Future of AMTRAK

I started to work on this project a few years ago, when I was carless and moving around Washington on AMTRAK Cascades a lot. As time goes on I find that I like this plan more and more, and would make a big difference for the United States.

When designing an infrastructure network you have to first understand how you need to look at it. With transportation in particular this becomes particularly important, since it makes a big difference if you are looking at it from a global, national, or regional perspective. In my previous post, Future of Airlines in the United States, I showed that over 50% of American flights are less than 616 miles (991 km) in length. For comparison, the distance between Chicago and New York is 790 miles (1271 km) according to Google Maps, and the distance between Chicago and Denver is over 1000 miles (1600 km). Coast to Coast travel which have a minimum distance of 2500 miles (4000 km) make up less than a quarter of all flights in the United States. Furthermore, a 600 mile (960 km) train ride running at 160 miles per hour (or 250 km/hr, the minimum speed required to be considered high speed rail by international standards) will take less than 4 hours, meaning that having AMTRAK dould be a reasonable alternative to a very large proportion of flights in the United States, saving people a lot of time waiting in airports for their flights to arrive. Furthermore, integrating our rail and flight network in an easy to use system could make travel in the United States a breeze.

So, when determining what cities are most important, the first step I had was to make this spreadsheet of the largest cities in America and calculate the approximate distances between them and the next largest American city. Being the largest, New York got a score of 20,000 km, being the radius of the Earth. I used kilometers instead of miles because I prefer the metric system for its ease of use and conversion, and since miles are defined in terms of kilometers I am basically skipping the middle man. I also personally wish we used the metric system because it is more robust as a system.

The results at the top are not surprising at all, and I used Combined Statistical Areas because I am not concerned with connecting San Jose and San Francisco, since they are part of the same metropolitan area, and the smaller more fine sized metropolitan areas the Census Bureau has were not useful for this purpose. Local government in my mind are responsible for connecting their metropolitan area, and more fine metrics would simply gum up my spreadsheet with no benefit. Only 7 cities are significantly more distant from the next larger American city than the 900 kilometers which half of American flights are done in, and Atlanta is the 8th largest at about 900 miles to Baltimore/Washington. To determine the most important hubs I also take into account their population, and by multiplying the distance by population rank, and also by population I get two easy to use metrics to determine the relative importance of American cities for regional planning purposes. There is no surprise that New York, LA, and Chicago dominate the index. Selecting all of the cities than which have a score of above one million gives us 19 cities which will be major hubs connecting to their surrounding cities. In order for the most efficient use of resources, building a high speed rail network would focus on these 19 cities first, ideally so that AMTRAK can be self-supporting as soon as possible.

The next step is figuring out which cities need to be connected by AMTRAK to these hubs. There are a couple ways we can determine this. First of all, we can look at the most flown air routes in the United States which are under 1000 km as prime targets for building high speed rail first. Using Python based on public data we can quickly find out the routes which are in the most demand. The top twenty routes for building high speed rail based on current passengers between city pairs are:

  1. Orlando (MCO) — Atlanta
  2. Fort Lauderdale (FLL) — Atlanta
  3. Tampa (TPA)— Atlanta
  4. Atlanta — Ronald Reagan, Washington DC (DCA)
  5. Jacksonville (JAX) — Atlanta
  6. Houston (HOU)— Dallas (DAL)
  7. Miami (MIA)— Atlanta
  8. Palm Beach (PBI) — Atlanta
  9. Raleigh/Durham (RDU) — Atlanta
  10. Detroit (DTW)— Atlanta
  11. Dallas/Fort-Worth (DFW)— San Antonio (SAT)
  12. Los Angeles — San Francisco
  13. Dallas/Fort Worth (DFW) — Austin (AUS)
  14. Orlando (MCO) — Charlotte (CLT)
  15. New Orleans (MSY) — Atlanta
  16. Southwest Florida/Fort Meyers (RSW) — Atlanta
  17. Orlando (MCO) — Miami (MIA)
  18. Baltimore (BWI) — Atlanta
  19. Charlotte (CLT) — Chicago (ORD)
  20. Las Vegas (LAS) — San Francisco (SFO)

The most impressive finding from this approach is the absolute dominance of Atlanta for southern travel, which is not too surprising when you look at the map of the largest cities in the United States. This implies that the South is the best market for starting to build our high-speed train network. When you look at the map of AMTRAK service in the South it is even more clear why so many people choose to fly in that part of the country.

This is clearly not enough service for this country, and we need to increase the amount of AMTRAK in order to increase the ease of travel in our country.

The advantage to this approach is that those 20 airline routes can be supplemented with just 6 high speed rail lines. Two will be in the south west comprising of 1 connecting LA and the Bay Area, and another 1 connecting Las Vegas to the LA- Bay line, 1 line connecting Austin/San Antonio with Dallas-Fort Worth, 1 line connecting Houston and Dallas-Fort Worth, 1 line connecting Atlanta with Detroit, 1 line connecting Atlanta with DC and the cities in the Carolinas, and 1 line extending south from Atlanta to Florida and then doing a loop around Miami and the Fort Meyers area. These lines can then be profitable, since there is obviously demand to get between these cities, and at the end of the day we will have a much better country.

If we decide however to determine our cities clusters mathematically we can determine those using Python’s tool DBSCAN. I’ve gathered the 60 largest primary statistical areas in the United States and graphed them on an x-y axis. We can then use DBSCAN to select cities within a certain radius of each other and graph them into clusters. Each separate color represents a different cluster. The results are like this for a distance of about 190 kilometers:

This clusters our cities into 12 groups, with 84% of cities included with at least one other pair. 190 kilometers however is a very short distance for high speed rail, and should only take about one hour, so we can double it to be around 380–400 km to get a more reasonable estimate of a 2 hour train trip. This looks like:

Now we can see that with a 3 hour train trip the entire northeast extending all the way to Birmingham, Alabama is part of one cluster, and we have a total of only 7 clusters in the country, represented by large dots of various colors, with 11 cities not quite meeting the threshold, represented by small red dots.

For full reasonableness we can then go one step further still and through this program extend it up to an approximate 475 km distance, or a 3–4 hour train ride the results become quite striking:

With the exception of Salt Lake City and Minneapolis every US metropolitan area with a populaton above 1 million is within 475 kilometers of another metropolitan area over one million people, and are easily grouped into only three clusters, with the Eastern US which extends from Maine to Texas, the Southwest, and Northwest with Seattle and Portland sitting out by ourselves.

We can then move forward and expand across other high density routes, and by only connecting major American cities which lie within 600 km (or a 3–4 hour train ride) of each other we will eventually end up with a map looking somewhat like this with only our most populous cities:

This is something that most people in the United States do not understand, which is that if you only take the 60 metropolitan areas with the highest weighting between size and isolation in the country there is not a single major American city which is more than 600 km from another major American city. By building a model which focuses on the close city pairs, which compose the majority of flights in the United States we can build a railroad network connecting every major American city to every other at high speed at a profit. These 60 metropolitan areas contain 203 million people, or 2 out of every 3 Americans as well.

If we decide to include metropolitan areas as small as 526,000 people, or the 100 largest cities in the United States the map ends up looking like this with a distance of about 190 kilometers between cities in order to be included in a cluster.

Given how AMTRAK currently runs a train out to Eugene, Oregon (ranked 131 currently) from Portland which isn’t even included in this map, I see absolutely no merit to the argument that the United States is too big or spread out to use AMTRAK.

More than that, it is not that high speed rail is unpopular, 67 percent of Americans when polled say that we would use it. People in my demographic, between 18 and 24 poll at 78% saying we want to be able to ride high speed rail. I believe that if we pressured Congress then we could end up with the greatest high speed rail network in the world.

On the question of whether it would be profitable, I believe based on how other countries work that it could be. Anti-rail spokespeople will point out China’s government owned corportation does not make a profit, but it is also true that China ranked 79th on the last corruption perceptions index, while America ranked at 18th. Comparing how China’s government does as a bench mark for the United States is an apples to oranges comparison and not valid. It is better to compare us to other countries near the top of the corruption perception index, and when you look at Deutsche Bahn in Germany you find that they pulled a €1 billion profit last year. Austria is in a practical tie with the United States in terms of corruption and their railway corporation pulled a €5 billion profit in 2015. Despite what many anti-mass transit activist will point out, France’s SNCF corporation pulled a €377 million net profit in 2015, although after impairment (depreciation of stock) they had a loss, and their most important routes are covered by high speed rail. It would be possible to change SNCF to a profitable corporation given sufficient oversight and investment. Looking at railway companies around the world shows that if left to be managed properly can be both publically owned and profitable.

Another important step is that America should own its railways through AMTRAK and let freight corporations use them the same way trucking corporations use the freeway. Pay a license fee, and stay on the right side of the road when you are going slowly. It is generally agreed upon by economists that natural monopolies don’t work in a free market, but this is the way almost all of our intercity railroads are managed today. Governments own almost all other types of infrastructure, eg roads and airports, and cities usually own the electric grid and charge far less than their counterparts in the countryside, so there is no reason railroads should be treated any differently. The only reason I can think for their special treatment is that the railroads were mostly built during the Gilded Age, when the separation between government and major corporations was all but non-existent, even compared to today, while other forms of public infrastructure, such as our roads and USPS started earlier.

If the United States let AMTRAK purchase out the railroads through eminent domain and operate them, we would have more competition transporting freight and they would be better maintained. AMTRAK does operate one line in the United States, the Acela line between DC and Boston, and it is the only profitable route in AMTRAK’s network today. Better management of our railways is essential to improve our quality of life. This is what I want for the United States. We have the right geography. Let’s do it.

The work behind this project is available on github.