As part of my research I am running into a small issue regarding stated choice experimental design and would appreciate your input on the matter. I noticed a similar forum post ("Second choice with some NEW attributes") although I am unsure if the same responses hold for my context.
Context:
Research is being done on transportation choices
Using Ngene to generate the design, have not run pilot test yet although priors are established in literature
Three alternatives: (alt1,alt2,alt3)
Three attributes, all continuous: (cost,time,distance)
One moderator: (air quality) [low,med,high] -> [0,1,2]
Problem:
If I want to measure the influence of the moderator on the beta of each attribute (i.e. does distance become more relevant?), should I create two models on Ngene (one with the moderator (four attributes) and one without (three attributes)) and make comparisons between the two or would it possible through one model?
Thanks for your time and any input is greatly appreciated.
For the analysis I would suggest to create interaction variables involving the potential moderator variable.
The four variable model would not tell you much about the moderating influence of this variable without interactions.
You could specify interactions already in Ngene, but I think it is not required (although more accurate), you could do that later in the analysis.
But I'm quite sure you need only one Ngene design. I would include only the product attributes.
This allows estimating a single model, and this kind of syntax can also be used to create a single design (there is trick to define dummy variables in Ngene while only specifying interaction terms, am happy to show you how). If you rewrite this utility function, you obtain:
In other words, the coefficient for distance is b1 if airquality is high, is b1 + b2 if airquality is low, and is b1 + b3 if airquality is medium. This way, you can analyse the impact of airquality on preferences for each attribute.