SP experiment to estimate the price elasticity only for trai

[Mathilde Ruyssen]

Hi everyone,

I would like to perform an SP experiment to estimate the price elasticity only for train passengers. Therefore, the respondents are only train passengers. I am not interested to know to which other mode they shift (I just want to know whether they shift to another mode or renounce to travel).

My question is: what would be the simplest (cheapest) SP experiment to perform?

- Is it possible to estimate the utility of train passengers if the SP questions ask their choice between: travelling by train, travelling by another mode of transport and not travelling?
In the SP questions, the attributes of the other modes would be shown but would be fixed (to their current value) and only the attributes of the train would change.
I will estimate a model with the utility of the train and two constant utilities for the options "travelling by another mode" and "no travel".
- If I am only interested in the impact of price change on demand (all else being equal), is it possible to let vary only the price or is it better to let vary at least price and travel time? Is it
possible to keep all the other attributes (frequency etc) to a fixed value, in the SP questions?
- One weakness of such a SP design would be that it gives the price-elasticity only with the current existing transport supply (for the alternative modes), no ?

Thanks in advance!

[Michiel Bliemer]

You will be able to estimate such a model with train, other mode, and no travel but it will likely have a lot of error variance since you are not providing much information to the respondents and therefore they will need to guess. For example, "other mode" could be car, bus, bicycle, etc., completely different alternatives. If one respondent has "car" in mind when evaluating the choice task, and another respondent has "bus" in mind, then it is unclear what the constant relates to for this alternative. You can estimate random parameters to account for such heterogeneity but preferably you remove such noise in the data. You could consider multiple alternative modes, namely "other public transport" and "car". Note that the utility for alternative "no travel" will need to be normalised to zero, you cannot estimate constants for all alternatives.

Further, information about each transport mode needs to be fairly complete, so for train you would expect access/egress time, in-vehicle travel time, waiting time, fare, number of transfers, etc. Leaving out attributes again means that respondents will be guessing what the attribute levels are when they evaluate the alternatives (for example, if "other mode" requires 2 transfers then it would be less attractive than if it requires 0 transfers), which again introduces a lot of noise in the data.

You can use fixed attribute levels for the "other mode" alternative, although I am not sure how generalisable your results will be. The other issue is deciding what attribute levels you choose for train. People with a high value of time would mainly focus on the travel time differences between the modes, while people with a low value of time would focus on the cost difference between modes, so it makes sense to vary both these attribute levels for train. You can use the same fixed values for access/agress, transfers, etc across the two mode alternatives.

Michiel