choice-metrics.com

Dear colleagues

Do you have any advice on what is the more important to randomize across respondents in DCE ?
- attribute ordering (but each respondant see the same order in every choice task to reduce cognitive burden)
- order of appearance of choice tasks
- order of appearance of alternatives (A,B) -some see B on the left and others on the right

I have checked papers and they mostly conclude that randomizing all these elements is important !
But randomizing everything is not feasible in my case (due to software limitation, since I have to code all the different survey version)
I already have a design with 3 blocks

What is your practice ?

Thanks
Marianne Lefebvre, Angers University

If you have labelled alternatives, you must randomise the order of the alternatives and add an order dummy variable to your utility functions for model estimation to account for left-to-right bias. If you have unlabelled alternatives, there is no need for randomisation (you simply add a constant to the utility function to account for left-to-right bias).

The attribute order is known to make a difference, e.g. price at the top or bottom may become more important than price in the middle. I generally do not randomise the order of the attributes but probably I should.

Randomising the order of choice tasks is often useful to ensure that learning and fatigue effects are spread across all choice tasks. If you have multiple blocks, then this already accounts for some of this randomisation. I usually have a relatively large design with many blocks, so I do not feel the need to randomise the order or the choice tasks, but if your design size is small then randomisation may be worth considering.

Michiel