Questions on efficient design

[JaimieleeM]

Hi,  
I am looking to run a DCE exploring the appeal program components to increase physical activity and those same components to reduce sitting time. Being new to this approach I have quite a few questions to please bear with me! 
 
The attributes have been informed by a previous qualitative study. The levels were originally not balanced however I have now done so giving 5 attributes with 4 levels each. I would also like to have an opt out option. I am having some difficulty assigning priors given the qual study origin. I have read is best to assign low priors if unsure to at least indicate direction however am still having some difficulty with this.  
 
Also ideally I would like to have questions on these program attributes being considered for increasing the individual’s physical activity levels and then consider the same components for decreasing sitting time. So say 10 questions considering these attributes for physical activity and 10 for sitting time. Will this complicate my design substantially? Would you recommend conducting them as separate experiments? I was thinking perhaps the separate blocks from each target behaviour could be combined into the one survey however unsure if this is possible. 
 
My current syntax is as follows: 
 
design 
;alts= programA*, programB*, programC*, optout 
;rows= 30 
;block= 3 
;eff= (mnl,d) 
;alg= mfederov(candidates=1000) 
;model: 
U(programA)= blength[0.01]*length[3,2,1,0]+ baccess[0.01]*access[3,2,1,0] + bmagnitude[0.01]*magnitude[3,2,1,0] + bfunding[0.01]*funding[3,2,1,0] +bform[0.01]*form[3,2,1,0]/  
U(programB)= blength*length +baccess*access +bmagnitude*magnitude +bfunding*funding + bform*form/ 
U(programC)= blength*length +baccess*access +bmagnitude*magnitude +bfunding*funding + bform*form/ 
U(optout)= Noprogram $

[Michiel Bliemer]

It looks like you have qualitative variables (otherwise I am not sure why you are using levels 0,1,2,3). You need to use dummy or effects coding for qualitative variables, e.g. baccess.dummy[..|..]. If they are continuous variables, you need to use the levels that you will use in estimation, e.g. if funding is in dollars then you can use 10, 20, 30 for $10, $20 and $30.

If you would like to test for differences in scenarios, then you can add a scenario variable. If physical activity has two levels, then you can create an attribute activity[0,1] and then either add as a main effect (in all programs or in the optout) or as an interaction effect (funding*activity). This would allow you to pool the data and estimate a single model in which you can explore the differences between physical activity and sitting. You can give respondents 5 or 10 questions of each.

So my advice would be to first carefully think about the model you will be estimating, including dummy coding, interaction effects, etc., and once you know what the utility functions will look like you can enter them in the Ngene syntax. Most of the work always goes into thinking about the exact model that you want to estimate to answer your research questions.

Michiel