ApolloChoiceModelling forum

Hi all,

I recently estimated a mode choice model based on RP and SP data. Some alternatives are only available in the SP data and hence, there are parameters (e.g., travel time parameters for these modes) which are only estimated based on the SP data. The mu of the SP data is around 0.5, whereas the mu of the RP data is fixed to 1. However, it seems to me that the estimates of the parameters which are only estimated based on the SP data are not scaled in the results file, since the VoT for those modes are around 1/mu as high as for modes for which the parameters are estimated on both RP and SP data.

So, can the parameters be used as they are shown in the estimates file or have the be multiplied with mu before using them in a prediction model?

Thanks!

Best,
Michael

Michael

if you set mu=1 for RP, then all the betas are at the RP scale, even if they are only used in the SP utilities, as long as the entire utility function is multiplied by mu.

Can you share your utility files and results?

Stephane

Hi Stephane,

unfortunately I cannot share any file since this work is confidential and not published yet. However, I use seperate models or each data source and combine them with apollo_combineModels. For the mu, I have
applied to the cnl_settings of each model and of course the mu_rp set to 1.

So does that mean that for prediction, the parameters can be used as estimated in the final mode choice model?

Best
Michael

Michael

yes, but in prediction, you of course need to decide which scale to use. The natural approach would be to use the RP scale

Stephane

Hi Stephane,

thank you for the quick reply. As I understood it, it is not common to use different scales for different alternatives in prediction, right?

Best
Michael

no, that wouldn't make much sense