choice-metrics.com

Dear moderators of the choice metrics forum,

I am currently designing a choice experiment, and I was wondering if there is a possibility to test (in NGENE or elsewhere) whether the generated design functions and if my models can be estimated using the design. Perhaps there is an option to generate random data to do some analyses with?

I was also wondering if and how in NGENE you can estimate the required (or recommended) minimum number of respondents for the generated design? Is there a tool to calculate this?

Many thanks in advance,
Sem

Hi Sem,

To answer your first question, YES, you can evaluate an existing design. You will need to put your design in an Excel file or CSV file in a specific format (see the manual) and then use something like the following in the Ngene syntax:

;eval = filename.xls

Ngene will then produce efficiency measures such as the D-error. If the D-error is finite, then you can estimate your model. If your D-error is NaN (not a number), Inf (infinite), or extremely large, then there is an issue with the design and not all parameters can be estimated. There is no need to simulate random choice data, the D-error computes the covariance matrix, which only requires the attribute levels in the design and does not rely on choice data. The covariance matrix is all you need to assess whether you can use the design for model estimation.

To answer your second question, YES, if you provide informative priors (e.g. from a pilot study) in your Ngene syntax then Ngene will automatically compute and report S-estimates (sample sizes estimates) for each parameter. It will tell you how many respondents you need at minimum for each parameter to be statistically significant in model estimation.

Let me know if you need further assistance in doing this in Ngene.

Michiel

Dear Michiel,

Many thanks for your response! This is very helpful and I will try to apply it in NGENE.

And sorry for the late reply, I was on holidays.

Best regards,
Sem