Asymmetric Mode Choice
In most models I have seen, ridership and mode choice are assumed to be symmetric: if I take the bus to work, I will also take it back home. Of course those models distinguish home from work: if a bus is full inbound in the morning it’s not expected to be full outbound in the morning. But the assumptions are that if the bus is full inbound in the morning, it should be full outbound in the afternoon. To a first-order approximation this is fine, but there are multiple situations in which people can choose differently in each direction. This is less relevant when discussing cars and bikes, because if you use them in one direction you must use them in the other, but it’s relevant to car-share, bike-share, various kinds of buses and trains, walking, and flying.
Most of this post will take the form of anecdotes. I have not seen any model that accounts for these cases, or any discussion elsewhere. The only exception is when large changes in grade are involved: people walk or bike down more easily than up, and this means that in bikeshare systems, the operators sometimes have to tow bikes back to high-elevation neighborhoods because people persistently take them downhill more than uphill. However, in addition to asymmetry caused by physical geography, there’s asymmetry caused by urban layout and transit system layout, as well as asymmetry caused by different characteristics of trips.
Case 1: Frequency Splitting
Consider the above image of a transit network. Point A is a major destination; area D is a neighborhood, and point E is an origin. The thick black line is a rapid transit line, passing through and stopping at B, C, and E. The red and blue lines going east from A are frequent rapid bus lines. The gray line going from A to E is a lower-grade bus line: less frequent, and/or slower.
In this image, traveling between A and either D or E, the frequent buses will be more useful going away from A than toward A. For an A-D trip, if I live in neighborhood D and travel to A, then I need to choose which of the two parallel streets to stand on, whereas going back from A to D, I can stand at the bus terminal and take whichever bus comes first. For an A-E trip, if I live at E and am going to A, then again I need to choose which of the two bus lines to use, that is whether to get off the rapid transit line at B or C, whereas going back this is not an issue. On the margin, I might choose to take the lower-grade but direct bus from E to A but not back.
Neither of the situations is hypothetical. When I went to college at NUS, my situation was similar to the A-E case: while the subway has since reached campus, in the mid-2000s the campus was connected by buses to two separate subway stops, Clementi and Buona Vista, and although some parts were definitely closer to one than to the other, the connecting buses served all parts of campus relevant to me. There was no equivalent of the gray bus, and I’d almost always take a taxi to campus, but usually take transit back. Now that I’m in Vancouver and work at UBC, where bus lines converge from parallel east-west streets, my situation is similar to the A-D case, since I can take Broadway buses as well as 4th Avenue buses; there is no alternative to the buses for me, but if there were, for example bikeshare, or walking if I lived closer to campus, I might well use it.
In those examples the asymmetry is for the most part unavoidable, coming from urban layout. In Vancouver, there are multiple east-west streets on the West Side that deserve frequent bus service. Consolidating everything on one street can come from the opening of a Broadway subway to UBC, but because the asymmetry is a second-order effect, the main argument for the subway has little to do with it.
Case 2: Waiting Facilities
Some bus and train stations are notorious for being unpleasant to wait at. Tel Aviv’s Central Bus Station is dark and labyrinthine. In New York, Penn Station and Port Authority are both unpopular. Many older airports are infamous for their poor amenities and confusing layouts. Because people need to wait going outbound but not inbound, this could affect mode choice.
In Vancouver, UBC has two separate bus loops, one for generally express diesel buses, and one for local electric buses; each loop has buses going on multiple streets, as in the above image. I find the diesel loop noisy and disorienting, and therefore avoid it, waiting at the electric loop or the next stop after the loops. Therefore, I usually take electric buses back home from UBC, while I almost always take diesels toward UBC. I have no direct experience with Kennedy Plaza, but other Providence-based bloggers think little of it; I think it was Jef Nickerson who noted that buses going on the same trunk routes are not co-located there. This could induce a similar asymmetry.
It gets worse when bus stops do not have shelter from the elements. Sheltered stops should be included in any bundle of best industry practices, but when they are present only downtown or at major stations, they can bias me to take the bus in just one direction.
Transit agencies can eliminate this asymmetry by making their facilities better. Usually the cost of shelter, clear signage, a bus bay layout that makes identifying the correct bays easy, and similar improvements is negligible, and the benefits are large. Of course, independently of any asymmetry there is no excuse for not having passable facilities, but in some cases, such as the UBC diesel loop, the situation on the ground is worse than it appears on planning maps and this worsens the passengers’ experience.
Case 3: Stress for Time
For some trips – going to an airport or intercity train or bus station, going to a meeting, going to class, going to a workplace where I need to be there at a specific time – there’s a more pressing need for timeliness in one direction than in the other. This biases in favor of more punctual or faster vehicles, even if they’re more expensive or less pleasant. This manifests itself in airport choice (when flying out of New York, I strongly prefer rail-accessible JFK, while my preference for flying in is much weaker), willingness to transfer to save a few minutes of trip time, willingness to ride a more expensive but faster train (for example, the LIRR versus the E), and bus versus train decisions (trains are almost invariably more reliable, often by a large margin). The few times I used transit to get to NUS, I used the subway, whereas when I went back home I’d often use a trunk bus, which was slower but had a station much closer to where I lived.
In this case, there’s not much the transit agency can do. If the bus versus rail issue is persistent, the best thing that can be done is encourage more mixed-use zoning and more symmetric morning travel demand, so that buses would be used in both peaks and not just in the afternoon peak and vice versa for trains.
Case 4: One-Way Routes
To the extent that the transit activist community has an opinion, it is strongly against one-way pairs, going back at least to Jane Jacobs’ criticism in The Death and Life. I’ve written briefly about them; Jarrett has written more extensively. One more issue is that if a bus route runs one-way on different streets (as a consistent one-way pair as in New York, or in a more complex arrangement as in Tel Aviv and Singapore) and I live closer to one direction than to the other, I might take it in just one direction. In Tel Aviv it was not a major problem because the bus I took to middle school run two-way in the segment relevant to me, but in Singapore it was an issue in both middle school and college: I lived next to a very wide one-way street without nearby crosswalks, and getting to the other direction of the buses required crossing it and walking some extra distance; this helped bias me for taking the bus only in the return direction.
The Effects on the Margins
Since asymmetry is small enough an effect that models can ignore it and still come very close to predicting actual ridership, its effect on transit planning is only on decisions that are very close to begin with.
I believe the most common case is the one in the image. A city that transitions from an idiosyncratic network of infrequent direct buses to a regular frequent grid where passengers are expected to transfer needs to decide which of the infrequent buses to keep. It might even have a few peak-only express buses it is considering keeping. In this case, it’s useful to note in which directions the infrequent or peak-only buses are more likely to get passengers, and potentially have an asymmetric number of trips on those in each direction, recycling the equipment for nearby routes whose asymmetry goes in the other direction.
Pedestrian Observations 
A variation of the problem would be if the red route departed the “A” on the hour and half-hour, and the blue route at :15 and :45. Going outbound to “D” you have effectively a 15 minute service, the difference being where in the neighborhood you are dropped off. Heading inbound though, you have to pick between which street to stand on. At low frequencies the decision is made easier by the fact that if you have a timepiece and the buses are somewhat reliable, you can make a more informed decision based on the schedule. If the red route just departed, walk to the street with the blue route, and vice-versa.
I have the opposite problem getting from Prospect Park on the BMT Brighton Line in NYC to Herald Square. Inbound I take either the B or the Q, whichever comes first, since they both come on the same track. They then diverge once they enter Manhattan, but meet once at Herald Square, on 34th Street. This is convenient for inbound travel, but for outbound travel, you have to pick one or the other, since the IND 6th Ave. Line (B) and BMT Broadway Line (Q) platforms at Herald Square are not near each other and neither line has a countdown clock.
I guess the obvious solution here is that the MTA should get its shit together and install countdown clocks.
Even on the trunk lines, you have this problem at Penn Station. Going from 14th to Penn Station, I usually take the A, C, or E. But at Penn Station I have to pick express or local in advance (like picking between the red and blue lines at position D on the map), and on some weekends it’s not even clear which tracks are functional, so I just walk to 14th.
If I had a very convenient phone app (or very simple SMS-based system for dumbphones) I could solve the “red or blue from D” problem very easily, but not even the best transit systems seem to have that worked out fully.
This might be the case with ESA as well. If your destination isn’t in the immediate vicinity of GCT or Penn, you might not care which station you end up at. But in the reverse direction you have to choose.
Hi. I have an additional case for you (a real live example from my commute): Walking distance and (in)convenience of a connection can be asymmetric. (Especially the amount of walking can change depending on whether you need to cross a big road or not).
My work trip consists of a tram or subway and a bus. The bus is always the same, but I can either use a subway (half a mile from my home) or a tram (right next to my apartment). With subway I’ll get to the bus station where the bus leaves from and with the tram I’ll get to a stop a few stops towards my work place (the bus is the same in both cases).
On my way home, I need to cross a big street through an underpass; it’s a relatively long way – and I need to be alert for the bus at all times, since they go very fast and it’s easy to miss it.
So on my way to work, the subway is both faster and more convenient and on my way home, the tram is faster and at least as convenient.
We are still applying rather ‘primitive’ techniques to mode choice prediction in travel demand models, but I personally am rather sceptical that we can forecast this with any degree of precision 20-30 years out. Composite and shared modes are certainly expanding the possibilities available to travellers…. Note as well that mode choice does not explicitly dictate which lines will be used – only which modes will be considered.
Concerning your point about ‘asymmetric’ mode choice, far too many models focus only on the average week-day peak period, and then scale up those values to represent a 24 hour period (how realistic is it to assign a full-day demand matrix onto the network?). There are some models out there which model different assignment periods independently (morning peak hour, evening peak hour, perhaps other periods as well) and these will typically include a separate mode-choice model for each period.
Another asymmetry that you haven’t pointed is more fundamental: there’s not always a matching pair of “home-work” and “work-home” trips. In the evening, it is rather more likely for there to be separate “work-somewhere else” and “somewhere else-home” trips.
Agreed. Try a run from Geneva + Mission (Outer Mission sector) to downtown San Francisco. That can be either Muni Metro (LRV) or BART (subway). The return could be a diesel bus (#38) out to Geary + Masonic to do some shopping followed by another diesel bus (#43) to get back to the Outer Mission. Another two-legger could be a trolley bus (#5) to Fulton + Masonic to shop or a couple of stops more to USF for a night class or program. Then follow that with the final leg on the #43.
P.S. Geneva + Mission does have a bus connection (diesels : 8X, 8BX,14L, 14X; trolley : 14) to downtown S.F. But the rail choices are more comfortable and as fast or faster (BART) in spite of the dog-leg over to the Balboa Park BART station.
2 & 3 are important in my particular commute: I’m willing to pay $130 for a student LIRR pass because it guarantees me an extra half-hour of sleep a day, but I’ll take the E and a bus back home to Jamaica because waiting at Penn is dismal (and post-Sandy, there have been a couple of times where they’ve shut down services out of Penn.) There’s also a social aspect, in that on my way home, I take the train home with friends, whereas in the morning it’s harder to find people on the train.
Another factor (which is harder to predict, but is somewhat important) is weather conditions. If someone walks or bikes to their destination, and the weather changes drastically by the time they have to make the return trip (heavy snow/ice, excessive rain or wind, etc.) a lot of people will opt to take a cab or mass transit instead. I find that the buses I use are generally more crowded on days with terrible weather conditions.
I have a frequently asymmetric commute/trip. There is a bus and a trolley which follow each other closely for a significant chunk of the route before diverging (the bus was formerly a trolley branch). The trolley runs up the hill, the bus stays in the valley. I can choose either one to get downtown. The trolley has more frequently scheduled arrivals but no predictability, while the bus has slightly less frequency but real-time tracking through NextBus. The bus usually travels faster than the trolley except when there is bad traffic. So my typical approach is to check NextBus and then decide which one to take. I see other people doing the same thing too, so I’m not alone in this. (I also see people who don’t seem so flexible, and are hard-set on one choice).
Coming back, I also have multiple choices. I can wait for the trolley branch I need to show up. Or I can take most trolleys to the bus departure point, since I know when it is scheduled to leave, and connect there.
So most days are asymmetric for me. That’s assuming I don’t go somewhere else in the afternoon, either.
Asymmetric trip patterns are one reason that transit systems have service planners. In Edmonton years ago, I came in with a proposal to improve the headways in the mornings on what we called an industrial route. Loads were exceeding our standards in the energy boom then underway. My boss questioned why not the afternoon? I just had the numbers, so I invested some time in watching the line at its Westmount transit centre anchor. There was nothing wrong with the data, I saw. But what I saw was pick-up trucks and other cars driven by industrial workers dropping their buddies off to make connections with the diverse services available at Westmount — especially the combined 5-minute headway Rtes 4/5 trolley coaches. Since then, I’ve come across this buddy-and-ride phenomenon in other situations, although particularly in industrial settings.
This happens in a variety of ways in large networks, tending to balance out the equipment requirement. The problem only is important when the imbalance becomes extreme or when multiple service providers are involved. An elderly history professor of mine told about the downhill walk in her girlhood southwest Portland neighborhood in the morning to the Oregon Electric Railway followed by the downhill walk home from the Red Electric line above her street. Luckily for what were then called the Traffic Departments of the respective railways, this was a small neighborhood.
Some interesting points raised here. A few anecdotes and comments:
1) There are some bike and car asymmetric transit issues, probably not a generally a significant component in general though. Getting dropped off by car and then taking transit home (or the reverse), biking and leaving the bike in an office or on campus for a day or two, or folding bikes on transit may play a role there. But a well-known car/transit asymmetry is the Bay Area morning “Casual Carpool” where people line up to fill cars for the inbound ride from North Berkeley BART and then take BART back in the evening. That surely affects train use with heavier rides outbound at the end of the day.
2) Pricing features may affect symmetry. In NYC, the free bus-to-subway transfer means that sometimes people take a different route home as there isn’t an extra fare to stop en route. There was a NYT article about this some time ago (found it: http://www.nytimes.com/1999/10/08/nyregion/nyc-metrocard-robs-pockets-of-a-jingle.html ) where this effect is described with some humorous anecdotes. Or in other systems, where some kinds of transfer are free in a window can affect how people plan their errands and routes.
3) Crowding can also play a role in mode choice. In NYC, there are people who take the MTA ExpressBus home, even though it costs more and can get stuck in traffic, just because you have your own seat, and can get signal generally on the way home. They’d rather be stuck in traffic browsing the net rather than standing in a subway car making progress home.
4) Value of punctuality can also affect symmetry- many people need to be at work at certain time, and value being on time in that direction than getting home, where often there isn’t the same concern. That is, a lower standard deviation/higher mean mode may be preferred for a morning commute, whereas a higher std dev/lower mean mode may be preferred going home.
5) Pleasantness and weather as well can play a role in asymmetry. There are people in Hoboken who will always take the PATH train in to NYC in the morning, but on a nice day will spring for the extra cost of a ferry if it is a nice day or if they are with friends (see #6.)
6) Social aspects can also play a role, alluded to above. The chance that you will be on the same train or bus as some friends inbound is generally low, but if you are leaving work at about the same time as some people whose company you enjoy, the benefit of pleasant fellow-travelling may outweigh some added cost or time.
In general, people who are doing the same thing every day really do optimize in some interesting directions (e.g. prewalking), so I think it is reasonable to consider some of these effects. As you mention, though, these tend to be at the margin rather than being dominant terms.
Don’t forget, even regular bikes can go on the bus or train in most cities (if not New York at rush hour), and folding bikes are almost always allowed. This makes it possible to take transit in the morning, and then ride home in the evening (or vis-a-versa). Here in Portland, I ride my bike with my daughter to daycare, take the train to work (with my bike on board) and then ride back in the evening. This lets me stop at a store on the way back, and I don’t have to worry about changing clothing if it is raining, or morning frost or snow in winter (which usually doesn’t last past noon, in these part). The two other bike commuters at my office also take transit – they put their bikes on the bus.
Which reminds me of this great video: http://www.youtube.com/watch?v=boiQKDHq7nk “I Put My Bike On The Bus”
For transit and biking, bikeshare systems introduce some asymmetry as well. The biggest benefit to bikeshare from my use of it in DC is the ability to use a bike for a one way trip and not have to worry about a folding bike or secure parking at all.
Pricing and fare schemes might cause asymmetries as well. EM mentioned mode transfers.
In Europe, many countries have cities or regions with discount card or subscription plans that give steep discounts outside peak time.
That is certainly the case here in the Netherlands, for instance, where I am currently living. Most people who use trains on a regular basis have discount cards that reduce train (but not bus or tram or subway) fares by 40% off-peak. So depending on schedule, some people might stick with buses or trams for the outward journey and trains the other way back.
Other issue to be considered is asymmetrical crowding. When I lived in Milan, Italy, and studied fairly away from home, I usually had the option of a slow + even slower tram combination line that originated on the stop closest to home, or a faster subway+not that slow tram (with a bit more walking on one of the extremities than the tram) combination. The subway was usually full on my outward journey since that line came from some dense housing areas on the outskirts of city. But that was not the case on my inward journey, since the station near my school was on a less used line and the tram-subway transfer station always had subway trains coming in with plenty of space and seats.
==========================
Comfortable shelter on transfer points is a complicated issue. You want them to be comfortable enough for transfer while designing them not to become rendezvous for homeless people or meeting point for bored youths or shoddy characters.
@Joseph E: carrying bikes on buses or tams without surcharges is only possible when the number of cyclists is exceedingly slow to interfere with the comfort of other users or schedule adherence. It cannot be a working model for cities where the share of cycling commute trips is really large, especially on short-spaced services like subways or trams.
Of all possible solutions, the front bike racks on buses are the most egregious ones, as they compromise not only operations, but safety, in order to provide an expectation of convenience for people who are using bikes by choice (contrary to those bound to wheelchairs for being not able to walk, for instance).
I meant *low not slow.
It does depend on the bike ~ you can come close to filling a bus or train with 20″ folders in bolsa bags, without any substantial impact.
It was when I was using a folder to bike commute to the downtown of a regional city in Australia that I had the most assymetric commutes ~ in by bike, home by bus, in by bike, home by taxi ~ most due to occasional inclement weather.
But then, as a primary bike commuter, with transit as a back-up, my ridership would have been down amongst the rounding error in the modeling.
Front bike racks are very much not scalable. We’ve seen in many places that they only work if there are really very few people using them.
I have a couple more situations that are slight variations of those already mentioned.
1. Variable morning/afternoon weather. This applies to hot arid cities like Sacramento where in the summer the morning temperature can be up to 40 degrees cooler than afternoon (ie. 60 f at 7:00 AM to 100 f at 5:00 PM). This would motivate some people to ride their bikes or walk in the morning, but take transit on the way home.
2. When I lived in Bay Ridge, I was two Avenue blocks from the R train. I always walked in the morning, because the bus varied by up to 10 minutes in when it showed up, plus I was just starting my day and had plenty of energy. On the way home, there were some differences. First, the asymmetry of loading/unloading. While in the morning, some 100-200 people amble into the subway station at various times over an 8 minute period to catch the next R train, in the afternoon they all spill out onto the street all at once. If a bus happens to show up at that time, people start to get on and now it is held for up to 5 minutes as 30-40 more people get on and pay fares. Therefore there was a higher likelihood of a bus just being there. Second, exhaustion. People who live in Bay Ridge commute one hour or more to their jobs, so anything that cuts out even 5 minutes of walking can be welcomed at the end of the day. As a result, I sometimes took the bus two blocks if I was sufficiently exhausted and it happened to be there anyway. From what I could tell there were quite a few of us opportunistic bus riders, even a good dozen or so who took the bus just one block.
Note that if the morning peak is more compressed than the evening peak, as it often is, asymmetries that boost total evening peak ridership can be a good thing. In that situation, its asymmetries that add to morning peak congestion without yielding the return fare in the evening that are more problematic.
I’ve had many asymmetric transit experiences, the biggest cause for me might be boredom. When I lived in Flushing there were many bus/train combos I could use.
One asymmetry factor was weather. In the morning, I could take a bus from my house to Queens Blvd. to get the train. During summer, I would often take the 7 home to Shea Stadium and walk across Flushing Meadows to get home (avoiding the bus portion of my trip). In winter when it was cold and dark I would not walk across the park and use one of my subway>bus options to get home.
In Providence, I think there is less asymmetry of routes as there is little redundancy, but there is certainly more asymmetry of stops with people perhaps willing to get off a bus in Kennedy Plaza and walk off to their destination, but not willing to wait in Kennedy Plaza on the way home. When at RISD I would ask people their routes and tell them where they could wait outside the Plaza if it would help convince them to try RIPTA.
I did my Masters Thesis on Travel Behavior and through my research I found many examples of asymmetric mode choice. Overall, I find that understanding type of activities the travel is serving can help explain some of these behaviours.
First, we need to consider what the requirements of the trip are. For instance, I found that in many car-less households, they would take transit to the grocery store, but take a cab home because the trip required them to transport groceries.
Second, we need to consider the time constraints of the activity (i.e. if they have set start and end times). A good example of this would be a doctor’s appointment, which has a set start time, but there is no set time when the person has to get home. In this case, they may choose to get dropped off at the appointment by a family member, because the transit schedule won’t get them there on time, but they will take transit home, because waiting at the bus stop for 20 min going home won’t make them late for anything.
Third, we should understand the duration of an activity and whether it is predictable or not. One example I had was a lady who took the bus to Walmart to go shopping, but walked home (which doesn’t make sense unless we understand the nature of the activity itself). The reason was that she could coordinate her trip to Walmart with the transit schedule (she had flexibility when she could leave the house), but she couldn’t control how long she was shopping. Therefore, she walked home.
Just some additional thoughts.
It seems to me that this phenomenon in transit riding behavior is, in the modeling sense at least, perhaps more accurately characterized as asymmetrical transit route choice rather than asymmetrical mode choice. On the highway travel modeling side of things, this type of behavior would be like taking an expressway to work in the morning but taking the local streets back home in the evening. It’s not so much the transit mode that is determining the choice, but the routing and operational characteristics of the particular transit services, whatever transit modes they are.
Then again, there are some very interesting modal implications of this discussion, but they seem to boil down to the effects and constraits of “modal encumberance.” As noted in several of the comments above, bike-share riders actually behave more like transit riders and cab users (and hearty pedestrians) in the sense that they’re essentially unencumbered by mode. Catching a ride with a friend might also fall into this category. Bicycle owners, on the other hand, are more like motorists (and car-share users). You either need to take the bike/car back home or find a place to store it and retrieve it later.
Even if people travel by the same mode, work-to-home trips tend to be more widely distributed by time than home-to-work, so crowding can be lessened anyway.
After all, CBDs are often good entertainment districts, so lots of people go out after work.