I have a student version of the survey engine for doing my master’s thesis, so I can only use the orthogonal design for a choice experiment. When I tested my survey with my friends, they got identical alternatives and dominant ones. Those will not be good while modelling the data. I couldn’t edit the design matrix as well. Is there any way to remove those issues? I am attaching a screenshot for reference.
Greeshma, This entirely expected with an orthgonal design. Orthogonal designs have no prior knowledge to avoid dominance. All they do is prevent co-linearity, so models can converge and identify all effects. In addition as you have generic alternatives and attributes with fewer levels than alternatives, some profiles will naturally repeat and importantly, orthogonality only exits within alternatives, not across them
So here’s the reality.
For most practical applications it doesn’t matter.
There will be enough colour in the data and linear independence to guarantee models will converge and part-worths are captured. Methodological deficiencies will end up in the error term. And generally these models will be useful for most commercial applications. This is the dirty secret about experimental designs. Even a random design will also probably be fine, models from real world data are just that. They are wrought from data with dominance problems and repeats and provided there is limited co-linearity, perform just fine.
It is not enough to perform some test and be concerned. If you are, then you need to read the published literature, understand it and then apply a design appropriate for your study’s aims and the properties you desire in your final model.
I suggest you start with Street & Burgess to understand orthogonal designs. Then read Bliemer & Rose to get a grip on D-Efficient designs which control for dominance. Perhaps then look at Bayesean designs as well.
Once that is done there are many options. You can use published designs or generate your own or use Ngene to help you get there. But I would strongly caution against generating designs without understanding their application if academic-review is of a concern.
Good Luck!