2_1*alt2_1 + alt2_2*alt2_2+ mod_2*mod_2 + mod_3*mod_3 + mod_4*mod_4 + mod_5*mod_5 + tim_2*tim_2 + tim_3*tim_3 + con_2*con_2 + con_3*con_3 + int_2*int_2 /
U(3) = asc + neith_1*neith_1 + neith_2*neith_2

Michiel