choice-metrics.com

Hi

I am doing a Bayesian efficient design based on a pilot survey. I am wondering whether I should increase rows for the Bayesian efficient design. I understand that increasing rows will significantly decrease D error and S estimate. However, increasing rows also means I have to add choice tasks in the following survey, which might not a good idea since I have already had 9 choice tasks in the pilot survey. Are there any guidance for how to do the trade-off?

Many thanks,
Shan

The type of design does not matter for the number of rows needed, only the model itself matters, i.e. the number of parameters to be estimated.

Note that having some variation in the data is good, therefore most people will increase their design and block it to create multiple versions of the survey. For example, instead of having a design of 9 rows, you can use a design of 18 rows and block it into two versions such that each respondent only faces 9 choice tasks. Such designs are sometimes referred to as "heterogeneous designs" where respondents to not all see the same choice tasks. For more information I refer to Sandor and Wedel (2005), "Heterogeneous Conjoint Choice Designs", Journal of Marketing Research, 42, 210-218.

Michiel