Reference alternative or current level alternative

[connie]

Hi Experts:

Hope you are well in such unsettling time.

I have tow designs for my study. The first one is about 'quasi' pivot. I created segment design for two groups, the employed and the unemployed. The attributes pivot around the reference levels. However, I did not use pivot design, as I have some conditions for certain attributes. Instead, I created traditional designs, with one alternative as reference alternative, fixed across choice sets. The idea is similar to pivot. Participants face three choices, altA, altB, and group average reference.

For the second design, I did not have reference alternative. The reference level is only used to identify the levels of attributes (attributes pivot around reference level). In this traditional design, I set an opt out. Opt out here means 'participants' current level alternative'. This information will be collected when participants answer the survey. In the data analysis, I will manually input current level alternatives for each participants. Participants face three choices: altA, altB and my current level.

I am struggling with which design I need to use. I think the first design is easy to implement, but it did not consider the individuals' actual level. If the reference level is quite close to individuals' actual level, then it may be good. While for the second one, I am not quite sure the potential issues. As each participant has different current level, how this actual choice will influence hypothetical scenarios? For example, income will be the price proxy in my study. The levels of this attribute are 1500, 2000, 3000. If a participant prefers his/her current level, whose levels of attributes may be quite different from alt A and alt B, and his/her the actual income is 5000, far higher than the designed attribute. I am not sure how this will influence the final analysis? Another possible issue I am thinking of is, if information on current levels is missing, for example, participants do not tell their income, then the whole choice sets can not be used. Any papers recommended?

Any suggestions about which design I may consider?

Thank you,

Connie

[Michiel Bliemer]

A reference alternative that is fixed across all participants is generally called a status quo alternatives.
An opt out alternative cannot have levels, so if you use reference levels of a participant it is a reference alternative.

The easiest is to create a library of designs around various reference levels, e.g. a design with a low reference level, medium reference level, high reference level. In the survey instrument, you essentially create different versions of the experiment.

For example, you create this design:

"low":
U(ref) = b1 * X1_ref[2] + b2 * X2_ref[20] /
U(alt1) = b1 * X1[1,2,3] + b2 * X2[10,20,30] /
U(alt2) = b1 * X1 + b2 * X2

"medium":
U(ref) = b1 * X1_ref[4] + b2 * X2_ref[40] /
U(alt1) = b1 * X1[2,4,6] + b2 * X2[20,40,60] /
U(alt2) = b1 * X1 + b2 * X2

etc.

When you conduct your survey, you can ask the respondent about the reference levels, and if the respondent provides levels around X1 = 4 and X2 = 40 then you present choice tasks of experiment "medium" where you show the actual reference levels in the reference alternatives.

In model estimation, you can pool all the data and estimate a single model for all participants.

Michiel