Efficient design, pooled data, evaluation of split designs
Posted: Tue Nov 17, 2009 9:37 amHello,
I am using Ngene to design choice experiments for a transport application. I have two groups of public transport alternatives and within each group "first class" and "standard class" alternatives, which share some attributes but have others that differ. I have set this up within each group as per 8.5 ("Designs within designs") in the manual. I also have had to take into account trip length, and have ended up with 5 separate designs for different trip length bands. I was interested in using Ngene's "pivot designs" functionality for this, but given my 4 alternatives it would only let me implement the "designs within designs" or the "pivot designs", but not both.
My question arises from the fact that I want to pool all the data together at the analysis stage, and am feeling slightly uneasy about this. I used the same priors for each of the 5 distance bands, so I think it should work, but it would be nice if there was a way of testing this or taking into account that it is going to be used like this, in advance. Is there? From my reading of the manual the "model averaging" functionality in Ngene lets the user test out different model specs with one experimental design, but not the other way round: in this case what interests me is using 5 different experimental designs with one underlying model to be estimated.
That brings me onto my second question. I was wondering if the "eval" feature could be used to evaluate the full blocked design with the 5 distance band-related designs, as a way of checking that the full SP experiment makes sense. The manual does not go into much detail of how to do this. Can you specify priors when you use "eval"? If I specify "block" in the last column, will Ngene take this into account when it evaluates the design?
Any help or advice would be very much appreciated.
Best regards
Alex Mitrani
I am using Ngene to design choice experiments for a transport application. I have two groups of public transport alternatives and within each group "first class" and "standard class" alternatives, which share some attributes but have others that differ. I have set this up within each group as per 8.5 ("Designs within designs") in the manual. I also have had to take into account trip length, and have ended up with 5 separate designs for different trip length bands. I was interested in using Ngene's "pivot designs" functionality for this, but given my 4 alternatives it would only let me implement the "designs within designs" or the "pivot designs", but not both.
My question arises from the fact that I want to pool all the data together at the analysis stage, and am feeling slightly uneasy about this. I used the same priors for each of the 5 distance bands, so I think it should work, but it would be nice if there was a way of testing this or taking into account that it is going to be used like this, in advance. Is there? From my reading of the manual the "model averaging" functionality in Ngene lets the user test out different model specs with one experimental design, but not the other way round: in this case what interests me is using 5 different experimental designs with one underlying model to be estimated.
That brings me onto my second question. I was wondering if the "eval" feature could be used to evaluate the full blocked design with the 5 distance band-related designs, as a way of checking that the full SP experiment makes sense. The manual does not go into much detail of how to do this. Can you specify priors when you use "eval"? If I specify "block" in the last column, will Ngene take this into account when it evaluates the design?
Any help or advice would be very much appreciated.
Best regards
Alex Mitrani