choice-metrics.com

Hello everyone,

I'm new to Ngene, and I don't know much about the theory and algorithmes that it uses.

My question is whether it is important to specify the real values of the levels when defining an attribute, or if it is only important to distinguish the levels between them, e.g. by numbering them ?

So for exemple, for an attribute of travel time, should I write in the utility function
U(O1) = b1[<priors>] * time[5,15,30,60]
or
U(O1) = b1[<priors>] * time[1,2,3,4] ?

Thanks in advance for any insight you might provide,
Paul

Hi Paul,

When you are generating an orthogonal design, the actual levels do not matter as orthogonal designs only look at correlations across attributes. Often people use design coding, i.e. 0,1,2,3.

However, if you are generating an efficient design, priors for the parameters are important and therefore the attribute levels need to match the levels that you would use in model estimation. So you would use the actual travel time levels in your specification, e.g. 5, 10, 15, etc. For categorical variables with dummy or effects coding the actual levels do not matter, so you could use either b1.dummy[..|..] * time[1,2,0] or b1.dummy[..|..] * time[15,20,10], where the last level is the reference level. With dummy coding, all levels are translated into sequences of 0 and 1 either way.

Michiel

Hi Michiel,

I see, I hadn't considered it could depend on the type of design, many thanks for your answer !

Paul