choice-metrics.com

Hi dear Ngene team,

I am trying to do a pivot design related to mode shift in Kuala Lumpur. We have 3 alternatives (drive alone, train and bus) and each with related attributes and levels. After reading Manual and some other papers from Ngene team, I ‘m still little (honestly more than little) confuse about design. I tried my best to figure it myself, even I am dreaming about pivot and Fisher matrix nowadays! BUT No way, I can’t do it myself. Please help me!
I am sharing my understanding of applying pivot design and I would be really thankful if you correct me.
1-Since, I have no idea about priors (except sign), so, I guess I need to have a pretest. For pretest, I am applying an efficient design with near to zero priors (I don’t mind about huge S estimate which software is giving). Here the question arises; whether I can use percentage again as levels or not? Reason of using percentage is because each respondent has different status quo.
2-Segments: This is the most confusing part, not in theory but in application! Do we need to define segments in advance, means even before pretest or after pretest? I would say from beginning we have to have some logic assumption about segments! Like travel time of respondents and each respondent would be belonging to one and only one segment, is it right?
3-Main design 1: If we suppose to determine segments in advance, or even after pretest, why coefficient of beta (b1 value page 166 of manual) should be the same for each small, medium and large model? Shouldn’t it different for each segment? Or I didn’t get the whole story of pivot design!! (Which is possible also)
4-Main design 2: If each respondent belongs to only one segment. In main survey ,e.g.,we suppose a respondent is belong to small segment; let me refer to page 166 again. If I put priors of my respondent and related levels in small segment (no question about that), what I suppose to put in medium and long model priors or their reference to get the final design? considering we are applying CAPI!

I am sorry, questions might look so silly! but I 'm also good in confusing myself. Thanks a lot.

Sara

1. Yes you can use percentages. The percentages apply to the attribute levels (reference), not to the parameter priors.

2. Yes you need to define segments in advance, each respondent only belongs to one segment.

3. That is a choice. In the example, we assume that the behaviour could be the same for small, medium, large, but we would like to show the respondent attribute levels that make sense. So a respondent in the large segment should not see a small value for example. But it is your choice to decide whether you think a different beta is appropriate (could very well be!) or that the behaviour is the same across all segments. Besides putting different betas for different segments, you can also create entirely different designs per segment. This is the difference between a homogeneous and a heterogeneous design. A homogeneous design is the same for all segments, but it uses pivots from different reference points, so the choice tasks are still different for each segment, although the underlying design is the same.

4. A respondent in the small segment, then only the beta in the small category applies. The beta in the medium and large segments should represent the behaviour of the respondents in these categories. I am not sure I completely understand this question.

Michiel

Hi Michiel,

Thank you. The confusing part was segment which I think I am clear about it now.

Sara