I want to investigate how people travel to various stations in the UK. I have found data on both passengers per year, and traffic by mode along the main roads leading to those stations, also per year.

With this data I should be able to get some estimate of how many passengers use each mode to get to the station eg)

change in passengers = A+ Bchange in cars + Cchange in bus +...

although this would give the amount of additional passengers resulting from additional traffic, not the amount of additional cars/buses per x new passengers.

Can anyone think of a functional form that answers my question with the available data? Many thanks.

  • $\begingroup$ Well the stations I'm looking at are parkway stations which tend to be accessed heavily by road. The DfT traffic count data splits traffic by type so the bus/coach travel is part of this. $\endgroup$ – Chris Jul 16 '15 at 14:53
  • $\begingroup$ Yes those are certainly obvious flaws in the approach but I guess I'm interested in a broad brush approach based on a large sample of stations. I have not seen any such direct observations except one small pilot survey. I would love any input into alternate approaches if you have experience though $\endgroup$ – Chris Jul 16 '15 at 15:06
  • $\begingroup$ I'm looking at the impact of location characteristics on method of travel to a station, specifically parkway vs city centre/suburban. Thanks for the ideas and effort, I've taken a look at the NTS but it doesn't seem to give the individual stations from the survey, will try with the ATOC $\endgroup$ – Chris Jul 16 '15 at 15:30
  • $\begingroup$ Let us continue this discussion in chat. $\endgroup$ – EnergyNumbers Jul 16 '15 at 15:39

Given that, as per your comments, you're "looking at the impact of location characteristics on method of travel to a station, specifically parkway vs city centre/suburban", one popular method is to estimate a multinomial logit model of modal choice for access to and egress from the rail station, where the probability of an individual trip $y_i$ having access mode $j$ is given by:

$$P(y_i = j) = \frac{e^{x_i\beta_j}}{\sum_{j\in J}e^{x_i\beta_j}}$$

where you'd estimate the $\beta_j$ coefficients from your observations: each $x_i$ representing some observed aspect of the level of service.

To look for some example estimations of such models, see work by some of the RAND Europe people, who will have done similar stuff: Charlene Rohr, Andrew Daly, James Fox are names to look out for, and other folks who work in this field of discrete choice modelling - Stephane Hess, maybe John Polak. There's probably documentation around calibration of the PLANET model and its relatives.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.