ApolloChoiceModelling forum

Hi! I would like to ask if someone has developed a choice model including spatial autocorrelation and if so, could you share your code. Thank you!

I have not done this, but am also very interested in knowing how it might be done.

Hi

you would need to make a decision on what model to use for this. Options could for example include a nesting structure that groups together alternatives that are closer to each other, or you could use error components and make the correlation a function of distance between the alternatives. Both approaches could adapt methods from departure time work, such as OGEV for example

Stephane

There are (at least) two different ways in which spatial correlation would show up in choice models.

1. Spatially correlated error between alternatives. This happens when there is some spatial attribute of the choice alternative, like in a destination choice model. That would seem to be handled with the GEV models that you suggest here, and that I have seen in Sener et al.http://citeseerx.ist.psu.edu/viewdoc/do ... 1&type=pdf and other places.

2. Spatially correlated utility between choice makers. This happens when the choice makers are located spatially, and the choices of one person affect the choices of others. Bhat https://ctr.utexas.edu/wp-content/uploads/DSTOP114.pdf appears to deal with this kind of thing.

I can see very readily how type 1. could be handled in apollo. Can type 2 be handled also? Is this what you mean by "you could use error components and make the correlation a function of distance between the alternatives"?

Hi

On point 1, yes, that's what I had in mind.

On point 2, you would have to think about coding this manually somehow, and using Apollo for estimation

Stephane