choice-metrics.com

Sorry if my terminology isn't quite correct, I haven't done a lot of this.

I've run into an issue with simultaneous orthogonal designs both in my code and the Demonstration code. In the syntax file 6.2.3 Orthogonal designs.ngs the design ends up with 8 choice situations, but the first one has the exact same levels for both alternatives... all zeros.

I ran very similar syntax for a comparison of 2 alternatives with 7 parameters/attributes(?) each with 4 levels, and the first choice was all zeros between the two alternatives as well.

Is this a bug?

Thanks!

No this is not a bug, this is a fine design for some labelled experiments. In the Ngene manual we discuss simultaneous orthogonal designs that are suitable for labelled experiments, and sequential orthogonal designs that are suitable for unlabelled experiments.

For example, consider two alternatives for travelling to work: bicycle and bus.

Bicycle attributes: distance (1km, 2km, 3km)
Bus attributes: fare ($1, $2, $3), travel time (10 min, 15 min, 20 min), transfers (0,1)

If a simultaneous orthogonal design states for the first choice task: (0,0,0,0) for the 4 attributes across the two alternatives, this means:

Bicycle: 1km
Bus: $1, 10 min, no transfer

This is perfectly fine. But clearly the choice task would not make sense for an unlabelled choice task, for example with 2 bus alternatives, because then (0,0,0,0,0,0) would mean:

Bus: $1, 10 min, no transfer
Bus: $1, 10 min, no transfer

In other words, two identical alternatives.

Ngene as ;orth = sim, seq, and seq2 as possibilities, and the analyst should choose the one that is most suitable. Maybe seq or seq2 is the one you are looking for.

Michiel

Okay, that makes more sense. Thank you. I'll take another look at the manual for the labelled and unlabelled orthogonal experiments.

I don't know enough about labelled and unlabelled, but I know there are some ways to alert Ngene that alternatives are labelled, no? Is the purpose of that to avoid situations like this?

Thanks.

An experiment is labelled if the alternatives have a meaning.

Some examples:
{Dell, Apple, Lenovo} are labelled alternatives, {laptop A, laptop B, laptop C} are unlabelled
{Bus, Car, Train} are labelled alternatives, {Mode 1, Mode 2, Mode 3} are unlabelled

Labelled alternatives can have a constant in their utility function, can have alternative-specific parameters, and can have different attributes per alternative. In contrast, in unlabelled experiments all alternatives have exactly the same attributes and generic parameters.

Depending on your research questions, you choose which type of experiment is most suitable. For willingness to pay studies, an unlabelled experiment suffices, while for demand and market share forecasting you want to use a labelled experiment.

Michiel