I've been busy staring at YouTube videos of trains lately, not because they're all that exciting, but because I'm trying to figure out what typical consists are for various Amtrak routes. A few videos have been fun, though, like this one featuring Metro-North Railroad maintenance-of-way crew moving through after the fourth train. It keeps going, and going, and going...
Anyway, thanks to Paul Druce's table of annual train-miles from earlier this year, it's possible to calculate average passenger and seat counts with some level of precision.
The good news is that this seems to solve one question Paul had about why different services seem to have such a variation in the cost charged per unit distance: There's a strong correlation between numbers of seats and cost per train mile.
The bad news is that the number of seats per train seems really bogus in a lot of circumstances. I've been fumbling my way around Wikipedia, some railroading forums, and spending time watching a lot of videos to try and determine typical consist sizes, and therefore actual seat counts. It's an error-prone process, but I think I've put together some figures which are at least in the ballpark.
The numbers I've calculated for what Amtrak is recording for seats-per-train are also subject to some error. I know there is some level of error in my calculations because I'm working backward from figures for overall contribution/loss per route and contribution/loss per passenger-mile found in Amtrak's September 2011 year-end report (FY2011), but most of those figures have at least 2 digits of precision, and would only be off by a small amount. There are a few situations where Amtrak's figures are only known to 1 digit of precision, but that should still generally be less than a ±10% variance from reality (or reality according to Amtrak, at least).
Similarly, the train-mile calculations I'm using are only accurate to within a few percent because of the method of calculation (number of trains per week multiplied by 52 weeks, though that gives some allowance for canceled trains).
Still, given all of those problems, several of Amtrak's corridor trains in the Northeast (though not necessarily NEC trains themselves) stand out as having highly questionable figures.
The standout case revolves around the Adirondack, Empire Service, Ethan Allen Express, Maple Leaf, Vermonter routes, which all appear to operate with exactly the same passenger car consist of 5 Amfleets: 4 coaches and 1 cafe car. There does seem to be some non-obvious mixing of the individual carriages: some are high-capacity Amfleet I cars, while others are lower-capacity Amfleet IIs (the second-generation Amfleets were generally intended for long-distance service). But even accounting for that variation, some numbers are really outlandish.
My calculations based off of Amtrak's figures come up with an average of 554 seats per Maple Leaf, for instance, an impossible number for a 5-car train. On the other end of the scale, my math produces an average of 87 seats per Vermonter. It would seem that each of those trains has a typical capacity around 340, so something isn't right.
I'm not sure what's happening here, but it seems that Amtrak—typically operating fixed-size consists—is charging routes based on the number of seats the trains would have had if they were actually shrinking or growing trains in accordance with demand.
In some ways, I can see how that's not really a problem—4 out of 5 of those trains terminate in New York City's Penn Station (the last, the Vermonter in Washington, D.C.), where the added hassle of switching out train cars might gum up the works so much that it's better to just run with interchangeable fixed-sized trains. In the name of operating efficiency, some accounting tricks might not be a bad idea in order to keep the charge per route equitable. But it still seems to be covering up the true costs of running these trains while also inflating the overall cost of operation. Trains would need to be sized for the maximum load on any of the interchangeable routes (which in this case appears to be on the Maple Leaf).
Nationally, there seems to be a pretty consistent practice of under-counting seats by about 15%, though some routes work out to be very precise: Some state-run and mid-continent corridor trains seem to have pretty close numbers, and the Acela Express, with a fixed consist of 304 seats, is just about bang-on as well. My local long-distance route, the Empire Builder, is also low by a considerable margin for FY2011, though I believe that was due to a long string of annulments west of Saint Paul due to flooding which dropped ridership by about 12%. FY2010 numbers were almost a perfect match in comparison.
Anyway, here are the basic results of all this work (with further details in my spreadsheet):