One of the big issues we have had with public transport modelling, which is used to make forecasts about how many people might use different routes in the future, is how the transport models deal with transfers. A transfer (or connection) is where someone changes from one service onto another. The transport models have long tended to predict that very few people would want to do this. Yet in most cities, transfers are a part of life and large numbers of people will transfer between different routes to get to their destination. It is after all the basis for metro networks.

Having observed huge numbers of people transfer between bus and train at Panmure on a number of occasions, I was sceptical of the modelling and decided to request actual information from Auckland Transport about the rail/bus transfers for Panmure station.

Information in the AMETI documents indicates that previous forecasts for Panmure Station 2016 AM (7-9am) demands were 1040 rail passengers. The previous forecasts for 2026 AM were 1439 passengers.

The Auckland Regional Demand Model passenger demands for 2016 AM are (total) 1236, of which 134 transfer from bus to rail. The transfer demands correspond closely to data provided In 2) below.

Panmure Transfers

So we were right the model was not anticipating the level of transfers, but what are AT doing reflect this, well I also asked that as well

3) Any work done on whether modelling tools should be adjusted in the treatment of transfers between bus and train over the past five years?

It has a confusing response as on the one hand AT respond with there is nothing wrong with our model, it works fine no need to question it thank you.

But, then, on the other hand says, however, we updated the model to reflect transfer penalties by giving Britomart its own class better which lowers the transfer penalty of the classes below as they move up.

I think the main point though is these models are not really scientific they are created with the inherent biases of their creators and should be treated with caution. If want PT to be on a level playing field then we need to update the models to reflect reality if not the case will always say expressway.

Share this


  1. Last time they tried forecasting on any scale was in their LTP/NLTP submission in February.
    Thought they would have learned a little more humility.
    Keep at them Harriet.

  2. Have these people who create these models never travelled overseas? I think most public transport trips I’ve taken when overseas involve a transfer, and often not at some fancy “transfer station”, but just get off at one stop, and around the corner is they next stop.

    1. Although most of those countries have much higher frequency. There aren’t many places in Auckland where you wont wait 10 minutes or more to transfer.
      Direct PT in Auckland can barely compete with a car in terms of duration, let alone PT with transfers.

      1. Agreed (as an exAucklander who’s lived overseas for several years). My comments below notwithstanding, overseas transfer-based systems work (only?) because you are talking about five-minute frequencies across *both* legs of a transfer-based journey.

  3. Great job Harriet. As a Panmure resident and PT user, it has been good to see the tide of passengers washing off the buses into the station.

    And the new bus schedule and extended high-frequency hours for trains has reduced the panic-stricken dashing from the bus to catch the train, which used to be the norm. All good for improving people’s acceptance of transfers as a useful PT tool.

    All bodes well for the Eastern Busway. Bring it on…

  4. I’d like to see transfer numbers for Panmure now that the New Network for East Auckland has been rolled out.

    I expect those to be far higher.

  5. The Auckland Regional Demand Model predicted 134 passengers transferring bus to rail. The reality for the year ending
    – January 2016 was 123 (this is the earliest full year available after the figures made that big jump up – presumably that happened after the upgrades.)
    – June 2016 was 149
    – December 2016 was ? – can we assume something like 175 (based on the increase between the two above, or do you have a better idea?)

    Harriet, do you think they don’t have the full 2016 figures? Surely they need them to compare them to the predicted 134 for 2016. From my estimate above it looks like the reality could have been 31% higher for 2016 than predicted.

    That’s not really acceptable, is it?

    1. I think I’ve misunderstood something there… never worked with passenger modelling before. Can you explain the figures some more, Harriet? Is the only modelled figure for 2016 the 1040 passengers, with nothing given for transfers? If so, what’s the 134 for 2016 from? Doesn’t seem to be the counted transfers? Confused…

  6. For a transfer-based system to work well, *both* legs of the journey need to have a high service standard, especially service frequency. I would proffer the view that the original spec of the models took a quite conservative view as to the numbers of transfers, because that is where past experience with transfers, in an Auckland context, would have pointed them.

    Think of it as the principle of the tipping-point, but this time a positive one; because both the legs of the transfer-based journey are now up to a sufficient standard, a transfers-based system will work much better than previously – or in others words, we are seeing a step-change,.

    1. You would hope that the modellers would be aware of this tipping point and would be able to build it into a model. I’m sure it has happened in many other places in the world so there must be some emperical data.

  7. I am not sure what you are expecting from their model. The mean for the 2016 data is 174 and the model suggested 134 or only 40 fewer. It is not a simulation model. If you want a simulation model for transfers at every station then they would have to spend millions of dollars to build them. But what would be the purpose of that?
    Any aggregate model that gets within 100 of the actual on a specific link is doing what it is supposed to do.

    1. So if lower-than-actual transfers are predicted by the model of the whole network, can you forecast better by forcing a higher transfer rate?

  8. It would be interesting to see more recent numbers. These ones cut out just before simpler fares were introduced. As a Pakuranga resident, I only started transferring to the train at Panmure once simpler fares were introduced, as it would have been more expensive otherwise. I expect there would be an uptick with simpler fares and then another one with the New Network.

  9. i will be fascinated to see what happens when Wellington gets integrated ticketing with rail – bus transfers at the station, I doubt there is much at all currently.

  10. To be fair to AT there have been lots of places having difficulty modelling PT patronage lately, especially when a new service is intended to change travel behavior. Even in places in Europe where PT mode share is much higher it remains a hard task.

    For that reason I wonder if we need a plan B? Can we use
    a benchmarking or econometric process to estimate patronage instead, or a likely range for patronage? We should start compiling a database of similar projects in cities with similar scale and density and observe their patronage, to get a feel for what we might expect. Most 4 step models were first built to forecast freeway project volumes, and they are starting to feel their age.

    1. One thing that they could do, is to calibrate their model against an existing service. ie, when anyone models for AMETI, they should be forced to check that their model gets close to a correct output for the northern busway.

  11. 1) Write back to AT and ask for the modelled transfers vs actual for the base year modelling.

    2) Pulsed transfers go along way to reducing user transfer times & user uncertainty about the PT system.

Leave a Reply

Your email address will not be published. Required fields are marked *