choice-metrics.com

Can we state anything about the relative importance of attributes (i.e. which is most important of attribute X and attribute XX)?
Or do we need a continuous variable to do this?

What does it add to results (or possible results) to include a continuous variable?

The importance of attributes is entered through the parameter priors in combination with the attribute levels. In order words, your priors need to come from actual expected behaviour (e.g. from a pilot study).

For example, if
U(alt1) = b1[0.2] * X1[1,2,3] + b2[0.1] * X2[10,15,20],

then X2 is more important than X1, since the average contribution of X1 to utility is 0.2*2 = 0.4, while the average contribution of X2 to utility is 0.1*15 = 1.5.

I am not sure I understand why it is relevant that a variable is continuous or not.

I quote from a "How to DCE" for the World Bank and WHO

To estimate trade-offs, a continuous attribute must be included in the DCE. Within the job choice literature, this continuous variable is commonly salary. Inclusion of this attribute allows estimation of how improvements in aspects of a job, such as housing and education opportunities, can compensate for lower wages - that is, how much salary an individual would be willing to give up for improvements in others attributes of a job

DCE
A. Can only say what is most preferred within an attribute (Gender: Man VS Woman).
B. Can find what attributes are most important (Gender more important than Religion with ... much percent..)

Which is right without continuous?
Does a continuous only give WTP sort of ?

It is common to have a price (or salary) variable in the design, and price can be modelled as a continuous variable. This allows the calculation of a willingness to pay (WTP). But note that it is not necessary to have all other variables also continuous, one can use discrete variables using dummy or effects coding and still calculate WTP values.

Note that gender is not an attribute of an alternative, but typically rather a socio-demographic that does not enter the design. But yes, when estimating the model, one can indeed say whether religion or gender is more important in the choice since these will be modelled with dummy or effects coding and one can simply look at the parameter values.

Note that these questions are not related to Ngene and should be posted in the other forum.