choice-metrics.com

Dear all,

we are conducting a DCE in partnership with a survey institute. They are using the "idefix" R-package. We asked them to generate 60 choice sets and 10 blocks (6 choice sets each). They asked us whether they should just randomly fill the blocks with the choice sets generated by the package.

This brings me to my question: When using blocks, are the choice sets randomly distributed to the blocks or is there a machanism behind it?

In addition, is there an R-package available that allows for a D-efficient design with blocking so we can maybe suggest it to the survey institute?

Thanks in advance
Tom

Blocking is a term that originates from orthogonal designs, and for orthogonal designs it is possible to generate blocks that are perfectly attribute level balanced. Perfect attribute level balance within blocks is not possible with efficient designs, but Ngene applies a special optimisation process after design generation that assigns choice tasks to blocks such that attribute level balance is as good as possible (but will never be perfect). I am not aware of any other software, or R packages, doing this.

Most survey platforms like SurveyEngine and Qualtrics can randomly allocate choice tasks from the design to respondents, without the need for blocks. In most cases this is fine, unless you want to be sure that all respondents in the survey see a good coverage of each attribute level (which is desirable but not required).

If you really want to ensure a reasonable level of attribute level balance within a design, Ngene can read in a design that was generated externally and then create a blocking column for it. If you do not have Ngene, then you could try to manually assign choice tasks to blocks such that attribute level balance is reasonably high, or perhaps use Solver in Excel to do this for you by minimising imbalance, see for example Equation (8.1) on page 154 of the Ngene manual, which you can download from the ChoiceMetrics website.

Michiel