choice-metrics.com

Dear Professor,

Quick question regarding attribute and attribute levels.
Say for instance I have:
Att1 3 levels
Att2 3 levels
Att3 3 levels
Att4 2 levels

Could this cause any issues with my design or estimation? My concern rose mainly with having 2 levels.
Or does it matter if i were to have 3 levels for ALL attributes, statistically?

Thank you

Kind regards,
Maria

With respect to estimation, you can mix numbers of levels across different attributes, you should choose the number of levels based on what makes most sense. For qualitative attributes often the number of levels is given (e.g. brands). For quantitative attributes you can choose the number of levels (e.g. price) and I would recommend choosing at least 3 levels such that you are able to estimate nonlinear effects.

With respect to design, you can also mix numbers of levels across different attributes when you are generating an efficient design, while for an orthogonal design it may be more difficult (or impossible) to find a design if the number of levels greatly vary. Orthogonal designs have been widely used in the past and this has often restricted researchers to choose the number of levels, but with more state-of-the-art efficient designs this is no longer an issue and you can essentially choose any number of levels you like. The number of rows (choice tasks) in the design should preferably be divisible by 2 and 3 in your case in order to have attribute level balance across the design, but this is not strictly necessary.

Michiel