32 views

Skip to first unread message

Oct 13, 2021, 4:52:27 AMOct 13

to Biogeme

Dear Michel,

I just followed the DCM course on EDX and also read a book on it. It was said that the estimated parameters cannot be compared directly to check the importance but I do not know why. Let's assume that there are two attributes with the same unit like Travel Time and Waiting Time. If the estimated parameter of Travel Time is bigger than Waiting Time, doesn't that mean that Travel Time is more important than Waiting Time? If not, why?

I also followed a thread in this forum about choice-based conjoint analysis and parthworth utilities. As far as I understood, the idea is to include both the estimated parameters and the attribute levels to calculate the relative importance. I would like to know why attribute levels should be considered when calculating the attribute relative importance?

Best,

Sara.

Oct 13, 2021, 6:17:59 AMOct 13

to sara.all...@gmail.com, Bierlaire Michel, Biogeme

On 12 Oct 2021, at 19:32, Sara Allison <sara.all...@gmail.com> wrote:

Dear Michel,

I just followed the DCM course on EDX and also read a book on it. It was said that the estimated parameters cannot be compared directly to check the importance but I do not know why.

Because their value depends on the normalization of the scale of the error term. What is actually estimated is mu * beta. Both are confounded.

Let's assume that there are two attributes with the same unit like Travel Time and Waiting Time. If the estimated parameter of Travel Time is bigger than Waiting Time, doesn't that mean that Travel Time is more important than Waiting Time?

In that case, yes, because they are associated with the same scale.

In general, it is better to compare elasticities, or willingness to pay.

If not, why?

I also followed a thread in this forum about choice-based conjoint analysis and parthworth utilities. As far as I understood, the idea is to include both the estimated parameters and the attribute levels to calculate the relative importance.

Conjoint analysis is when you have several variables in the utility function. Parthworth analysis is about the relative importance of each attribute in the utility function.

I would like to know why attribute levels should be considered when calculating the attribute relative importance?

Best,Sara.--

You received this message because you are subscribed to the Google Groups "Biogeme" group.

To unsubscribe from this group and stop receiving emails from it, send an email to biogeme+u...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/biogeme/ea65fa84-d282-41b2-9bef-a9ec67993d88n%40googlegroups.com.

Oct 13, 2021, 4:09:37 PMOct 13

to Biogeme

Thanks for the reply, Michel. There is a special case here. I have two monetary attributes with the same unit including parking cost and toll cost. The estimated coefficient of parking cost is bigger than the coefficient of toll cost. According to your response, we can say parking cost is more importance than toll cost given that they have the same unit/scale , right?

The thing is that when I calculate the partworth utilities and the relative importance, toll cost becomes more important than parking cost probably due to the higher attribute level range of toll cost.

I wonder what I should conclude here? One method suggests that parking cost is more important and the other one results in an inverse outcome. I want to generalise the results and make a conclusion. Which conclusion/method is more reliable or generalisable in this case?

Thanks for your help.

Best,

Sara.

Oct 14, 2021, 2:42:04 AMOct 14

to sara.all...@gmail.com, Bierlaire Michel, Biogeme

On 13 Oct 2021, at 21:59, Sara Allison <sara.all...@gmail.com> wrote:

Thanks for the reply, Michel. There is a special case here. I have two monetary attributes with the same unit including parking cost and toll cost. The estimated coefficient of parking cost is bigger than the coefficient of toll cost. According to your response, we can say parking cost is more importance than toll cost given that they have the same unit/scale , right?

If your variables are continuous, yes. If your variables represent levels, as i appears from your next comment, it is not the case.

The thing is that when I calculate the partworth utilities and the relative importance, toll cost becomes more important than parking cost probably due to the higher attribute level range of toll cost.

I wonder what I should conclude here? One method suggests that parking cost is more important and the other one results in an inverse outcome. I want to generalise the results and make a conclusion. Which conclusion/method is more reliable or generalisable in this case?

I usually calculate arc elasticities. They are unitless, and are always interpretable and comparable.

To see how to calculate elasticities with Biogeme, see https://transp-or.epfl.ch/documents/technicalReports/Bier18a.pdf

To view this discussion on the web visit https://groups.google.com/d/msgid/biogeme/3e7144b6-eaaa-483b-a11c-6a583d9c092cn%40googlegroups.com.

Reply all

Reply to author

Forward

0 new messages

Search

Clear search

Close search

Google apps

Main menu