Web survey and efficient design

[Christoph]

Dear all,
We are currently preparing a web-based SP survey (using lime survey) on mode choice in freight transport. We have three requirements: (1) We would like to use an efficient design due to the small sample size. (2) Since there are so many different options in terms of commodities, distances etc. that impact freight mode choice, we do not provide uniform SP experiments, but like to pivot the alternatives from reference transport orders of the respondents (-> each respondent has its own reference transport order). (3) From our point of view it is important to finish the interview without interruption in order to decrease the drop-out rate.
Thus we need a solution to “determine an efficient design on the fly (yields a separate design for each respondent)” during the interview as mentioned on page 158 of the manual. Has anybody experience with linking NGene and a web-based survey? How can we automatically embed NGene into a web-SP from a technical point of view?
Thanks a lot in advance for all answers
Christoph

[Andrew Collins]

Hi Christoph

This is an interesting technical problem. There are two key issues - specifying the design, and extracting the resulting experimental design.

First specifying the design. I'm not familiar with the capabilities of lime survey. Most internet surveys I have been involved with have been coded in PHP, and in this case it would be trivial to generate Ngene syntax based on the responses provided. This could be considerably harder with a "packaged" survey solution, although some solutions like Qualtrics allow Javascript scripting, and will expose the responses.

Extracting the experimental design is harder. An instance of Ngene would have to be executed, the syntax run, and the design levels extracted. I can't see how this could be implemented as Ngene currently stands. The most robust way would be to have the Ngene functionality in a dynamically linked library (DLL) file, and then access this through a C# program or wrapper. A little while back I worked on a proof of concept of this with Christian Schlereth and his DISE survey system (http://www.dise-online.com/o/Survey.aspx?id=3634&pos=0&user=000093-cdbc4cad-d10c-455f-8d69-a3c2f8a0d476), but I'm not aware how far this was ultimately taken. You could contact him if you would contemplate using that survey system.

There is an easier approach that might be feasible, depending on how many survey variations you could have. Generate all possible design specifications and associated designs beforehand, then retrieve the correct design on the fly. If there are too many combinations, sample various reference points (commodities, distances etc), and chose the closest one on the fly.

Basically, it is not an easy thing to implement, but perhaps the above suggestions give you some ideas.

Andrew

[Christoph]

Hi Andrew,
Thanks for your prompt answer. We will discuss your remarks in our team. I am sure we can handle the problem described in the first paragraph, but unfortunately this is only the first part of the story
Maybe we have to switch to a two-stage survey.
Thanks a lot
Christoph