Visualizing LA Metro’s Ridership data, 2009 until 2016

Attention Conservation Notice: Here’s the animation I’ve been working on in order to understand the ridership changes at LA Metro, beware that the y-axes are all different, and that distorts the amount of variation going on. I did this mostly as timeline; YMMV.


The backstory:

About a week or so go while I was gardening, it occurred to me: if we various transit nerds were seeing the same trend we are seeing for the past few years with LA’s Metro ridership, only labeled for VMT instead, we’d be declaring that the market for vehicle travel was saturated.

Relax, I don’t think the transit market in LA is saturated. That was just me, getting grumpy with myself for being too lazy to examine my own biases.

But I sure would like to know what is going on. There are limits as to what you can explain with descriptive data analysis, but doing some critical visualizations put me a lot further along in my own understanding of what I think is going on, so that I thought I would share what I got.

Every so often, the Times reports on these numbers, and Laura Nelson wrote-up a nice story a day or so ago about bus ridership loss, and Sahra Sulaiman wrote up this depressing (but important) piece for Streetblog.

We’ve invested a lot of money recently in Los Angeles, and with capital investments at the scale we have undertaken, the last thing we want to do is put more money into obtaining fewer rides. Yes, I know, decades of neglect, yada yada, but we should be seeing nice, big jumps with early investments–diminishing returns should show up later.

Credible explanations: a) new rail supply is moving passengers from the bus to rail so that we are having fewer bus transfers and thus, lower counts; b) retirements and aging has prompted less commuting by transit as well as car (egads, let’s hope not as that is a demand effect); c) gasoline prices are low so that more people drive; d) the introduction of Uber and Lyft (then Zimride, thanks for the info Kendra Levine) into the LA travel market means that people handle the last mile problem (or the entire trip) with those services instead of buses; e) fare increases; f) reduced overall bus supply; g) the routes need to be reconfigured; h) bus transit is an inferior good*, so that we saw the highest possible usage during the worst of the recession, falling off as price-sensitive consumers at the lowest incomes leave the systems for other means; i) all that talk about fighting obesity and active transport hit home and more people started walking and biking; j) fare increases have forced bus riders to ride less.

It certainly looks like the UBER/higher fares/low gas prices combo are not helping, at least in the above timeline.

What did I miss? Single factor explanations are not likely here.

Then, there are various assertions that aren’t credible.

a) that “There’s actually more ridership, you just don’t understand how to use the numbers” directed at the Times’ Laura Nelson. Nah. As I demonstrate below, you can beat up on the numbers, and you’ll still have a trend. She’s reporting this correctly as far as I can tell even if transitlovers don’t like the framing being anything less than the Times’ usual, breathlessly supportive pro-development tone.

b) A hardy perennial: you’re using the wrong measure, that’s not fair! (Ridership counts are an incomplete measure of transit output–it takes more work to move 10 passengers 10 miles than it does to take 10 passengers 1 mile–but you need way better data than Metro is putting out there for the rest of us if you want to passenger-miles for the bus system, where the ridership loss is occurring. But I don’t buy this one. If that were the case, Metro would have an easy answer to the Times when the ridership story comes up. If anybody from metro wants me to diddle with better numbers, they know where I am at. Holler at me; I’m just here not getting my book done and having a midlife crisis.)

c) The averages do not capture peak ridership well, and our trains are doing that for us! That could be. But we don’t (or shouldn’t) build to peaks for any capital investment. Building to peak is one reason our auto infrastructure is so over-capitalized.

d) But, but, if you go back multiple decades when Los Angeles had fewer people, you would see that we have more rides, not fewer. Moving the end points around on analyses is certainly a way to manipulate what you see for trends. But that didn’t convince me either: of course we have more rides taken now than we did in 1970. But I still don’t see get why we’ve had the ups and downs we’ve had since 2009.

To create the graphic, I went to Metro’s Ridership statistics page and downloaded the data for all the years from 2009 to 2016. For the gas price data….eh, all I could find was a statewide average, but it should do well enough to indicate trends. These are data from the Energy Information Agency.

So all of these data have strong seasonality, and while you can see trends past the noise, it’s hard to figure out what is noise, and trend. I used a Fournier transform process to detect for seasonality, then decomposed each dataset into seasonality and trend. The data for the Metro 720 looks like this, where the top panel shows you the raw data, the second the seasonal variation, the third the deseasonalized trend, and the final panel the remainders.

Metro 720 Rapid Ridership, Jan 2009 to December 2016, Time Series Decomposition

I can show you a bunch of these, and they are all interesting, but here’s the bus and the rail overall. Rail, a strong upwards trend towards the end due to the Expo line and its extension, and buses, falling, long before the fare hikes in 2014 (but, I suspect, those fare hikes are not helping; the bus riders are going to be the most price sensitive customers due to their demographics).

Metro Rail Ridership, Jan 2009 to December 2016, Time Series Decomposition


Metro Bus Ridership, Jan 2009 to December 2016, Time Series Decomposition


Ok, yeah, let’s look at the red line. That’s a lot of ridership loss there I don’t understand:

Red Line Ridership, Jan 2009 to December 2016, Time Series Decomposition


Fuel prices Jan 2009 to December 2016, Time Series Decomposition


So here is a correlation matrix of the deseasonalized bus and rail ridership numbers plotted against gas prices. I could get fancy and start fitting the curve, but…it’s fairly clear that falling gas prices since 2012 track well with the ridership declines. But I do think we are seeing a rail supply effect here: rail patrons are less likely to be sensitive to fuel or fare prices than bus patrons, and we see that rail in general correlates positively with fuel prices, though not to the same degree as bus ridership. That’s kind of a weird result, but not really when you think that a big boost in ridership came from new supply, and supply to an area–downtown Santa Monica–where riders are going to be comparatively better off than the rest of the region.

There is evidence of reshuffling exiting patrons–it would be nice to have the Santa Monica Big Blue Bus data to compare–as bus and rail ridership are negatively correlated at around -.40. We know that there are times where rail and bus act as complimentary goods, and there are other times when they are substitutes. If this is evidence that some riders got moved from buses to trains…well, that’s not the wonderous “we solved traffic and air quality” victory lap we might want, but those folks are probably getting a more comfortable ride, and I’m ok with that.

Correlation Plot, Fuel Price and Ridership by Mode, 2009 to 2016


That said, it’s not all reshuffling. At the end of the time period, we have about 70,000 more rides on an average weekday on the rail system–and remember, that even with some pretty big losses on the Red line (whyyyyy?) and Green line. But we’ve lost nearly 280,000 rides on the bus side, so reshuffling isn’t the whole story, either.

So here’s the correlation matrix between the routes I examined.There are network spillover effects in action, generally, with connections between lines being weak in instances you would expect (gold line, Metro 720: they don’t really feed each other) and very strong in instances where lines intersect. Interestingly, both the Green and the 720 are parallel E-W routes, but with a lot of real estate in between. The Expo Line, which runs parallel to them both, perhaps became the the E-W route of choice, which might explain some of those losses. Maybe.

Correlation Plot, Routes by Ridership, 2009 to 2016


Falling gas prices and higher fares are probably not helping; lags in land use probably are not helping; Uber/Lyft are probably not helping; reshuffling is a potential explanation but not, really, a problem from my perspective.

Bah. I have a better handle on what is going on, but not a convincing story (at least not for me, not yet) for why. But the pictures are kind of cool.

*If you have ever attempted to lecture me about housing supply and demand on Twitter, but you yell and scream about this term, then go back to your micro 101 class. It’s a technical term that describes a good where consumption increases when incomes falls. I still love transit, but–again technically–the term applies empirically, so I use it.