choice-metrics.com

Hi! I am learning choice experiment and doing my first one this semester. This platform has been really helpful. I have a doubt regarding my first design (more of a confirmation if I am doing things right). If anyone can help, that would be greatly appreciated.

I have three alternatives with five attributes asking respondents to choose between policy A(alternative 1), policy B (alternative 2), and without policy (alternative 3). Policies show what happens to the attributes if a particular policy is chosen and alternative 3 shows what happens without a policy.

The 5 attributes are black, red, blue, green, and cost.
Levels of attributes are as follows: black[0,1,2,4,6], red[15,25,34,40,50], blue[10,25,35,50,60,70], green[0,1,10,25,50], and Cost [12,36,72,120,240] .
The levels for the 3rd alternative (without policy) are fixed and don't change across choice situations. The third alternative has the same attributes as alternatives 1 and 2,

Is the following design correct for this situation?
I renamed the attributes for alternative 3 in the design below so that new levels are considered, however, all attributes are same across all alternatives. Only the levels differ.
Thank you in advance! Any help is appreciated.

Design
;alts = alt1, alt2, alt3
;rows = 24
; eff = (mnl, d, mean)
;block = 4
;model:

U(alt1) = b1[(n,-22, 8.80)] * black[0,1,2,4,6] + b2[(n, -5.49, 2.19)] * red[15,25,34,40,50] + b3[(n, -6, 1.71)] * blue[10,25,35,50,60,70] + b4[(n, 3.84, 2.69)] * green[0,1,10,25,50] + b5[ (n, -1, 0.3)] * Cost [12,36,72,120,240] /

U(alt2) = b1 * black + b2 * red+ b3* blue+ b4 * green + b5 * Cost /

U(alt2) = b1 * black1[6] + b2 * red[50]+ b3* blue[70]+ b4 * green[50] + b5 * Cost[0]
$

There are a few issues:
- Your priors are entirely wrong, they do not make any sense, they are far too large. Your priors should preferably come from a pilot study and must be consistent with the units in your attribute levels.
- You twice specified the alt2 utility function
- You need to give attributes a different name if they have different levels, see my proposed status quo alternative
- I would add a constant for the sq alternative
- I would block the design since 24 rows is too many for a single respondent

Code: Select all

Design
;alts = alt1*, alt2*, sq
;rows = 24
;block = 3
;eff = (mnl, d, mean)
;block = 4
;model:

U(alt1) = b1[(n,-0.22, 0.08)]      * black[0,1,2,4,6]
        + b2[(n, -0.0549, 0.0219)] * red[15,25,34,40,50]
        + b3[(n, -0.06, 0.0171)]   * blue[10,25,35,50,60,70]
        + b4[(n, 0.0384, 0.0269)]  * green[0,1,10,25,50]
        + b5[(n, -0.1, 0.03)]      * Cost [12,36,72,120,240]
        /
U(alt2) = b1 * black + b2 * red+ b3* blue+ b4 * green + b5 * Cost
        /
U(sq)  = b0[0] + b1 * black_sq[6] + b2 * red_sq[50]+ b3* blue_sq[70]+ b4 * green_sq[50] + b5 * Cost_sq[0]
$

Michiel

Thank you so much for your help in pointing out the errors and for providing a revised code.
I do not have a pilot study yet but we plan to have one soon.
Hopefully, I can get better priors through that.
Thank you, again!