Pivot Design

[Andy20160616]

Hi dear Ngene team,
Now I am trying to create the stated choice design using the pivot design, and some questions confuse me:

Background: there are 12 attributes, and I want to use three of them to do the pivot design, remains will do the efficient design. These three attributes are: distance to the bus station, metro station and shopping mall respectively. Here are my questions:

1) From the former questionnaire (RP section), I have already known the status of every respondent. I would like to set 4 groups for the design: small, medium1, medium2, large. However, how to select the right group for each respondent? (I.e. A respondent could select attribute1 as the small group; select attrbute2 as medium1 group and select attribute3 as the large group considering their status quo and the reference value.)

2) Pivot is for the reference value or the status value of respondent?
For example, this is one line of my syntax: ……+b HOMEaccmetro [-0.10] * HOMEaccmetro .ref [0.5] +……
The status value of my respondent is 0.6 and the group reference is 0.5. My question is , when I do the pivot, the attribute value of the new alternative is based on 0.5 or 0.6?

3) Does heterogeneous pivot design better than homogeneous pivot design for the modeling part? For I could these pivot design both, I am a little confused about which one is better for my design?

4) Lastly, could Pivot design add constrains like the orthogonal design or efficient design?

Thanks in advance, I will be very appreciated for your suggestions and advices.
Best, Jia

[Michiel Bliemer]

Apologies for the late reply, sometimes the system does not notify us of new posts, and some of us have been away.

1) If attributes appear in variations and cannot be classified into a small number of groups, you may have to extend the number of groups.
For example, instead of small and large you may need combinations [small1, small2], [small1, large2], [small2, large2], and [large2, large2] to represent different categories for different attributes. This of course leads to a much larger number of designs. For a specific study I created more than 1,000 pivot designs for this reason.In Ngene you therefore have to specify different models for each combination, i.e. ;model(small1large2) etc.

2) In the Ngene syntax you use the group average, for example for the group with 0.4<small1<0.8 you use value 0.6. Then in the survey instrument you use the pivot design optimised for 0.6 but apply the levels pivoted around the respondent value, namely 0.5.

3) A homogeneous design applies the same pivots to all respondents and groups, while a heterogeneous design applies different pivots to different groups (and hence creates multiple pivot designs). Best would be to generate heterogeneous designs for each combination mentioned under 1). A homogeneous design is simpler to implement, but will be less efficient. So it depends on whether you are able to create a library of designs and use the right design based on the attribute levels the respondent provides, or whether you only want to implement a single design in the survey.

Michiel