1. Hysteresis &
Urban Rail
The effects of past urban rail on
current residential & travel choices
David Block-Schachter
October 15, 2012

2. 1. Intro
2. History
3. Ideas
4. Hypotheses & mechanisms
5. Methods
6. Evidence
Agenda

3. Ideas
• Heterogeneous response
• Buffalo
• Delorean, iran hostage crisis, soviet union, mtv, mubarak, AIDS
• Things accumulate on top of intentions and accidents
Predicting the future is hard

4. Ideas
Historical
vision
Qualita0ve
Quan0ta0ve
predict
&
model
Ahistorical
Using history as guide

5. History
• The tape of history is played uniquely forward (Gould)
• Ports & harbors
• Factories & housing
• Roads & bridges
• More buildings
• Streetcars
• More buildings
• And so on
Emergence of
cities

6. Pre-history

7. Fill and wharfing out

8. Post WWII

9. History
• Post roads 1673 New York à Springfield à Boston
(washington street roxbury silver line )
• Turnpikes turn, pike. Middlesex (hampshire st) chartered 1805, free 1846
25 cents per vehicle + 4 cents per man or horse
• Bridges tolls to free
• Omnibus 1793 stagecoach over west boston bridge
• Horsecar 1856 central square across west boston bridge to bowdoin
• Steam Rail 1830 Boston & Lowell – later Green Line D branch
• Streetcar 1889 electrified in Allston
• Subway 1897 “first”
• Elevated 1901 last “major” american city
Transport history primer

11. History
Source: Binford

12. History
Source: Binford

14. Source: BPL

15. History

16. History

17. History

19. History

20. Methods
Era Speed Fares Reach Competition Urban trends
1865 Horsecar + Variable + - Immigration
1925 Streetcar+ ++ Standard +++ + Expansion
1960 Consolidation +++ + ++ Flight
2000 Expansion +++ + +++ Gentrification
4 eras

21. History
1865 - Horsecar

22. History
1925 – Streetcar, Elevated & Subway

23. History

24. History
1960 – Heavy and Light Rail

25. History
2000 - Heavy and Light Rail

26. East Cambridge Bridge, 1912 (Detroit Publishing Company)

27. Harvard Square, 1912 (Detroit Publishing Company)

28. History
12
80%
x
100000
Growth
rate
Popula0on
70%
10
60%
8
50%
6
40%
30%
4
20%
2
10%
-­‐
0%
1810
1820
1830
1840
1850
1860
1870
1880
1890
1900
1910
1920
1930
1940
1950
Fares, frequency, transfers, ubiquity
Effective 50% decrease due to inflation & free transfers from system consolidation
Concurrent inner ring growth
Boston, Brookline, Cambridge, Chelsea, Somerville
Horsecars v. streetcars

29. Ideas
This is why history matters
Source: HowStuffWorks.com

30. Ideas
• It is not possible to predict the next step without knowing the history of the system
• Iron ore retains magnetization after magnetic field removed
• Urban rail is magnet. Built environment is iron ore
Hysteresis
Source: HowStuffWorks.com

31. Ideas
Source: Wikipedia

32. Ideas
• Self evident: current options determined by past choices
In transportation
• Durable capital—long-lived residential, industrial, and commercial buildings—is an
order of magnitude greater than the public or private investment in transportation
infrastructure
• Changing the transport path requires significant incremental value.
• And there are coordination problems
Path dependence

33. Hypotheses
Historical
• Exploring one historical process – this
the evolution of urban rail from single vision
cars pulled by horses, to those powered
by electricity, and eventually multi-car Qualita0ve
Quan0ta0ve
trains running on elevated, surface, and
underground tracks
predict
&
• In one city – Boston – over an extended model
Ahistorical
time period – from 1865 to the present
• Examine impacts of proximity to rail on residential density and travel (auto
ownership, mode choice)
• HYP. Urban rail has permanent direct and indirect effects on the geography of
density and behavior over exceptionally long time frames, and that these effects
outlast the urban rail itself—they are persistent and hysteretic.
What is this work about?

34. Ideas
We don’t do it often
• Capital is durable
• Land Use and transport inter-related

35. Hypotheses
Strong Density
Moderate Auto ownership
Minimal Mode choice
Proximity to past rail
influences current behavior

36. Hypotheses
• Influences present rail and bus location
• Behavior persists
• Built environment: direct or via preferences / attitudes
• Culture: family hist., neighbor pressure, pos. externalities, BE
• Municipal action: zoning, N/IMBY, political power
• AND/OR proxy for omitted and unique
Why?

37. Hypotheses
Mechanisms

38. Hypotheses
• Cumulative causation
• Durable capital built around infrastructure
• Useful life, staged development, capitalizing access
• Legal and institutional rationales for re-use
à Hard to make new paths: incremental value
à Easy to serve existing market
Mech: Infrastructure

39. Hypotheses
Habitual travel choice (Garling)
+ Transit agency incentive to replicate or add frequency on new
mode
= Existing riders may not revisit mode choice
à Reduced effect over time
Mech: Behavioral persistence

40. Hypotheses
Direct. Past rail affects quality of BE (connectivity é routes)
BUT objective v. perceived (Gim); indicators (Crane, Lee)
à Changes attributes of residential and travel behavior
Indirect. BE as mnemonic device retaining signs and symbols
associated with use of rail
à Influences weighting of existing attributes
Mech: Built environment

41. Hypotheses
Heuristic that simplifies complex decision making
• Property of BE (unique to places near past rail)
• Result of historical travel behavior
Perceptions of behavioral control (Ajzen)
• Past rail à more usage in past
• Slow migration + habitual choice à more usage today
Mech: Culture

42. Hypotheses
• Planning horizon too short by a century
• Locating growth and local incentives
• Urban “renewal” was even worse than previously thought
• Multimodal complements to enlarge growth effects of rail
If I’m right

43. Prove it

44. Methods
HYP: Proximity to past rail influences current behavior
MECHANISM: Plausible direct and indirect effects:
rail persistence, demographics, BE, culture
Behaviort = f ( BEt−1, Demographict−1, RailAccesst−n , RailAccesst , Exogenous)
Proximity to. Buffer around routes. Lack of stop, frequency
information, consistency between eras
Past rail. Horsecars (1865), streetcars (1925), pre-MBTA (1960)
Influences. Has a statistically significant effect on
Current behavior. Density, auto ownership, mode choice
Basics

45. Methods
Tract level for density, auto ownership, mode choice
• Spatial error model with adjacency matrix corrects for violation of OLS errors due
to misfit to tract facets
• Sensitivity testing for matrix, correction
Multinomial logit for household auto ownership
• v. ordered logit / probit
Multinomial logit for individual mode choice
• All trips (not just JTW)
• Individual panel structure
• Origin & destination attributes
• Home location (restriction to non-home trips)
• Validation of VOT against CTPS published estimates
Methods

46. Methods
• IPUMS 1860, 1930
• Maps, turned into vectors
• Census 1960-2000, UTPP, CTPP
Data

47. Evidence

49. Evidence
CHAPTER 4. NETWORK PATHS 11
1925 1960 2000
1865 0.89 0.88 0.87
1925 0.92 0.88
1960 0.92
n= 2,210 blockgroups
CHAPTER Table 4.1: Access correlation by actual distance to rail
4. NETWORK PATHS 13
modern rail makes use of the sameEstimate way as older Estimate
right of Std. incarnations of urban
rail but does not use all of these historical paths. The test for this hypothesis is
Intercept 0.352 ***
the degree to which the modern city is composed of areas that (a) always***
Dist. to 1960 Rail (d1960) 0.283 0.578 had
accessto 1925 Rail never had access to rail, and (c) had access in given period,
Dist. to rail, (b) (d1925) 0.177 0.127 ***
Dist. to others. If (d1865)
but not 1865 Rail those areas that had urban rail access in 2000 also had the
0.233 0.326 ***
d1960level of access during prior periods, then this hypothesis is verified. **
same * d1925 0.017 0.043
d1960 * d1865seen in Table 4.2, the strength of findings depends on how access
As can be 0.052 0.192 *** • High degree of correlation
d1925 is defined. At the blockgroup level, 83% (one minus the last column) of
to rail * d1865 -0.031 -0.192 ***
d1960 areas with access to urban rail in 2000 had similar access to urban rail in
those * d1925 * d1865 -0.002 -0.048 ***
between access in eras
previous periods—when access is defined by a radius 1/10 mile from the centroid
dependent: distance to urban rail, 2000; r = 0.876, n = 2,210 blockgroups
2
of each blockgroup. By 1/2 miles, 96% of areas that had access in rail in 2000
‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1
also had it in previous periods. Furthermore, beyond 1/10 mile, current access to • Not entirely mediated by
rail is more dependent on access to rail in periods prior to 1960—approximately
Table 4.3: Relative effects of distance to rail: 1865-1960 on 2000 access in most recent era
the advent of the modern system in Boston—than it is to access to rail in 1960.
The implication of these findings is that the visual inspection also rings true
Findings: infrastructure
Estimate Std. Estimate
from a statistical perspective. Urban rail extensions in the past 4 decades have
Intercept to areas that had 0.362 to rail prior to 1960, but had lost that access
largely been access ***
Dist. to 1865 Rail
by 1960. These conclusions are not 0.994 merely the products of spatial definition.
0.893 ***
dependent:rail in theto urban rail, in Table 4.2 is also strongly blockgroups on where
Access to distance modern era 1925; r2 = 0.797, n = 2,210 dependent
‘***’existed in prior‘*’ 0.05 ‘.’on an area and population basis.
rail 0.001 ‘**’ 0.01 periods 0.1
Table 4.4: Relative effects of access to horsecars on access to streetcars
4.3 Permanence to persistence
4.4 Section explores the relative weight that access
This Buses in prior periods has on
subsequent access. The main method used is regression of distance to rail in
Left unexamined thus far is the role of the motor bus in replacing the streetcar.
period t on periods t 1 through t n. If rail is permanent but not persistent,
Between 1925 and 1960 the entirety of surface running routes in Boston were
previous periods (t 2 through t n) effects on the present (t) are mediated
removed, save for the branches of the (not-yet-named) Green Line. Did bus
through access in the most recent period (t 1). Access in periods t 2 through
routes fill the void in access left by this massive change? Or was the planing
t n will be insignificant. If rail is both permanent and persistent, access to
paradigm for bus coverage un-moored from dependence on past routes by the
urban rail in periods t 2 through t n will also be significant. There will be a
same forces that removed the fixed streetcar guideways?
measurable effect on where new rail is placed based on historical rail corridors.
Some anecdotal evidence points to bus routes as largely consistent with
As is evident in Table 4.3, 1960 plays the largest role in determining access

50. Evidence
• The monocentric model implies a specific functional form—negative exponential—
that results from regressing the natural logarithm of density on distance from the
CBD. Access to rail in each period is an additional binary regressor. If the coefficient
on rail proximity in past periods is significant and positive, the associated density
gradient is taller.
Density: theory

51. Evidence
20,000
2000 Density (ppl. / sq. mi.)
18,000
16,000
14,000
12,000
10,000
8,000
6,000
4,000
2,000
-
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Dist to CBD (mi.)
Actual w.out 1925 Rail (endogenous network)
w.out 1925 Rail (exogenous network)
Density comparisons

52. Evidence
Density gradients

53. Evidence
Auto ownership models

54. Evidence
Mode share models

55. Evidence
(1) Past access to rail à density > auto ownership, but both significant.
Mode share not significantly influenced by past access to rail after controlling for
current access.
(2) Demographic and built environment controls, as well as controls for additional
causal mechanisms reduce the measured effect, but do not eliminate it.
(3) Density + auto ownership: past rail > current
Mode share vice versa
Summary

56. Evidence
• Travel behavior lit: behavior is a function of the current attributes of the person and
the environment.
• This finds: past existence of rail is an indicator both of some omitted
characteristic(s) of the BE unique to those places that once had rail and a cultural
inheritance, but the mixture of the two is unknown.
v. literature

57. Evidence
Use history
• Efforts to understand how the history and present of a place may influence its
future, for time frames beyond BCA or design charette
Where and when to build within cities
• Costs of (re)development, neighborhood opposition
• Mechanisms for patience over long time frames
Planning timeframes and goals
• Scenario planning, built environment endowments
Cultural interventions
• Local policies to support national goals; direct and indirect
Why it matters

58. Thank you