Link Light Rail has had some rough patches over the last month with a couple of major incidents occurring during peak commute times. The first was a Link-involved accident on MLK at the tail end of the Sept. 15 afternoon commute. A week later, Link was disrupted by a drive-by shooting in Columbia City that resulted in a crash near the tracks. These incidents are a timely reminder of the vulnerabilities (and dangers) of a rapid transit system that shares right-of-way with vehicle traffic. But how often is Link service disrupted? I’ll dig a bit into the data to shed some light on the statistics of Link interruptions.
My data source here is the Sound Transit alert feed which emails notices to subscribers when service interruptions occur. Service interruptions are anything causing delay or temporary closure, such as a mechanical problem with a train, power outage, blockage, or medical emergency. I’ve been subscribed to the alerts since December 2015, so I have about a year and nine months of service interruption records to analyze. In my experience, ST is pretty consistent with notices when an interruption happens, but it’s possible this dataset leaves out some incidents. Numbers reported here are, hence, minimum rates.
Overall, I count 76 separate Link service interruptions between Dec. 4, 2015 and Sept. 25, 2017. Given that there are about 660 days in the whole period, the high-level conclusion from the data is that you should expect a Link interruption about once every 9 days. That’s only an average, though, so sometimes there are more. What does the data look like? Below, I’ve graphed interruptions* by date and the duration of the service disruption in minutes.
I classified the interruptions into four groups (quartiles) based on their durations. The “Level 4” interruptions are generally an hour or more – and are the ones that daily Link riders may hold in their memories for some time – while “Level 1” events are 10 minutes or less. The median interruption is about 25 minutes.
It’s also interesting to just count the interruptions occurring in different time windows. In the chart below, I broke the observation period into 30 day intervals and plotted the number of interruptions in each interval.
The average number of interruptions per 30 days is between 3 and 4, with a notable peak of 8 interruptions around March 2016. Maybe some of these were glitches related to the U-Link opening? It can be tempting to claim a downward trend in the interruption rate with later dates, but the recent reductions are well within the natural variation of a statistical process** like this. In other words, there’s not enough data to claim that we will see fewer interruptions in the future.
Are interruptions more common on certain days of the week? Emphatically, yes. Service interruptions have overwhelmingly happened on weekdays. That’s not at all surprising given the larger numbers of both trains and cars during the work week. The chart below shows the breakdown by day-of-the-week.
There were zero Saturday interruptions in my dataset and only a couple of Sunday interruptions. Again, it is tempting to pose theories about why there were more interruptions later in the work week. However, I will again play the wet blanket card and say that the variation amongst the weekdays is well within the natural range. There’s not conclusive evidence here that Thursdays will see more interruptions than Tuesdays in the long term. What I can do is refine the interruption rate statement a bit: On any given weekday, there is about a 15% chance of a Link interruption.
What about time of day? I’ve got a chart for that too:
Evidently, the afternoon and early evening are more hazardous times to ride Link.
One final question: How does Link compare with other light rail systems when it comes to service interruptions? This is hard to answer because incident data is not generally available, but I do have some glimpses into other systems. For instance, The Oregonian reports that Portland’s MAX had around 300 interruptions in 2015. Considering interruptions per track mile, that puts the MAX rate in 2015 (~5 interrupt/t-m/yr) at more than double Link’s current rate (~2 interrupt/t-m/yr). MAX has a lot of operational complexity, including many at-grade crossings, so that may help explain such a large difference. On the other hand, Denver’s RTD system – which also has many at-grade crossings – seems to have an interruption rate that is closer to Link’s – with 88 interruptions in 220 days of 2016 over about 80 track miles (~1.9 interrupt/t-m/yr). It’s hard to know if these are apples-to-apples comparisons, but they at least provide some context for the numbers I’ve reported in this post.
[*] I’m missing duration estimates for 6 of the 76 interruptions. This is either a shortcoming of my interruption detector or due to a missing email from Sound Transit’s alert system.
[**] For those who care about such things, the Link interruption data exhibit the properties of a classic Poisson process when aggregated at 30-day intervals.