They “already at home” compare various combinations of lines and modes, with travel time components, prices, departure and arrival time
Travellers minimize generalised cost (G) according to Sampers Vips and Visum. But G varies between individuals in different ways.
Thus the models deal with different types of random variation
In Sampers variation wrt taste, measurement errors etc.
In Vips and Visum variation wrt difference between ideal departure or arrival time and actual times
Sampers and Visum generate demand between centroids
Vips does not generate demand
Mapping of routes and modes in the models
Sampers deals with “main modes”, while Vips takes into account all combinations of lines and modes
For allocating demand between main modes Sampers uses the average ride time and average price of the lines in each mode
Vips specifies separate prices and ride times for lines, implying that the price and ride time for the whole journey, O-D, is the average price and time, weighted with all acceptable combinations (not “main modes”)
Sampers – Emme/2
Passengers know the travel time components and headway of all routes; but not the timetable
Passengers choose only the perceived best stop (one mode) and assigns on routes in proportion to frequency only, without regard for time and fare. (Correct with no knowledge of timetable)
No distinction between ride times and prices of lines of a mode (e.g. problematic for high-speed rail)
Violates the assumption that travellers know the timetable and compare times and prices of lines and modes
Sampers – Logit model
Assigns on ”main modes” with travel time components per mode from Emme/2
Exogenous price per main mode in each O-D pair
Does not take into account combinations of modes
Does not take into account several airports (e.g. in Stockholm) or several operators
The logit model in use and the logsum for public transport
Larsen , Lang, Jansson:
” On combining discrete choice and assignment models”
HongKong paper December 2010:
“ long distance models that apply this approach, are miss-specified, leaving us with models that may have biased parameters and unknown properties when used for evaluations of changes in transport systems whether logsums or “rule of the half” (RoH) are applied. The core of the problem is that the implicit random terms of the discrete choice models will not have the assumed properties.”
Sampers – Samkalk
Consumer surplus is based on change of generalised cost only for the alternative mode that has been subject to change
For existing/remaining passengers a rectangle
For new/lost passengers a triangle (rule – of –half).
Correct if price or ride time is changed
Not correct if headway is changed
Headway changes mean that travellers shift between departures, so there is nothing like existing and new travellers
Changes of headway M9 (air) and B7 (rail) Headway of M9 is reduced, for ride times, 30 – 90 min . Samkalk produces much larger change of CS
Main mode concept - consequences We want to evaluate the effects of introducing high-speed rail between B and C for travellers going from A to C via B. The change in consumer surplus according to Vips is large but zero according to Samkalk. The reason is that Sampers does not take into account the possibility to go by air first between A and B. Since air is not taken into account when rail is subject to change, Samkalk cannot calculate the effect.
Main mode problem – Example Luleå-Örebro Vips gave a change in CS by 2 500 and Samkalk by 2 only. Only night train Luleå-Göteborg, via Örebro is taken into account. The combination airline Luleå-Arlanda, Arlanda express train Arlanda-Stockholm plus the rail lines Stockholm-Örebro is not taken into account. Evaluate faster train Stockholm-Örebro
Vips and Visum
One step for travel paths, lines and modes
The algorithm in these network models is called RDT – Random Departure Time. Both ideal times compared to actual times and departures are randomly distributed.
Vips does not generate demand, but Visum does
Vips does not take into account randomness wrt taste, measurement errors etc.
Vips has no scientific method for calibration
Sampers does not take into account randomness wrt departure/arrival time vs actual times
Emme/2, logit model and Samkalk fails to take into account the assumptions on travellers´ behaviour (known timetables)
Logit model and logsum in use do not work well for public transport (miss-specified)
Sampers fails to calculate CS when headway is changed and cannot handle combinations of modes