-
Notifications
You must be signed in to change notification settings - Fork 0
/
A_Comparative_Illustration_of_Trip-_and_Activity-Based_Modeling_Techniuqes.tex
3544 lines (2913 loc) · 159 KB
/
A_Comparative_Illustration_of_Trip-_and_Activity-Based_Modeling_Techniuqes.tex
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
\documentclass[fancy, twoside, mastersfancy, ms]{byuthesis}
\usepackage{bookmark}
\providecommand{\tightlist}{%
\setlength{\itemsep}{0pt}\setlength{\parskip}{0pt}}\usepackage{longtable,booktabs,array}
\usepackage{calc} % for calculating minipage widths
% Correct order of tables after \paragraph or \subparagraph
\usepackage{etoolbox}
\makeatletter
\patchcmd\longtable{\par}{\if@noskipsec\mbox{}\fi\par}{}{}
\makeatother
% Allow footnotes in longtable head/foot
\IfFileExists{footnotehyper.sty}{\usepackage{footnotehyper}}{\usepackage{footnote}}
\makesavenoteenv{longtable}
\usepackage{graphicx}
\makeatletter
\def\maxwidth{\ifdim\Gin@nat@width>\linewidth\linewidth\else\Gin@nat@width\fi}
\def\maxheight{\ifdim\Gin@nat@height>\textheight\textheight\else\Gin@nat@height\fi}
\makeatother
% Scale images if necessary, so that they will not overflow the page
% margins by default, and it is still possible to overwrite the defaults
% using explicit options in \includegraphics[width, height, ...]{}
\setkeys{Gin}{width=\maxwidth,height=\maxheight,keepaspectratio}
% Set default figure placement to htbp
\makeatletter
\def\fps@figure{htbp}
\makeatother
% definitions for citeproc citations
\NewDocumentCommand\citeproctext{}{}
\NewDocumentCommand\citeproc{mm}{%
\begingroup\def\citeproctext{#2}\cite{#1}\endgroup}
\makeatletter
% allow citations to break across lines
\let\@cite@ofmt\@firstofone
% avoid brackets around text for \cite:
\def\@biblabel#1{}
\def\@cite#1#2{{#1\if@tempswa , #2\fi}}
\makeatother
\newlength{\cslhangindent}
\setlength{\cslhangindent}{1.5em}
\newlength{\csllabelwidth}
\setlength{\csllabelwidth}{3em}
\newenvironment{CSLReferences}[2] % #1 hanging-indent, #2 entry-spacing
{\begin{list}{}{%
\setlength{\itemindent}{0pt}
\setlength{\leftmargin}{0pt}
\setlength{\parsep}{0pt}
% turn on hanging indent if param 1 is 1
\ifodd #1
\setlength{\leftmargin}{\cslhangindent}
\setlength{\itemindent}{-1\cslhangindent}
\fi
% set entry spacing
\setlength{\itemsep}{#2\baselineskip}}}
{\end{list}}
\usepackage{calc}
\newcommand{\CSLBlock}[1]{\hfill\break\parbox[t]{\linewidth}{\strut\ignorespaces#1\strut}}
\newcommand{\CSLLeftMargin}[1]{\parbox[t]{\csllabelwidth}{\strut#1\strut}}
\newcommand{\CSLRightInline}[1]{\parbox[t]{\linewidth - \csllabelwidth}{\strut#1\strut}}
\newcommand{\CSLIndent}[1]{\hspace{\cslhangindent}#1}
\usepackage{flafter}
\usepackage{float}
\usepackage[section]{placeins}
\floatplacement{table}{htbp}
\hyphenation{Front-Run-ner}
\usepackage{microtype}
\usepackage[all]{nowidow}
\usepackage{booktabs}
\usepackage{longtable}
\usepackage{array}
\usepackage{multirow}
\usepackage{wrapfig}
\usepackage{float}
\usepackage{colortbl}
\usepackage{pdflscape}
\usepackage{tabu}
\usepackage{threeparttable}
\usepackage{threeparttablex}
\usepackage[normalem]{ulem}
\usepackage{makecell}
\usepackage{xcolor}
\usepackage{siunitx}
\usepackage{booktabs}
\usepackage{longtable}
\usepackage{array}
\usepackage{multirow}
\usepackage{wrapfig}
\usepackage{float}
\usepackage{colortbl}
\usepackage{pdflscape}
\usepackage{tabu}
\usepackage{threeparttable}
\usepackage{threeparttablex}
\usepackage[normalem]{ulem}
\usepackage[utf8]{inputenc}
\usepackage{makecell}
\usepackage{xcolor}
\makeatletter
\@ifpackageloaded{bookmark}{}{\usepackage{bookmark}}
\makeatother
\makeatletter
\@ifpackageloaded{caption}{}{\usepackage{caption}}
\AtBeginDocument{%
\ifdefined\contentsname
\renewcommand*\contentsname{Table of contents}
\else
\newcommand\contentsname{Table of contents}
\fi
\ifdefined\listfigurename
\renewcommand*\listfigurename{List of Figures}
\else
\newcommand\listfigurename{List of Figures}
\fi
\ifdefined\listtablename
\renewcommand*\listtablename{List of Tables}
\else
\newcommand\listtablename{List of Tables}
\fi
\ifdefined\figurename
\renewcommand*\figurename{Figure}
\else
\newcommand\figurename{Figure}
\fi
\ifdefined\tablename
\renewcommand*\tablename{Table}
\else
\newcommand\tablename{Table}
\fi
}
\@ifpackageloaded{float}{}{\usepackage{float}}
\floatstyle{ruled}
\@ifundefined{c@chapter}{\newfloat{codelisting}{h}{lop}}{\newfloat{codelisting}{h}{lop}[chapter]}
\floatname{codelisting}{Listing}
\newcommand*\listoflistings{\listof{codelisting}{List of Listings}}
\makeatother
\makeatletter
\makeatother
\makeatletter
\@ifpackageloaded{caption}{}{\usepackage{caption}}
\@ifpackageloaded{subcaption}{}{\usepackage{subcaption}}
\makeatother
\title{A Comparative Illustration of Trip- and\\
Activity-Based Modeling Techniques}
\author{Hayden Atchley}
%\author{Hayden AtchleyKamryn MansfieldGregory S. Macfarlane}
% On the custom title page, use the same title, but format as you like
\customtitle{A Comparative Illustration of\\
Trip- and Activity-Based\\
Modeling Techniques}
% This is the date of graduation
\date{27 May 2024}
% If your degree is not a PhD or MS, then you can overwrite the degree using
% the \degree command: \degree{Bachelors of Basics}
% Your department
\department{Civil and Construction Engineering}
% The names of your committee members
\committeechair{Gregory S. Macfarlane}
\committeemember{Grant G. Schultz}
\committeemember{Gustavious P. Williams}
\keywords{
travel demand model; activity-based model;
ActivitySim
}
\AtBeginDocument{%
\ifdefined\contentsname
\renewcommand*\contentsname{Table of Contents}
\else
\newcommand\contentsname{Table of Contents}
\fi
}
\begin{document}
\frontmatter
\titlepage
\cleardoublepage
\customtitlepage
\cleardoublepage
\begin{abstract}
Activity-based travel demand models are generally considered superior to
their trip-based counterparts, as activity-based models (ABMs)
explicitly model individuals in contrast to the aggregate nature of
trip-based models. There have been a number of comparisons between trip-
and activity-based models, but these comparisons focus almost
exclusively on the technical ability of the two model types, while not
considering the practical benefits an ABM may or may not have to a
transportation agency. This research performs a more holistic comparison
between trip- and activity-based models, focused specifically on the
practical differences between model types, both in terms of usability
and capability for complex analysis. We use the existing Wasatch Front
model as a representative trip-based model, and an ActivitySim
implementation in the same area as a representative ABM. We create three
hypothetical scenarios in both models: a change in land use, an
improvement to commuter rail service, and an increase in remote work. We
discuss the process of creating each scenario in both models, and
perform several example analyses with each scenario and model. We find
that many commonly-cited reasons for the lack of ABM adoption may not be
as applicable as previously thought. ABMs are often considered more
complicated than trip-based models, requiring more data and
computational resources. While ABMs do require more input data, we found
that in our case the complexity of the model and the computational
resources required were similar between model types. Additionally, the
ABM allows for much more intuitive and straightforward interpretation of
results.
\end{abstract}
\cleardoublepage
\begin{acknowledgments}
I would like to acknowledge the Utah Department of Transportation for
providing funding for this research. Additionally, I would like to thank
the modeling teams at the Utah Department of Transportation, Wasatch
Front Regional Council, Mountainland Association of Governments, and
Fehr \& Peers for their help and input at several stages of this
project. I would specifically like to thank Chad Worthen and Chris Day
at Wasatch Front Regional Council for answering my questions about their
travel demand model. I would also like to thank my peers in the BYU
transportation lab for their friendship and support, and especially for
their help dealing with miscellaneous issues that arose throughout this
project. Lastly, I would like to especially thank my graduate advisor,
Greg Macfarlane, for his support and encouragement.
\end{acknowledgments}
\cleardoublepage
\phantomsection
\addcontentsline{toc}{chapter}{\contentsname}
\tableofcontents*
\cleardoublepage
\listoffigures
\cleardoublepage
\listoftables
\cleardoublepage
\bookmarksetup{startatroot}
\chapter*{List of Acronyms}\label{list-of-acronyms}
\addcontentsline{toc}{chapter}{List of Acronyms}
\markboth{List of Acronyms}{List of Acronyms}
\begin{description}
\tightlist
\item[\phantomsection\label{acronyms_ABM}{ABM}]
activity-based model
\item[\phantomsection\label{acronyms_ASC}{ASC}]
alternative-specific constant
\item[\phantomsection\label{acronyms_DAP}{DAP}]
daily activity pattern
\item[\phantomsection\label{acronyms_TAZ}{TAZ}]
transportation analysis zone
\item[\phantomsection\label{acronyms_WFRC}{WFRC}]
Wasatch Front Regional Council
\end{description}
\mainmatter
\bookmarksetup{startatroot}
\chapter{Introduction}\label{sec-introduction}
In travel demand modeling, activity-based models (ABMs) have been
championed by researchers and many practitioners as being theoretically
superior to the trip-based models historically used in transportation
planning efforts since the 1950s (Rasouli and Timmermans 2014). ABMs
explicitly model individuals, in contrast to the aggregate nature of
trip-based models, and so in theory are able to represent travel
behavior more accurately. Additionally, the focus on individuals in an
ABM can allow for more detailed post-hoc analysis of model outputs
compared to a trip-based model.
There have been a number of comparisons and case studies between trip-
and activity-based models (Ferdous et al. 2012; Mouw 2022; Zhong et al.
2015), but these comparisons focus almost exclusively on the technical
ability of the two model types. Though there are potential
\emph{theoretical} benefits to ABMs over trip-based models, there is
little discussion in the literature of the \emph{practical} benefits an
ABM has, if any. In fact, while trip-based models are almost ubiquitous
among transportation agencies, many agencies have delayed or declined to
transition to an ABM citing additional data requirements, staff
training, computational resources, and related concerns (Miller 2023).
In this research, we perform a more holistic comparison of ABMs to
trip-based models, with a particular focus on the practical
considerations an agency would need to make in transitioning to an ABM.
We additionally discuss the potential practical advantages regarding the
quality and characteristics of travel analyses that an ABM allows.
Though this research occasionally makes quantitative comparisons between
model types, we do not focus heavily on model \emph{accuracy} (either to
each other or to observed data), as this can be adjusted in any model
type through model calibration. Instead, this research seeks to
illustrate the differences between trip- and activity-based models in a
way that would be practically useful to an agency considering
transitioning to an ABM, noting potential pain points both in the
literature and in our experience in this research itself.
To compare the model types, we first identify three main goals of travel
demand modeling, which are to model travel behavior in response to
changes in land use, transportation infrastructure, and social/economic
factors. We then create three hypothetical model scenarios, one for each
goal identified. These scenarios are the addition of a new development,
an increase in commuter rail service, and an increase in remote work,
respectively. Each of these scenarios is created in both a trip-based
and activity-based model representing the Wasatch Front (Salt Lake City)
region of Utah, USA. We discuss the process of implementing each
scenario, as well as perform a variety of post-hoc analyses, for both
model types.
The document proceeds in a typical fashion: Chapter~\ref{sec-literature}
provides an overview of the literature discussing the differences
between trip-based models and ABMs, including the theoretical and
analytical benefits of each framework. Chapter~\ref{sec-methods} first
describes the models used in this research, namely the existing regional
trip-based model and an activity-based model constructed to support
research activities in the region; this section also describes the
scenarios designed to test the usefulness and applicability of the
different model frameworks. Chapters \ref{sec-landuse}--\ref{sec-wfh}
describe the findings from each scenario, alongside a discussion of
related limitations and implications. Chapter~\ref{sec-conclusions}
provides a summary of our findings and a discussion of our conclusions,
along with a set of recommendations.
\bookmarksetup{startatroot}
\chapter{Literature Review}\label{sec-literature}
Travel demand modeling in the modern sense has its origins in the
1950's, with the Chicago Area Transportation Study (Chicago Area
Transportation Study 1959) being one of the first urban planning studies
to use the now-ubiquitous ``four-step'' modeling framework (McNally
2007). Up to this point, most urban transportation planning used
existing demand or uniform-growth travel forecasts to model travel
demand, but the Chicago Study used a combination of trip generation,
trip distribution, modal split, and network assignment models to more
accurately represent travel behavior (Weiner 1997). Since then, there
have been numerous studies iterating on the ``four-step'' (more
appropriately termed ``trip-based'') framework, and trip-based models
are now the primary tool used in forecasting travel demand across the
United States (Park et al. 2020).
These trip-based models are not without problems, however. Rasouli and
Timmermans (2014) give several shortcomings of trip-based models. First,
they use several sub-models that are (implicitly or explicitly) assumed
independent, and this can result in a lack of consistency or integrity
between sub-models. For example, the assumed value of time in the mode
choice model might be radically different than the assumed value of time
in the tolling assignment model. Second, these models are strongly
aggregate in nature, which can cause significant aggregation bias with
high and low values excluded. Finally, they lack ``behavioral
realism''---that is, they do not have a concept of individuals making
decisions, which is what travel behavior actually is.
Jones (1979) proposed an alternative to the trip-based paradigm, namely
an ``activity-based'' framework that models travel behavior at an
individual rather than aggregate level. An ABM places the focus on
``activities'' rather than ``trips'' as the basic unit of analysis, and
predicts a sequence of activities for each individual and household,
with information such as activity location, start time, and duration,
using a high level of temporal and spatial granularity. ``Trips'' are
then journeys from one activity to the next (Pinjari and Bhat 2011). By
adopting this activity-centric framework, ABMs provide a more consistent
and comprehensive representation of travel behavior. They take into
account complex dependencies and interactions within the model as a
whole and at an individual level. ABMs acknowledge that travel choices
are not made in isolation, but rather influenced by the preceding
activities. This means that, for example, if an individual takes transit
to work, they will not be able to drive home. ABMs therefore attempt to
present a more conceptually accurate model of actual travel behavior
than traditional trip-based models.
Despite these advantages, many agencies have yet to adopt ABMs, and
instead continue to use trip-based models (Miller 2023). While ABMs may
be superior in certain aspects, they may also have disadvantages, such
as requiring more detailed input data and greater computational
resources. It is also not always clear if ABMs provide substantially
better forecasts than their trip-based counterparts, nor if this
tradeoff is worth the increased costs for every agency. This literature
review presents an overview of both modeling frameworks, and discusses
the advantages and disadvantages of using an ABM.
\section{Overview of Model Types}\label{overview-of-model-types}
Trip-based models are often referred to as ``four-step'' models due to
their four fundamental sub-models: trip generation, trip distribution,
mode choice, and network assignment (National Academies 2012 p. 28).
Models can be more complicated than these four steps, possibly including
integration with a land use forecast, iteration between mode and
destination choice, etc., but the ``four steps'' are the central
component of any of these models (McNally 2007).
In a typical trip-based model, travel demand is predicted based on
aggregate population data, often delineated by transportation analysis
zone (TAZ). Each sub-model relies on this aggregate data; for example,
the modal split sub-model will often use average TAZ income as an input
(National Academies 2012 p. 14). Many trip-based models include a
disaggregation step, where this aggregate data is segmented along
variables such as household size and vehicle ownership. Regardless of
the segmentation variables used in the first three model steps, the
resulting trip matrices by mode and time of day are then assigned to a
transportation network.
ABMs differ significantly from this approach. Rather than using
aggregate data, ABMs use data representing an actual or synthetic
population, with individual person and household data (Vovsha et al.
2005). These models use an activity or tour scheduler to assign a daily
activity pattern (DAP) of zero or more tours to each individual, where a
tour is a series of trips that begin and end at home. These DAPs are
restricted temporally, spatially, and modally; i.e., each person has a
logical and followable sequence of trips and activities (Bowman 1998). A
``drive alone'' trip from work to lunch, for example, cannot be made if
transit was taken to work. ABMs output a list of tours and trips by
person, time, location, and type, and these can then be assigned to a
transportation network in a similar manner as in a trip-based model. In
effect, an ABM replaces the first ``three'' steps of the traditional
``four-step'' approach.
\section{Comparison of Modeling
Frameworks}\label{comparison-of-modeling-frameworks}
In discussing the differences between ABMs and trip-based models, there
are really two comparisons that need to be made: how the population data
is structured, and how travel is organized. Trip-based models generally
use aggregate population data while ABMs use a synthetic population, and
trip-based models organize travel into trips while ABMs organize travel
into activities and tours. The following sections explain these aspects
of travel demand modeling and discuss the claimed advantages and
disadvantages of each model type.
\subsection{Population Data}\label{population-data}
The aggregate population data used in trip-based models can vary in
origin and level of detail, but the basic concept is the same: the study
area is organized into generally small zones, and certain demographic
and socioeconomic data is known or obtained for each zone (National
Academies 2012 p. 14). This includes data such as number of households,
average household income, population, number of workers, etc. Rather
than predict travel behavior using only this zone-level aggregate data,
many models include a ``disaggregation'' step, which classifies the
households in a zone along variables such as household size, vehicle
ownership, and number of workers. For example, a 1000-household zone
with an average household size of 3 may be classified into 500 2-person
and 500 4-person households.\footnote{The specific method for
classifying households may differ between models, so different models
will have a different distribution of households along each variable
used for classification.} This disaggregation is useful, as travel
behavior (such as the number of trips made) can vary significantly based
on a household's classification.
Subsequent model steps then use this disaggregated data in their
estimations. A 2-worker, 1-vehicle household, for example, may be
modeled to make 3.8 work trips on an average weekday, while a 1-worker,
1-vehicle household may make fewer. The trips are then added to obtain
the total number of trips produced by each zone (National Academies 2012
p. 37).
This approach is relatively straightforward: the required input data is
usually easy to obtain, the trip generation models are often simple, and
it is computationally inexpensive (National Academies 2012). However,
the types of analyses possible are limited by the initial segmentation
of the aggregate population data. An analysis based on
parents'\slash adults' highest received education, for example, would
require determining the number of households in each TAZ with each
possible combination of education level. This can theoretically be done,
but more detailed and varied analyses would require more levels of
segmentation, greatly increasing the number of classifications needed.
Since these segmentations need to be carried through each model step,
trip rates, mode choice equations, etc. need to be estimated for every
classification, and while relevant real-world data may exist, sample
sizes approach zero very quickly, and so the estimates have little
statistical value (Moeckel et al. 2020; National Academies 2012).
Further, combining these segmentations at any point precludes that
segmentation from use in subsequent model steps as well as in any
post-hoc analysis.
This approach becomes a particular issue in equity analysis because it
is perhaps impossible to determine equitable distribution of ``winners''
and ``losers'' of a potential policy without using demographic variables
in the trip generation and destination and mode choice steps (Bills and
Walker 2017). Though many studies have shown that trip production and
mode choice behavior differ by ethnic group even after controlling for
income (Bhat and Naumann 2013; Yum 2020; Zmud and Arce 2001), including
such variables in travel demand models can be problematic. Does coding
such a variable in a mode choice model represent discrimination? Or does
doing so assert that present differences resulting from unequal
opportunity will persist into future planning years? Regardless of the
reasons for their exclusion, in a trip-based model these variables
consequently cannot be used in a post-hoc analysis of a transportation
policy because the trip matrices do not contain the adequate
segmentation.
An alternative approach to population data, and the approach that ABMs
use, is to use a full synthetic population. A synthetic population takes
demographic and socioeconomic data at various levels of detail to create
a ``population'' with generally the same attributes as the study area
(National Academies 2012 p. 93). The goal is to have a population that
is functionally similar to the actual population, but without the
privacy concerns of using real data. Castiglione et al. (2006) argue
that the major advantage with this approach is that the demographic and
socioeconomic data is known at the person and household level, rather
than aggregated at the zone level. In an ABM, decisions in each model
step are tied to a specific individual, and so the individual-level
socioeconomic data remains available throughout the modeling process
regardless of the specific variables used in each model step. This
allows, for example, an equity analysis to identify the ``winners'' and
``losers'' of a proposed development without needing to encode
demographic variables into each step of the model.
Bills and Walker (2017) used the 2000 Bay Area Travel Survey to create a
synthetic population and compare the effects that certain scenarios had
on high income and low income populations. With a 20\% reduction in
travel cost, they found that high income workers benefited more than low
income workers. They did similar comparisons for scenarios involving
reduced travel times for different mode choices and saw the effects each
scenario had on the high and low income workers. These types of
analysis, which are difficult with aggregate population data, can be
very valuable in transportation planning and policy making, particularly
when equity is a priority.
It is important to note that while many connect them only with ABMs,
synthetic populations can be used in running trip-based models as well.
Trip-based models using a synthetic population---often called trip-based
microsimulation models---do exist (Moeckel et al. 2020; Walker 2005),
but these are relatively rare.
Figure~\ref{fig-pipeline-example} gives a visualization of an example
``information pipeline'' for a model using aggregate data and a model
using a synthetic population. In the aggregate data model, it is
impossible to know which trips are made by, for example, 2-worker,
1-vehicle, low-income households after the mode choice step; it only
describes which trips are made by households with fewer vehicles than
workers. With a synthetic population, however, \emph{individuals} are
being modeled, and so each trip can be traced to a specific person. All
information is known at each point in the model regardless of the data
used in previous steps.
\begin{figure}
\begin{minipage}{\linewidth}
\centering{
\includegraphics{qmd/../images/aggregate.png}
}
\subcaption{\label{fig-pipeline-example-1}Aggregate data}
\end{minipage}%
\newline
\begin{minipage}{\linewidth}
\centering{
\includegraphics{qmd/../images/synthetic.png}
}
\subcaption{\label{fig-pipeline-example-2}Synthetic population}
\end{minipage}%
\caption{\label{fig-pipeline-example}Example ``information pipeline''
for aggregate data vs.~a synthetic population.}
\end{figure}%
\subsection{Travel Behavior}\label{travel-behavior}
The other primary difference between trip-based models and ABMs---and
the main difference from trip-based microsimulation models---is that
ABMs organize travel into ``tours,'' a sequence of trips that begin and
end at the home, rather than just trips. It should be noted that Miller
(2023) argues that many current ``activity-based'' models ought to be
labeled ``tour-based'' due to this focus on building tours. This is
contrasted with ``activity scheduling'' models, in which activity
participation is modeled explicitly and trips emerge as the means to get
from one activity to the next. However, in practice there are few true
``activity scheduling'' models, and the term ``activity-based'' is
commonly used to refer to both activity scheduling and tour-based
models.
In a typical trip-based model, trips are forecasted based on empirical
trip rates, usually by trip purpose and by household type (for example,
low-income, 1-vehicle households make a certain number of ``home-based
work'' trips) (McNally 2007). These trips are then assigned an origin
and destination, mode, and often a time of day (peak/off-peak, etc.),
resulting in a list of trips between each zone by mode and purpose. A
trip-based microsimulation model may use choice models rather than
aggregate data for some of the model steps (Moeckel et al. 2020), but
the end result is similar: a list of trips by person, noting mode and
purpose. However, this trip list may be inconsistent, and the forecasted
trips may not be physically possible to complete in any sequence, as
there is no sense of ``trip-chaining.'' The hope, though, is that over
an entire population the inconsistencies would cancel out, leaving an
overall accurate forecast.
ABMs, on the other hand, explicitly model this trip-chaining in the form
of ``tours'', sequences of trips that begin and end at the home. This
approach attempts to create consistency in trip origins/destinations,
mode choice, and time of day: since each trip is a part of a tour, the
trips within a tour are dependent on each other (Rasouli and Timmermans
2014). The open-source ABM ActivitySim (Association of Metropolitan
Planning Organizations 2023a), for example, has a tour-scheduling model
that determines the number of ``mandatory'' (work, school, etc.) and
``discretionary'' tours each individual will make, and performs
tour-level mode and destination choice for each tour. After the
tour-level decisions are made, trip-level mode/destination choice is
done for each trip in the tour, including the possible addition of
subtours (see Vovsha et al. (2005), fig.~18.1).
Figures \ref{fig-network-aggregate} and \ref{fig-network-synth} show
examples of the trips distributed across several TAZs in the various
model types. Figure~\ref{fig-network-aggregate} depicts the distribution
in a typical trip-based model where the total number of trips between
each zone is modeled. With these results, the mode and purpose of each
trip is known, but because trip-based models can only model trips at the
zone level, there is no way of telling who made which trips other than
the segmentation used through each model step (see
Figure~\ref{fig-pipeline-example-1}). It is also not possible to
construct a coherent daily list of trips for individuals.
Figure~\ref{fig-network-synth}, on the other hand, depicts visual
representations of an \emph{individual's} travel made possible by the
use of a synthetic population. Figure~\ref{fig-network-synth-1} depicts
the trip distribution that could be given for an individual in a
trip-based microsimulation model. Though each individual's trips are
known, there is no guarantee of consistency between trips. For example,
a trip-based microsimulation model could predict that the individual
takes transit to work but then drives home, or that the individual makes
two trips to recreation without ever making a return trip. The
activity-based approach, depicted in Figure~\ref{fig-network-synth-2},
attempts to add consistency by modeling tours, and only generating trips
consistent with each tour.
\begin{figure}
\centering{
\includegraphics{qmd/../images/tbm.png}
}
\caption[Example network assignment using aggregate
data.]{\label{fig-network-aggregate}Example trip distribution using
aggregate data. There is little information on who is making which
trips, and it is not known how trips are related to each other.}
\end{figure}%
\begin{figure}
\begin{minipage}{0.50\linewidth}
\centering{
\includegraphics{qmd/./../images/trip.png}
}
\subcaption{\label{fig-network-synth-1}Trip-based microsimulation}
\end{minipage}%
%
\begin{minipage}{0.50\linewidth}
\centering{
\includegraphics{qmd/./../images/tour.png}
}
\subcaption{\label{fig-network-synth-2}Activity (tour)-based}
\end{minipage}%
\caption{\label{fig-network-synth}Example trip distribution using a
synthetic population allows an individual's travel to be tracked.}
\end{figure}%
In addition to intra-person dependencies, Rasouli and Timmermans (2014)
note that ABMs can model dependencies between members of a household as
well. A vehicle can't be used by multiple people in the same household
at the same time to travel to different destinations. Because the people
within the household will have travel patterns that depend on the
patterns of others in the household, a policy affecting one person in
the household can affect everyone in the household no matter how
directly the policy connects to them (Macfarlane and Lant 2023; Vovsha
et al. 2005). These effects are not possible to forecast in a trip-based
model.
Another advantage of organizing travel into tours comes regarding
accessibility analyses. Dong et al. (2006) note that when trip-based
models are used to analyze accessibility, each zone must be analyzed
independently of travel behavior. This approach only analyzes zones'
proximity to each other and does not take into account individual travel
patterns. They argue that this is a limited view of accessibility, and
discuss the ``activity-based accessibility measure,'' which is evaluated
based on all trips in a day rather than particular trips. As an example,
if an individual does not live within a 20-minute drive of a grocery
store, traditional measures might rate this as poor accessibility.
However, if a grocery store lies on their path between work and home,
then in reality the accessibility should be rated much higher. Overall,
they found that the ``activity-based accessibility measure'' predicts
more reasonable accessibility outcomes compared to traditional measures.
\section{Lack of ABM Adoption}\label{sec-literature-lack-of-adpotion}
Though ABMs have many clear theoretical advantages over trip-based
models, adoption among agencies has been relatively slow. Many ABMs are
implemented in proprietary software, which creates difficulty in
maintaining and iterating on the model, Miller (2023) argues. Even in an
open-source model like ActivitySim (Association of Metropolitan Planning
Organizations 2023a), Miller notes several disadvantages of ABMs:
\begin{itemize}
\item
Computational inefficiency and complicated program design: ABMs take
more time, more computing power, and more money to run. This is
because the synthetic population needed to run an ABM uses much more
data. In areas with thousands of TAZs and millions of people, a
supercomputer is needed, and it will cost much more than what is spent
to run trip-based models. If a region can see similar results using a
trip-based model, they may decide not to invest in an ABM.
\item
Absence of a standard model system: The modeling systems are often
designed with different approaches and for specific areas making it
hard to transfer from one urban area to another. This also makes it
difficult for agencies to determine which approach is the best and
decide which to implement. In relation to this, Miller also states
that the pressures of publishing unique and ground-breaking research
in academia can deter researchers from converging towards best
theories and methods.
\item
Lack of resources: Most of these models were developed in academic
settings which often lack resources, and possibly desire, to put them
into practice. This leaves it up to governments and consultants to put
the models into practice, but they can be hesitant to promote software
development and to invest in new systems.
\end{itemize}
For these reasons, as well as the inertia of current practices, many
agencies and organizations in the US remain using trip-based models for
demand forecasting and policy analysis.
\section{Research Gap}\label{sec-literature-research-gap}
Although there has been much research on ABMs and their theoretical
advantages, practical comparisons of the model frameworks have been
limited. It is often taken as a given that ABMs are unilaterally
superior to traditional trip-based models due to their better
theoretical foundation, but it is not clear if that better foundation
always yields better results in terms of analytical flexibility or
policy outcomes. Ferdous et al. (2012) compared the trip- and
activity-based model frameworks of the Mid-Ohio Regional Planning
Commission and found that the ABM was slightly more accurate to observed
data at the region level, but about the same at the project level. Zhong
et al. (2015) found significant differences in the predictions from an
ABM compared to a trip-based model in Tampa, Florida, but Mouw (2022)
found that both model types had similar prediction quality when compared
with observed data.
These comparisons have somewhat contradictory findings, and certainly do
not present an overwhelming victory for ABMs. Each of these comparisons,
however, is focused on the \emph{accuracy} of the two frameworks, but do
not address the methodological differences between model types. What
types of data collection/synthesis are needed for each model type? Are
there analyses that can only be done through (or that are made easier
by) one of the model types? What would an agency need in order to
transition from a trip-based model to an ABM? Are certain types of
scenarios suited to one model type? Though some of these questions have
been discussed in the literature (Lemp et al. 2007), a holistic
methodological comparison is lacking. The answers in the current
literature are mainly theoretical, with little use to an agency
considering the transition. Additionally, much of the existing
literature comparing the two model types is outdated, and the technology
of both model types may have significantly changed in recent years.
This research aims to answer these questions by providing a side-by-side
comparison of a potential trip-based and activity-based modeling
methodology. Several ``proposed development'' scenarios are run in each
model, and the strengths and weaknesses of each approach are compared.
It is important to note that this research is not focused on model
accuracy, as in any model type this can be adjusted dramatically through
calibration efforts. Rather, the focus is on the methodological
differences between the approaches, and the types of analyses that can
be done with each model type.
\bookmarksetup{startatroot}
\chapter{Methodology}\label{sec-methods}
This paper seeks to compare methodological differences between trip- and
activity-based modeling frameworks. Both model types have a wide variety
of implementations, as individual agencies will adjust the basic model
framework to match their specific needs. It would be unreasonable to
compare each of the various implementations of both model types.
Instead, we use a representative model for both types, and care is taken
to note when results apply to trip- or activity-based models generally,
and when results are specific to the models used.
The representative trip-based model is the 2019 Wasatch Front travel
demand model, and is the current production model used by the Wasatch
Front Regional Council (WFRC). This model covers much of the Salt Lake
City-Provo-Ogden, Utah Combined Statistical Area. An ActivitySim
implementation in the same study area is used as a representative ABM.
Both models are discussed in detail in the following sections.
Note that the focus is not on comparing model accuracy or performance,
but rather on comparing the process of using each model, including the
types of analyses that can be performed. There are therefore few direct
comparisons of model outputs between each type. Instead, this research
highlights the strengths and weaknesses of each model type in planning
and policy analysis, and illustrates these differences.
\section{WFRC Model}\label{wfrc-model}
The WFRC model is implemented in the CUBE software by Bentley (Bentley
Systems 2023), and is currently used by WFRC for modeling travel in the
Salt Lake City, Utah area. WFRC provided the model directly, including
land use forecasts and the current long-range transportation plan. The
model is taken essentially as-is, with no changes other than those noted
in Chapters \ref{sec-landuse}--\ref{sec-wfh} to implement the scenarios
studied in this research.
The WFRC model, like many trip-based models, requires the following
inputs:
\begin{itemize}
\tightlist
\item
Land use data, including information about population, employment, and
socioeconomic variables such as income, delineated by TAZ. This is
provided by WFRC directly, as an output of their land use forecasting
model(s).
\item
Travel skims detailing travel time, cost, etc. between each
origin-destination pair of TAZs. The WFRC model uses an iterative
process of assigning volumes to the transportation network and
recalculating the skims, which are used in the destination and mode
choice model steps.
\item
Transportation networks, including highway, transit, etc. networks
which connect the TAZs to each other. These networks contain
information such as link speed and capacity. Though the WFRC model
assigns travel volumes to the network, this paper does not analyze the
model's network assignment results. However, the network volumes are
still used to calculate the loaded network skims.
\item
Lookup tables, used in many model steps for information such as trip
rates by household type. These are taken directly from the WFRC model
without modification.
\item
Model constants and coefficients, which some model steps such as mode
choice require for calibration. These are also taken directly from the
WFRC model.
\end{itemize}
Figure~\ref{fig-wfrc-flowchart} gives an overview of the WFRC model,
showing broad model steps in a flowchart. Like many trip-based models,
the WFRC model follows the ``four-step'' approach and has main steps of
trip generation, trip distribution, mode choice, and network assignment.
The model also includes a household classification step at the beginning
where the TAZ-level data is used via lookup tables to estimate the
number of households by size, income group, number of workers, and auto
ownership. This does not create a fully synthetic or disaggregated
population, but is more segmented than the initial TAZ-level data.
\begin{figure}
\centering{
\includegraphics{qmd/./../images/wfrc_flowchart.png}
}
\caption[WFRC model flowchart.]{\label{fig-wfrc-flowchart}WFRC model
flowchart. The distribution step includes a feedback loop where
preliminary loaded network skims are used to perform subsequent
iterations of trip distribution until the distribution converges.}
\end{figure}%
The classification step takes TAZ-level socioeconomic data (such as
population, number of households, and average income) and estimates the
number of households belonging to each category of household size,
number of workers, income group, and vehicle ownership. The categories
of household size, number of workers, and vehicle ownership are
``capped'' at 6, 3, and 3, respectively (e.g., every household with 3 or
more workers is grouped into a ``3+ workers'' category). The specific
income groups used in the WFRC model are given in
Table~\ref{tbl-income-groups}.
\begin{table}
\caption{\label{tbl-income-groups}Income Groups in the WFRC Model}
\centering{
\centering
\resizebox{\ifdim\width>\linewidth\linewidth\else\width\fi}{!}{
\begin{tabular}[t]{cc}
\toprule
Income Group & Income Range\\
\midrule
1 & ≤ \$45,000\\
2 & \$45,000–\$75,000\\
3 & \$75,000–\$125,000\\
4 & ≥ \$125,000\\
\bottomrule
\end{tabular}}
}
\end{table}%
There is an additional distribution estimated, which is termed ``life
cycle'' in the WFRC model. This distribution places households into one
of three categories, intended to represent the presence of children
and/or working adults in the household. This is done by estimating the
age distribution in each TAZ and categorizing each household based on
Table~\ref{tbl-lif-cyc-categories}.
\begin{table}
\caption{\label{tbl-lif-cyc-categories}Life Cycle Categories in the WFRC
Model}
\centering{
\centering
\resizebox{\ifdim\width>\linewidth\linewidth\else\width\fi}{!}{
\begin{tabular}[t]{c>{\centering\arraybackslash}p{.9in}>{\centering\arraybackslash}p{.9in}>{\centering\arraybackslash}p{.9in}}
\toprule
\multicolumn{1}{c}{ } & \multicolumn{3}{c}{Presence of persons in household aged:} \\
\cmidrule(l{3pt}r{3pt}){2-4}
Life Cycle & 0–18 & 18–64 & 65+\\
\midrule
1 & — & ✓ & —\\
2 & ✓ & ✓ & —\\
3 & ✓ & — & ✓\\
\bottomrule
\end{tabular}}
}
\end{table}%
The segmented household data is then used in the trip generation step to
estimate the number of trips produced from each TAZ. The trips are
estimated using lookup tables which assert an average number of trips
for each household type. There are separate lookup tables for each trip
purpose, and depending on the trip purpose the lookup table uses a
different household classification. The trip rates in the lookup tables
are multiplied by the number of households in each category, and this
gives a total number of trips by purpose produced in each TAZ.
The WFRC model contains the following trip purposes: Home-based Work,
Home-based Shopping, Home-based School, Home-based Other,
Non--home-based Work, and Non--home-based Non-work. The Home-based Work
and Non--home-based Work purposes use only the number of workers per
household in determining trip productions, and all other trip purposes
use the cross-classification of household size with life cycle.
Trip attractions are estimated for each purpose based mostly on the
number of jobs by industry in each TAZ. Home-based other and
non--home-based trip attractions also are affected by the number of
households in a TAZ, and school attractions are based on the school
enrollment by TAZ. Each purpose has a different coefficient for each
variable, and these are left unchanged from the existing values.
Trip distribution uses a gravity model of the form\\
\[
T_{ij} = P_i \times \frac{A_j F_{ij}}{\displaystyle \sum_J A_j F_{ij}},
\]\\
where \(T_{ij}\) is the number of trips from zone \(i\) to \(j\),
\(P_i\) is the productions at \(i\), \(A_j\) is the attractions at
\(j\), \(F_{ij}\) is the cost term/function from \(i\) to \(j\), and
\(J\) is the set of all zones trips from \(i\) can be attracted to. The
WFRC model includes a ``distribution feedback loop,'' where preliminary
highway assignment is performed to obtain congested network skims, and
then the distribution process is repeated iteratively until the trip
distribution converges.