choice-metrics.com

Hi Experts,

I have a question about model analysis. My choice experiments consist of three alternatives, two are unlabeled (A and B) and one is respondents' current status. All attributes were numerical but coded as dummy when using Ngene, to account for non-linearity. However, when doing analysis, those attributes have to be considered as continuous variables as the levels of attributes in current status vary by individuals (for example, patients' current waiting time is different from each other).

I applied latent class conditional logit model and identified two classes. I would like to compare preference weights between those two classes. I think relative importance is a way to do so. I know how to calculate RI when attributes are categorical, but I am not sure how RI is calculated for continuous variables. To my understanding, the coefficients of continuous variable represent the unit change in utility. Suppose I have four continuous attributes with coefficients a1, a2, a3, a4. May I calculate RI for first attribute as: RI (1)=a1/(a1+a2+a3+a4) ?

No that is not correct because attributes may all have different units. You need to compute the relative importance by looking at the contribution to utility.

Suppose you have:

... + a1[0.1] * X1[1,2,3,4] + a2[-0.05] * X2[5,10,15] + ..

Then attribute X1 contributes 0.1 to 0.4 to utility, that is, a range of 0.3.
Attribute X2 contributes between -0.25 and -0.75 to utility, that is, a range of 0.5.
Therefore, the relative importance of X1 is 37.5% (namely, 0.3/0.8).

For more information about relative importance, please refer to book by Orme (2010) Getting started with conjoint analysis: strategies for product design and pricing research.

Note that the above computation of relative importance depends on the range of the levels of the attributes, so it is not entirely independent of the design of the experiment.

Michiel

Many thanks for the clear explanation, Michiel.