After yesterdays post on bus performance this comment from Peter got me thinking.

One thing that’s always annoyed me about reliability and punctuality statistics is that they generally don’t account for varying passenger loads throughout the day. For example, if the 10.30pm train runs 10 minutes late then not too many people are affected. If the 8am train runs late then a huge number of people are affected. It’d be good to get the stats to account for this somehow.

I had seen information on rail performance in the patronage reports but had always dismissed it as it never seemed that useful. Here’s what the AT report says:

Now my understanding is that delay minutes are counted only for trains more than 5 minutes late and that each the delay is multiplied by an estimate of the patronage on those services. Going back and looking at it more closely I found out why I have never really liked this measurement, the problem is it doesn’t really give good context as to how things have changed from month to month. As an example the months of August and September have some of the most delay minutes but the are also some of the biggest months patronage wise so how does that really compare to say December and January which have less delay minutes but also much less patronage.

So what I have done is go back through the AT and ARTA reports to get the delay minutes. I have then divided that by the number of passengers carried each month to see the average delay per person. The result is below:

This is the average delay in seconds across all passengers but you can clearly see some trends that are to be expected like increased delays around the Christmas shutdown. It does show that while there were less overall delays in Jan compared to September, the Jan delays were worse for those catching trains. Another thing I noticed is that on average delays seem to be not much better now than they were four years ago which means that despite all of the works going on we haven’t really seen any major improvements.

I’m still not happy with this analysis as it doesn’t really take into account things like what the punctuality is so I will keep working on it with what data I can find. One thing mentioned by others in that last post is that it would be great to see all of the raw data to get a better handle of things, something I really agree with. It is also an area where integrated ticketing could really help in seeing things like exactly how many people are on each service. It also only counts delays for trains that are more than 5 minutes late meaning that a train carrying 500 people that is 4 minutes late isn’t considered but one carrying 100 people that 5 minutes late is considered even though the total delays for the first train is much more. To me this buffer means that we don’t consider the impact of delays fully and while I can understand the reasons it, when it comes to measuring the impact of delays, it means that not everything is counted.

Share this


  1. All of the work would be hopefully reducing delays due to signalling faults, but since there’s still a heap of work going on in that area it’s not surprising there’re still a lot of delays. Ultimately, I don’t think we’ll see any major improvement until the EMUs start running and the signalling is complete and bedded down, so basically I don’t expect much change for a couple more years yet.

  2. Another thing that needs to be taken into account with these numbers is service frequency: If a train is running every 5 minutes to Britomart from Otahuhu station I couldn’t care less whether they’re all actually 10 minutes behind schedule. Likewise on B-Line routes – for most passengers they won’t catch a specific bus, so frequency is more important. This is where we need to aim for – most passengers not catching a specific train/bus, but just showing up at stops because there’ll be one along any minute.

    When frequency is lower (every 20/30/60 minutes) – either off-peak or on smaller routes – whether a specific bus/train turns up and leaves on time has a massive impact on individual passengers. If my 7pm bus doesn’t turn up, and the next one to my house is 8pm then I’ll be home an *hour* later. On a B-Line route I may not even notice that I didn’t catch the “right” bus.

    Maybe a measure of Delay Impact? the time between delayed/cancelled services and the next service on that route for each stop along the way.

    Basically, we need all the GPS data from trains, buses, and ferries for every month, and the hop-card data for every stop & station. That way advanced analysis can be done to come up with better performance indicators.

    Rob 🙂

  3. From NZ Herald :
    “KiwiRail spokeswoman Jenni Austin apologised to people who missed the start of the rugby because of a combination of incidents that caused delays. As well as the dog being hit at 4pm, there had been two points failures at Papakura about 4.30pm which were not repaired until 7pm, and a train had to be stopped at 6.30pm after over-running a signal.”

    How is it after spending a lot of money upgrading the rail network we still have points failures, let alone 2 points failures in one critical evening?

    I understand that AT is going to be paying Kiwirail a lot for track access fees going forward — I sure hope the contract has penalty clauses for getting money back when the network isn’t performing.

  4. I think the new signalling has yet to be commissioned at Papakura so the points machines that failed may still be old ones

  5. Measuring rail performance is my day job. Some thoughts:

    * The main measure of performance used in Great Britain includes an element to account for cancellations. Areas with a less frequent service have a higher weighting put on those cancellations than areas with a higher frequency. The number reported is thus a joint measure of reliability and punctuality.

    * A lot of effort is made to track down what causes the delays, where they are caused, and when. This does help in addressing the problems; over the time I’ve been in my job, I’ve seen our network company and the main operator make a huge dent in the level of delay generated in the system.

    * There is also a performance regime between the network and the operator which, because of the monies involved, also act as a major incentive to get it right.

Leave a Reply

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