In a number of recent presentations Network Rail have flashed a diagram that they call the “Galaxy Plan”. It’s been up for only a few seconds and the print has been too small to read. But we’ve used the Freedom of Information Act to obtain a copy, and we’ve been exploring.

Network Rail are using “data science” to drill down into the data on delays and their causes. Published figures only show the division of delays among Network Rail, Southeastern and other Train Operating Companies. But the Galaxy analysis shows both a much more detailed breakdown and what is being done, at least in some areas, to improve specific causes of delay. Overall this looks like a good initiative to use data and management science to relentlessly drive out causes of delay, although some of the commentary shows that there is still a way to go.

We’re still studying the fine detail, but some features are already evident in the figures for delays between 16 October and 4 February:

• “Signal failures” are only 2.7% of the total delay minutes – even though they often seemed to be blamed. Indeed “signalling failures” – where the signals are working but have been set wrongly by the signallers – are nearly as bad (2.5%).
• Overall failures of Network Rail assets (track, signals, etc) only account for 16% of the delay minutes. (And it’s unlikely that much of the sub-three-minute delays are due to asset failures, because they could not be fixed that quickly.) Contrary to some of the spin, the poor performance is not mainly caused by asset failure.
• A total of 10.4% of the delays were attributed to external events, of which just 4% was due to weather – in the period October to February with both leaf fall and high winds (though not, in the last winter, much snow). However Southeastern’s autumn performance in 2016 was significantly worse than in 2015.
• The attribution of delay minutes is to the original cause. That flatters some of the figures, such as delays due to lack of train crew: if a train cannot run because the crew have been delayed on an incoming train then the delay appears to be attributed to why the incoming train was delayed. That provides no incentive for Southeastern to relax unrealistically tight crewing schedules.
• There’s a strange category of “Unexplained/Un-Investigated” delays. There seems to be an agreement with Southeastern that these will entirely count against Network Rail. We’re not sure what the financial implications of this are.
• There’s no attempt to explain 37.5% of the total delay minutes at all. That’s because the train reached its final destination less than 3 minutes late, so the cause is not investigated because it does not count for compensation payments between Southeastern and Network Rail. Arguably most passengers are less concerned with very short delays (all 10 trains of a five-day commuter delayed by 3 minutes is less noticeable than 9 trains on time and one delayed by 30 minutes). But overall it’s an enormous figure.