Demand Model
From TriboroRX
The typical transportation demand model includes four steps -- trip generation, trip distribution, mode choice, and trip assignment (or route choice). In other words, how many people traveling to where are using what kind of transportation on which route? To estimate how many commuters would use the Triboro RX, we have constructed a crude version of this kind of model for the New York City subway system, and tested the addition of the Triboro RX to the network.
Contents |
[edit] Trip Generation, Trip Distribution, and Mode Choice
Using the the Journey-to-Work tables from the 2000 US Census takes care of the first three elements of the model, telling us how many people commute between each pair of census tracts by each mode of transport. However, because we have only modeled walk access to transit stations, we cannot study commuters who live or work beyond a reasonable walking distance from a subway station. Those commuters who might ride a bus, drive a car, or otherwise travel a substantial distance to get to or from the subway are unaccounted for entirely.
[edit] Census Tracts
The basic geographic unit of analysis used is census tracts, which are designed by the US Census to vary in physical size but hold similar numbers of people. It is assumed that the whole population of a census tract is served by a station if half the tract's area is within one half of a mile from that station. The map below shows the census tracts included in our study with blue indicating access to the existing subway system, and red indicating access to the Triboro RX.
[edit] Commuters
Our model considers only commutes that both begin and end within one of the served census tracts. The following map show all commutes that begin and end within the four boroughs at their origins and destinations, respectively. Commuters whose trip begins and/or ends in an unshaded area do not factor into any of our estimates. These represent a significant number -- 22% of subway commutes and 38% of all commutes that begin and end within the four boroughs. For this reason, we believe that our model yields fundamentally conservative estimates of Triboro RX ridership.
| All Commutes | Subway Commutes | |
|---|---|---|
| Total | 2,642,492 | 1,197,623 |
| Served* | 1,648,152 (62%) | 936,547 (78%) |
| Unserved | 994,339 (38%) | 261,075 (22%) |
* Served: census tracts that have at least 1/2 of area within 1/2 mile of a station
[edit] Route Choice
Given the complexity of the New York City subway system, there exist multiple choices of routes among many origins and destinations. To compare these routes from home to work, and to compare new routes created by the addition of the Triboro RX, we have selected the generalized cost function described below. The cost function was designed to minimize the number of inputs (for simplicity's sake) while accounting for different parts of a rider's experience -- time spent riding the subway, waiting for trains, and transferring between trains.
[edit] Generalized Cost Function
The fundamental unit used in this cost model is seconds. However, not every second is created equal -- time spent riding a train is experienced differently by riders than time spent waiting for a train.
- SPEED = 18.3 Miles per Hour
- Assume a uniform speed of 18.3 mph for all trains in the system. [1]
- DWELL_TIME = 10 seconds
- The time that trains spend in the station while passengers board and alight. MTA/NYC Transit policy is that dwell time should be as close to 10 seconds as possible [2]. Also represents the time it takes a passenger to board or alight a train.
- TRAVEL_LINK = [straight-line distance between 2 stations] / SPEED + DWELL_TIME
- The total time spent riding a train from one station to the next, including dwelling (or alighting) at the second station. We do not account for the geometry of actual subway tracks, but rather assume straight-line paths between successive stations.
- WAIT_TIME = 120 seconds
- Rather than account for complicated train schedules, we assume a consistent 4 minute headway between trains throughout the system. Expected waiting time is 1/2 of headway, in this case 2 minutes.
- TRANSFER_TIME = 120 seconds
- Assume that, on average, it takes 2 minutes to navigate from one platform to another when changing trains.
- PLATFORM_TIME = 0 seconds
- Transfers between trains on the same platform do not incur any excess travel time.
- WEIGHT = 1.75
- A coefficient to represent the extra pain experienced by riders for time spent not traveling (i.e. waiting or transferring) [3]
- PLATFORM_LINK = WEIGHT * WAIT_TIME + DWELL_TIME
- Total cost of changing trains on the same platform, including boarding the next train.
- TRANSFER_LINK = WEIGHT * (TRANSFER_TIME + WAIT_TIME) + DWELL_TIME
- Total cost of changing trains on the different platforms, including boarding the next train.
TOTAL COST = [# platform transfers] * PLATFORM_LINK + [# walking transfers] * TRANSFER_LINK + [sum of TRAVEL_LINK for each segment ridden]
As mentioned above, this cost function does not, for simplicity's sake, include access time (i.e. at the origin traveling from home to the station entrance and through the station to the platform, or at the destination traveling from the platform to the exit and from there to work). Neither does it include waiting for and boarding the initial train, because under this model those are constant terms for every trip and add no useful information for comparisons.
[edit] Adding the Triboro RX
Our model predicts three different groups of people who would ride the Triboro RX:
- Subway commuters who live and work in census tracts currently served by the subway, and whose commutes will be made more efficient by the Triboro RX
- Subway commuters whose trips are not currently served on one end by the subway, but will be served on that end by the Triboro RX
- Non-subway commuters who will switch modes in response to the additional convenience resulting from the Triboro RX
To assess the first group, we start by applying the cost function to the current subway network to determine the lowest cost route between each pair of tracts that are served by the system. For each commuter flow served by one of these routes, we assign all subway riders from that flow to the corresponding route. Next, we repeat the process but with the Triboro RX added to the network. The members of this first group are those commuters assigned a route in both scenarios, and whose trip in the second scenario includes at least one leg on the Triboro RX.
The second group is identified as those subway commuters who are successfully assigned a route after the Triboro RX is added to the network but not before. Although these people do use the subway as it exists today, we believe it is reasonable to assume they will use the Triboro RX without comparing it to their current trip because of how far they live or work from an existing subway station.
The third group, new riders, is estimated by accounting for the effect that increased accessibility and mobility would have on the percentages of people using the subway for different kinds of trips, and working backwards from there. Specifically, we use our model of the existing system to determine the current subway mode share for trips between each pair of boroughs and by number of transfers (shown in the table below). These fractions are then applied, with the Triboro RX included in the network, to commuter flows between the corresponding boroughs with the same number of transfers (which in many cases will have decreased because of the Triboro RX). When this fraction is higher than the current fraction, the difference is counted as new riders who have converted from other modes of transport.
Please see the Results section for a detailed presentation of what actual numbers are produced by this methodology.
[edit] Subway Mode Share, by Borough-Pair Flows, by Number of Seats (Before Triboro RX)
| Mode Share (and # Trips) by # Seats* | |||||
|---|---|---|---|---|---|
| Origin | Destination | 1 Seat | 2 Seat | 3 Seat | 4 Seat |
| Bronx | Bronx | 22.6% (11391) | 22.1% (4033) | 16.5% (333) | 86.2% (25) |
| Bronx | Brooklyn | 65.6% (4165) | 64.3% (2677) | 55.1% (161) | 0.0% (0) |
| Bronx | Manhattan | 74.6% (70953) | 63.5% (7604) | 28.1% (402) | 0.0% (0) |
| Bronx | Queens | 64.1% (453) | 52.8% (3510) | 31.8% (78) | 0.0% (0) |
| Brooklyn | Bronx | 65.4% (1795) | 51.4% (1779) | 44.3% (122) | 0.0% (0) |
| Brooklyn | Brooklyn | 32.6% (52555) | 26.2% (19083) | 24.2% (3055) | 16.0% (467) |
| Brooklyn | Manhattan | 84.4% (192799) | 81.1% (40693) | 89.1% (616) | 0.0% (0) |
| Brooklyn | Queens | 45.4% (4208) | 46.1% (5485) | 44.4% (1734) | 34.7% (328) |
| Manhattan | Bronx | 48.2% (5310) | 35.6% (1041) | 22.5% (237) | 0.0% (0) |
| Manhattan | Brooklyn | 71.2% (10848) | 58.4% (3034) | 47.5% (76) | 0.0% (0) |
| Manhattan | Manhattan | 54.2% (233905) | 50.4% (28671) | 40.9% (370) | 0.0% (0) |
| Manhattan | Queens | 67.4% (3607) | 61.9% (3241) | 41.8% (49) | 0.0% (0) |
| Queens | Bronx | 44.7% (197) | 39.3% (1717) | 52.5% (176) | 0.0% (0) |
| Queens | Brooklyn | 44.7% (6628) | 47.3% (6818) | 36.6% (1515) | 31.2% (200) |
| Queens | Manhattan | 85.5% (111808) | 79.1% (44784) | 62.4% (571) | 0.0% (0) |
| Queens | Queens | 32.2% (15470) | 30.4% (6636) | 25.5% (2499) | 25.9% (1235) |
* Number of seats is simply the number of transfers plus one (i.e. a trip with no transfers is a "one-seat ride," with one transfer is a "two-seat ride," etc). For this part of the analysis we have not counted same-platform transfers as an extra seat, indicating a belief that transfers that do not require extra walking or stairs do not affect peoples' decisions about whether or not to use the subway.
[edit] Caveats
more specific detail?
