informatics homework

Importance of traveler attitudes in the choice of public transportation to work: findings from the Regional Transportation Authority Attitudinal Survey

Yasasvi Popuri • Kimon Proussaloglou • Cemal Ayvalik •

Frank Koppelman • Aimee Lee

Published online: 14 April 2011 � Springer Science+Business Media, LLC. 2011

Abstract The commute mode choice decision is one of the most fundamental aspects of daily travel. Although initial research in this area was limited to explaining mode choice

behavior as a function of traveler socioeconomics, travel times, and costs, subsequent

studies have included the effect of traveler attitudes and perceptions. This paper extends

the existing body of literature by examining public transit choice in the Chicago area. Data

from a recent Attitudinal Survey conducted by the Regional Transportation Authority

(RTA) in Northeastern Illinois were used to pursue three major steps. First, a factor

analysis methodology was used to condense scores on 23 statements related to daily travel

into six factors. Second, the factor scores on these six dimensions were used in conjunction

with traveler socioeconomics, travel times, and costs to estimate a binary logistic

regression of public transit choice. Third, elasticities of transit choice to the six factors

were computed, and the factors were ranked in decreasing order of these elasticities. The

analysis provided two major findings. First, from a statistical standpoint, the attitudinal

factors improved the intuitiveness and goodness-of-fit of the model. Second, from a policy

standpoint, the analysis indicated the importance of word-of-mouth publicity in attracting

new riders, as well as the need for a marketing message that emphasizes the lower stress

level and better commute time productivity due to transit use.

Y. Popuri (&) � K. Proussaloglou � C. Ayvalik Cambridge Systematics, Inc., 115 South LaSalle Street, Suite 2200, Chicago, IL 60603, USA e-mail: [email protected]

K. Proussaloglou e-mail: [email protected]

C. Ayvalik e-mail: [email protected]

F. Koppelman Midwest System Sciences, Inc., 1122 Hinman Avenue, Evanston, IL 60202, USA e-mail: [email protected]

A. Lee Strategic and Long-Range Planning, RTA, 175 West Jackson Blvd, Suite 1550, Chicago, IL 60604, USA e-mail: [email protected]

123

Transportation (2011) 38:643–661 DOI 10.1007/s11116-011-9336-y

Keywords Mode choice � Public transportation � Attitudes and perceptions � Factor analysis � Logistic regression � Elasticities

Background and motivation

The choice of commute transportation mode is one of the most fundamental aspects of

daily travel. Models of mode choice are perhaps as old as discrete choice modeling theory

itself (Domencich and McFadden 1975). Early research on mode choice had little or no

acknowledgment of the impact of attitudes on the mode choice decision. Instead, the focus

was on readily observable travel times, costs, and trip maker socioeconomics. However,

researchers soon appreciated the need to understand mode choice as a behavioral process

informed by individual attitudes toward transportation.

One of the first bodies of research in this area was due to Stopher (1967, 1969), who

recognized that attitudinal data related to comfort, convenience, and safety of various

transportation modes may add to the predictive power of mode choice models. However,

Stopher did not find a satisfactory way of quantifying these variables. Beginning in the mid

1970s, researchers were able to successfully quantify and include attitudinal data in mode

choice models. Spear (1976) compared models based only on time and cost with those

including comfort, convenience, safety and reliability measures. He concluded that such

attitudinal variables significantly improved the explanatory power of mode choice models.

Recker and Golob (1976) included variables expressing satisfaction or dissatisfaction with

mode features in their models of mode choice, and found that the model’s performance was

at least as good as models using time and cost variables. Research on the impact of

attitudes on transportation choice continued in the 1980s. One such interesting paper by

Proussaloglou and Koppelman (1989) used attitudes of commuter rail riders as explanatory

variables in a mode choice model to derive the relative importance of different attitudes

and to inform service design improvements. Train et al. (1986) explored the inclusion of

attitudes in econometric models of consumer choice.

More recently, Kuppam et al. (1999) found that the contribution of attitudinal factors

was greater than that of demographic variables in explaining mode choice behavior, and

emphasized the need for greater consideration of attitudinal and preference variables in

travel demand modeling applications. In a study for San Diego’s transit system, Prouss-

aloglou et al. (2001) and Lieberman et al. (2001) incorporated attitudes into transit plan-

ning by developing attitudinal market segments and by using segment-specific explanatory

variables and attitudinal factors within a mode choice model framework. Golob (2001)

developed joint models of attitude and behavior to explain how both mode choice and

attitudes regarding the San Diego I–15 Congestion Pricing Project differ across the pop-

ulation. This study recognized that attitudes and behavior are interdependent, and there-

fore, need to be analyzed simultaneously. In contrast to Kuppam et al. (1999), this study

did not find any significant effects of attitude on choice, but found causal links from choice

behavior to attitudes. Outwater et al. (2003) used stated-preference and attitude informa-

tion from a survey of San Francisco Bay Area residents to identify market segments, and

subsequently, to explain mode choice for each market segment. Their study showed sig-

nificant differences in time/cost tradeoffs across these market segments, reinforcing the

importance of attitude information in travel modeling. Zhou et al. (2004) used a structural

equation modeling approach along with cluster analysis to identify market segments in San

Mateo County, California. Although attitudes were not used as explanatory variables in

644 Transportation (2011) 38:643–661

123

mode choice models, separate mode choice models for each market segment were esti-

mated to quantify differences in behavior. Shiftan et al. (2008) applied a similar approach

to an empirical study for the Utah Transit Authority (UTA). Johansson et al. (2006)

discussed the importance of attitude and personality traits in mode choice using a

sequential latent variable model and discrete choice model estimation.

This paper extends the existing body of literature by examining the choice to use public

transit in the Chicago area as a function of transportation level of service, traveler

socioeconomics, and attitudes. Data from a recent Attitudinal Survey conducted by the

RTA in Northeastern Illinois were used to pursue three major steps. First, a factor analysis

methodology was used to condense scores on 23 statements related to daily travel into six

underlying constructs or factors. Second, the factor scores on these six dimensions were

used in conjunction with traveler socioeconomics, travel times, and costs to estimate a

binary logistic regression for the choice of public transit. Travel times and costs for auto

and transit were obtained directly from the Chicago Metropolitan Agency for Planning

(CMAP) travel demand model for the entire Northeastern Illinois region. Third, elasticities

were computed for the six factor variables to help rank the factors in order of importance.

As a side note to the reader, the terms ‘‘public transportation,’’ ‘‘public transit’’ and

‘‘transit’’ will be used interchangeably throughout this paper to represent the three major

transit services in the region: Chicago Transit Authority (CTA) buses and trains, Metra

commuter rail service, and Pace suburban bus service. In developing models of public

transportation choice, we do not make a distinction between these three services for two

reasons. First, a transit commute trip in the region may include one or more of these

services. Second, the objective of this paper is to understand the determinants of the choice

of public transportation versus the private automobile. Therefore, the emphasis is on

broadly defined transit service competing with the automobile.

The remainder of this paper is organized as follows: ‘‘RTA Attitudinal Survey overview

and data preparation’’ provides a brief overview of the RTA Attitudinal Survey and

describes the data preparation process. ‘‘Attitudes and factor analysis’’ section presents

results from the factor analysis of the travel-related statements. ‘‘Choice model method-

ology’’ section elaborates the choice model estimation process, while ‘‘Model results’’

section presents the empirical results. Finally, ‘‘Summary and conclusions’’ summarizes

the paper and presents the conclusions.

RTA Attitudinal Survey overview and data preparation

Transit in the Northeastern Illinois region faces both challenging and growing needs. The

RTA’s Moving Beyond Congestion Strategic Plan (RTA et al. 2007) outlined a variety of

initiatives that seek to maintain, enhance and/or expand the existing system. However, in

light of constrained resources, difficult decisions need to be made about future investments.

As a result, the RTA sought to prioritize future investments on the basis of market needs

and customer input through a comprehensive market analysis, which had two major

components:

1. Development of a baseline understanding of the regional travel patterns, and

documentation of the role of transit in serving different geographic markets; and

2. Analysis of the attitudes and preferences of transit riders and nonriders to categorize

them into distinct market segments, evaluate existing or perceived barriers to transit

use, and identify potential target segments.

Transportation (2011) 38:643–661 645

123

The first component relied heavily on the recent CMAP Travel Tracker Survey, while

the second component was based on an Attitudinal Survey of riders and nonriders in the

six-county Northeastern Illinois region. This survey is the principal source of data for the

research presented in this paper, and was conducted between June and August 2009 via two

separate methods:

1. A Computer-Aided Telephone Interview (CATI) survey was conducted with a random

sample of transit riders and nonriders. This survey generated 1,392 completed surveys.

2. A web-based survey was conducted to supplement the CATI records. The sample for

this survey was drawn from respondents of a previous CTA on-board survey and a

sample of Illinois Tollway users. The web-based survey generated 897 completed

surveys.

The survey collected information on the following items:

• The socioeconomic attributes of the respondent and the respondent’s household. • The most frequent trip in a typical week including the mode used, the origin and

destination, and the time-of-day of travel.

• Transit captivity in terms of the availability of a private vehicle for their travel. • Stated availability of and familiarity with transit services. • Ratings for 23 statements capturing time sensitivity, flexibility, travel experience,

safety, reliability, stress, social values, and cost associated with travel.

• Relative prioritization among transit travel time, frequency of service, cost of travel, and information availability using Maximum Differential Scaling (MaxDiff) experi-

mentation techniques. MaxDiff experimentation involved providing the respondents

with varying levels of four transit attributes at a time, and then asking them to choose a

‘‘most important’’ and ‘‘least important’’ for each set. As part of the RTA Attitudinal

Survey, respondents were given eight sets of such experiments. Using the ‘‘most

important’’ and ‘‘least important’’ features from the eight experiments, a statistical

model was estimated to assign a utility value to each level of each feature. These

utilities were then normalized to a scale of 100 and ranked in descending order.

• Relative prioritization among transit travel times, frequency, cost, bus stop features, and on-board comfort for a proposed premium bus service for the reverse commute and

suburban travel markets.

This paper studies the choice of public transportation to the workplace. This endeavor is

meaningful only if public transportation is at least theoretically available. Therefore, only

those trips that had both transit and highway options in the CMAP regional travel demand model were considered for analysis. The home and work locations from the survey were

each geocoded and associated with a traffic analysis zone (TAZ) from the CMAP regional

model and the level of service variables for highway and transit were attached to each

survey record. Nonmotorized trips were excluded from the analysis sample because these

generally tended to be intrazonal trips for which highway and transit level of service data

were not available from the CMAP regional model. Figure 1 presents a summary of the

data assembly process.

Table 1 presents the unweighted frequencies of key variables in the final data set. Respondents residing in Cook county, which contains the City of Chicago, constituted 80%

of the data sample, while the five suburban counties accounted for the remaining 20%. The

sample consisted of respondents over 16 years of age (by design), with all the major age

cohorts represented in the sample. Full- and part-time workers, respondents from both

small and large households, and respondents from households with different levels of

646 Transportation (2011) 38:643–661

123

vehicle ownership and number of workers were included. The ‘‘surplus’’ or ‘‘deficit’’ of

vehicles over workers has been found to be a significant variable in mode choice, as will be

described in ‘‘Choice model methodology’’ section. Finally, the sample had a high pro-

portion of transit riders (57%) compared to the observed transit market share of 14% for all

work trips in the six-county region.

The RTA Attitudinal Survey was weighted to ensure that the sample proportions

matched the observed proportions from the American Community Survey (ACS) data for

key household and person-level attributes. Further, the weights were developed so that the

fraction of trip interchanges between each county pair as a percentage of total trip inter-

changes in the region matched the CMAP Household Survey estimates. Finally, the

weights were also designed to match the transit market share in the sample to the transit

share observed in the CMAP Household Survey. Both the factor analysis results presented

in ‘‘Attitudes and factor analysis’’ section and the choice model results discussed in

‘‘Summary and conclusions’’ section incorporate these weights.

Attitudes and factor analysis

As part of the RTA Attitudinal Survey, respondents were asked to state their level of

agreement or disagreement with 23 statements pertaining to the following key dimensions

of their day-to-day travel: time sensitivity, flexibility of schedule, travel experience and

comfort, safety, reliability, stress, social perceptions, and cost. Each of these statements

was rated on a scale of 0 to 10, with 0 indicating complete disagreement and 10 reflecting

complete agreement. An average rating of 5.0 indicated that the respondent was neutral to

the statement.

Table 2 presents a summary of the statements and the average ratings provided by

transit riders and auto users. There were significantly different perceptions regarding the

fastest mode to work. Transit riders rated the statement ‘‘Driving is the fastest way to get to

the destination’’ with 4.9 compared to the auto users who had a very high rating of 8.9.

There were also major differences in respondents’ social perceptions pertaining to transit.

Specifically, transit riders had much higher ratings of 8.8 and 6.4, respectively, for the

statements ‘‘I am the kind of person who rides transit’’ and ‘‘My family and friends

1. All Data Records from RTA Attitudinal

Survey

N = 2,289

2. Select commute Trips N = 1,335

3. Select commute trips where both transit and highway are available, as per CMAP regional

model. N = 1,095

4. Select trips for people with valid ratings for all

attitude statements

N = 887

5. Select trips where either transit or auto

was chosen, leaving out short non-motorized

trips N = 868

Fig. 1 Data preparation

Transportation (2011) 38:643–661 647

123

Table 1 Sample characteristics Frequency Percent

Total sample size 868 100

Home county

Cook 695 80

DuPage 71 8

Lake 43 5

McHenry 15 2

Kane 24 3

Will 20 2

Age group

16–24 40 5

25–34 194 22

35–44 193 22

45–54 223 26

55–64 162 19

65–74 30 3

75 or older 8 1

Missing 18 2

Gender

Male 408 47

Female 460 53

Employment status

Employed full-time 780 90

Employed part-time 88 10

Household size

One person 201 23

Two people 316 36

Three people 130 15

Four people 135 16

Five people 60 7

Six or more people 26 3

Number of household vehicles

No vehicles 109 13

One vehicle 332 38

Two vehicles 298 34

Three or more vehicles 129 15

Number of household workers

No workers 14 2

One worker 339 39

Two workers 406 47

Three or more workers 109 13

Commuter type

Traditional commuter (suburbs to city) 101 12

City commuter (city to city) 645 74

Reverse commuter (city to suburbs) 40 5

648 Transportation (2011) 38:643–661

123

Table 2 Travel-related statement ratings—transit and auto users

Travel-related statement Transit users N = 491

Auto users N = 377

All N = 868

Mean SD Mean SD Mean SD

The fastest way to get to work/school is driving 4.9 4.0 8.9 2.1 8.3 2.9

If it would save time, I would change my form of travel 6.5 3.2 6.6 3.5 6.6 3.4

More than saving time, I prefer to be productive when traveling

5.4 3.0 5.7 3.2 5.7 3.1

During the day, I often make trips to a wide variety of locations

3.8 3.3 4.8 3.5 4.7 3.5

I often need to change my daily travel plans at a moment’s notice

3.5 3.2 4.4 3.7 4.2 3.6

I often make a lot of stops along the way to work/school 1.8 2.5 3.1 3.2 2.9 3.1

Privacy is important to me when I travel 5.2 3.0 5.5 3.2 5.5 3.2

I am willing to walk a few minutes to get to and from transit

7.7 2.5 7.0 2.9 7.1 2.8

I don’t mind transferring between buses and trains 6.0 3.2 5.2 3.2 5.3 3.2

It is important to be able to control heat and air conditioning when I travel

5.4 2.7 6.7 2.9 6.5 2.9

I feel safe walking near my work/school location 8.0 2.5 7.8 2.8 7.8 2.8

I feel safe on a bus or train 7.0 2.6 6.5 2.6 6.6 2.6

When I drive, I worry about getting into an accident 4.3 3.3 4.4 3.4 4.4 3.4

As long as I am comfortable, I can tolerate delays 4.8 3.0 5.7 3.0 5.5 3.0

Riding transit is more reliable than driving during bad weather

7.6 2.7 6.4 3.0 6.6 3.0

Predictable travel time is more important than a faster trip

7.4 2.3 6.8 2.5 6.9 2.5

To avoid highway congestion, I leave earlier or later than usual

6.4 3.0 7.4 3.0 7.3 3.0

Riding transit is less stressful than driving on congested highways

8.0 2.6 7.5 2.8 7.6 2.7

I am the kind of person who rides transit 8.8 1.9 4.4 3.2 5.1 3.4

My family and friends typically use public transportation

6.4 2.9 4.5 3.2 4.7 3.2

Regardless of cost, I choose the fastest way to travel 4.3 3.1 6.5 3.0 6.2 3.1

Improving transit is as good a use of tax dollars as improving roads

8.6 2.0 7.8 2.4 7.9 2.4

Increasing fares is necessary to avoid any cuts in transit service

5.0 3.1 5.4 2.9 5.4 2.9

Table 1 continued Frequency Percent

Suburban commuter (suburb-to-suburb) 63 7

Missing 19 2

Transit user?

Yes 491 57

No 377 43

Transportation (2011) 38:643–661 649

123

typically use public transportation,’’ compared to scores of only 4.4 and 4.5, respectively,

for auto users.

Transit riders appeared to be much less likely to pursue intermediate stops en route to

work compared to auto users. Transit riders’ rating on the statement ‘‘I often make a lot of

stops along the way to work/school’’ was only 1.8 compared to 3.1 for auto users. An

interesting finding relates to the scores on the statement ‘‘Improving transit infrastructure is

as good a use of tax dollars as improving roads.’’ Although transit riders expectedly had a

higher score than auto users, both groups had notably high ratings of 8.6 and 7.8, respectively. This is an important finding with respect to the public’s support for transit

improvements.

While the 23 statements presented in the table capture information on various aspects of

daily travel, using all of these statements as variables in a choice model is not advisable for

two reasons. First, there is a high degree of correlation between these statements. Second,

from the standpoint of model parsimony, using 23 variables is not desirable. To condense

the information captured by these 23 variables into a more manageable and uncorrelated

set of variables, a factor analysis methodology was adopted. Factor analysis assumes that

the ratings on the 23 statements are really ‘‘produced’’ by some underlying and unobserved

attitudes (Lehmann et al. 1998). The basic form of the factor analysis model is as follows:

Xji ¼ Xm

k¼1 kjk Fki � �

þ eji; 8j ¼ 1; 2; . . .; J and 8i ¼ 1; 2; . . .; N ð1Þ

where, Xji is the rating on statement j for person i; Fki is the value of the kth factor for the person i; kjk is the relation of the jth variable with the kth common factor, also known as the loading; and eji represents the error term. The model in (1) assumes that there are J statements, m factors, and N observations in the sample. It must be noted that the factor scores, Fki, are not observed. Factor analysis computes both the factor scores and the loadings so as to maximize the information maintained from the original statements.

The first step in factor analysis involved the computation of pairwise Pearson cor-

relation coefficients between the 23 statements. The factor loadings were then estimated

using principal component analysis (Lehmann et al. 1998). Six factors were retained

based on a combination of professional judgment and percentage of total variance in the

original variables explained by the factors. Figure 2, popularly known as the Scree Plot,

0.0%

2.0%

4.0%

6.0%

8.0%

10.0%

12.0%

14.0%

16.0%

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

% V

a ri

a n

c e

E x

p la

in e

d b

y E

a c

h F

a c

to r

Factor

Fig. 2 Scree plot from principal components analysis

650 Transportation (2011) 38:643–661

123

shows the percentage of total variance explained by each additional factor. The incre-

mental variance explained was very low beyond six factors. To enable easy interpreta-

tion, the factors were ‘‘rotated’’ using the Varimax technique (Kim and Mueller 1991)

Table 3 Rotated factor loadings for travel-related statements

Travel-related statement Factor 1 Factor 2 Factor 3 Factor 4 Factor 5 Factor 6

Riding transit is more reliable than driving during bad weather

0.66 0.10 0.08 0.16 0.11 0.09

Improving transit is as good a use of tax dollars as improving roads

0.66 -0.02 -0.12 0.09 -0.04 0.05

Riding transit is less stressful than driving on congested highways

0.65 -0.15 0.04 0.03 0.11 0.28

If it would save time, I would change my form of travel

0.45 0.02 0.04 0.19 0.00 -0.33

More than saving time, I prefer to be productive when traveling

0.44 0.18 0.31 -0.22 0.09 -0.20

Predictable travel time is more important than a faster trip

0.37 0.36 0.05 -0.09 0.33 0.30

Privacy is important to me when I travel -0.12 0.68 0.02 0.12 0.06 -0.25

It is important to be able to control heat and air conditioning when I travel

-0.06 0.65 -0.03 -0.10 -0.08 -0.06

Regardless of cost, I choose the fastest way to travel

0.04 0.59 0.14 -0.01 -0.14 0.19

To avoid highway congestion, I leave earlier or later than usual

0.15 0.50 0.11 -0.14 0.26 0.11

As long as I am comfortable, I can tolerate delays

0.13 0.49 -0.09 0.46 -0.21 0.24

When I drive, I worry about getting into an accident

0.32 0.40 0.04 0.34 -0.12 -0.16

I often need to change my daily travel plans at a moment’s notice

-0.03 0.04 0.86 0.10 0.02 0.01

During the day, I often make trips to a wide variety of locations

0.00 -0.02 0.81 0.05 0.06 0.05

I often make a lot of stops along the way to work/school

0.09 0.11 0.73 -0.08 -0.17 -0.10

I don’t mind transferring between buses and trains

0.07 0.07 -0.02 0.73 0.19 0.18

I am willing to walk a few minutes to get to and from transit

0.18 -0.22 0.12 0.65 0.13 -0.01

My family and friends typically use public transportation

0.12 0.05 -0.06 0.26 0.66 0.05

I am the kind of person who rides transit 0.41 -0.36 0.04 0.27 0.58 0.02

The fastest way to get to work/school is driving

-0.22 0.25 0.06 -0.22 -0.42 0.20

Increasing fares is necessary to avoid any cuts in transit service

0.37 -0.04 0.02 0.25 -0.61 -0.03

I feel safe walking near my work/school location

0.01 0.12 -0.05 0.11 0.02 0.66

I feel safe on a bus or train 0.32 -0.35 0.03 0.20 -0.01 0.59

Transportation (2011) 38:643–661 651

123

so that each variable loaded heavily onto a single factor. This helped in the clear

identification of variables that measured each factor, and minimized the overlap across

factors. Once the factor loadings kjk were obtained, the factor scores Fki were computed using the relationship in Eq. (1).

Table 3 presents the rotated factor loadings from the six-factor solution. For each factor,

the statements with the highest loadings are highlighted. The first factor captures the need

for reliable and stress-free commute. The second factor represents the degree of intrinsic need for privacy and comfort. The third factor captures the extent of dynamism in the work schedule and the complexity of trip-making behavior in terms of number of intermediate stops and need to pursue activities at multiple locations. The fourth factor represents the

trip maker’s tolerance to the out-of-vehicle components of a transit trip. The fifth factor reflects the trip maker’s general attitude toward public transportation. Finally, the sixth factor reflects the perceived safety of the travel environment.

The attitudinal factors uncovered as part of this study were compared to those reported

by other similar studies, specifically, Lieberman et al. (2001), Proussaloglou et al. (2001),

and Shiftan et al. (2008). Table 4 provides a comparative summary of attitudinal factors

from the current study and relevant past studies. Reliability of travel, avoidance of stress,

privacy and comfort, and sensitivity to safety were the common factors that emerged from

this paper as well as from the past studies. Proussaloglou et al. (2001) and Lieberman et al.

(2001) also found concern for the natural environment as a key attitudinal construct, a

finding supported by Shiftan et al. (2008). The RTA Attitudinal Survey did not include

statements capturing the respondents’ concern for the natural environment. The RTA

study, however, included statements pertaining to the social perceptions of the respondents

and their immediate friends and family on the importance of public transportation. Will-

ingness to use public transit was an important attitudinal construct uncovered by the

analysis presented in this paper, a finding supported by Shiftan et al. (2008).

Six standardized factor scores were computed for each respondent in the data sample.

These scores were then used as explanatory variables in the choice model estimation

described in ‘‘Choice model methodology’’ section.

Choice model methodology

A binary logistic regression methodology was used to model the choice between public

transportation and the private automobile for the work trip. Logistic regression is one of the

most commonly used statistical techniques in marketing research and travel demand

forecasting (Ben-Akiva and Lerman 1985).

The basic principle behind the binary logistic regression is that the trip maker associates

a certain utility to each transportation mode. These utilities are not observed by the analyst

and are implicit to the decision-making process. Let Ui,PA represent the utility that trip maker i associates with the private automobile, and Ui,PT be the utility associated with public transportation. The utility of the public transportation mode, Ui,PT, consists of a deterministic component Vi,PT, and a random unobserved error term ei,PT.

Ui;PT ¼ Vi;PT þ ei;PT ð2Þ

Similarly, the utility of the private automobile can be written as follows:

Ui;PA ¼ Vi;PA þ ei;PA ð3Þ

652 Transportation (2011) 38:643–661

123

Also, since only the differences in the utilities matter and not the absolute values themselves, we assume Vi,PA = 0.

The deterministic term Vi,PT was modeled as a function of three sets of attributes: first, the socioeconomic characteristics of the trip maker, represented by the column vector Si; second, the in- and out-of-vehicle times, and costs for transit and auto, represented by the

column vector Ti; and third, the attitudes of the trip maker, as measured by the six factor scores described in ‘‘Attitudes and factor analysis’’ section, represented by the column

vector Fi. Therefore,

Vi;PT ¼ a þ b0Si þ c0Ti þ u0Fi ð4Þ

where, a is the alternative-specific bias constant; b, c, and u are column vectors of parameters corresponding to each constituent variable in Si, Ti, and Fi, respectively. The parameter estimates are the output of the binary logistic regression methodology.

The trip maker i will choose public transportation over the private automobile if the utility associated with public transportation exceeds that associated with the private

automobile:

Ui;PT [ Ui;PA ð5Þ

From Eqs. (2), (3), (4), and (5), the equation above can be restructured as:

a þ b0Si þ c0Ti þ u0Fi þ ei;PT [ ei;PA ð6Þ

Table 4 Attitudinal factors from previous studies

Paper Major attitudinal constructs

Lieberman et al. (2001) and Proussaloglou et al. (2001)

Factor 1: Need for flexibility and speed

Factor 2: Concern about natural environment

Factor 3: Sensitivity to personal travel experience

Factor 4: Sensitivity to personal safety

Factor 5: Sensitivity to travel time

Factor 6: Sensitivity to transportation costs

Factor 7: Sensitivity to crowds

Factor 8: Sensitivity to stress

Shiftan et al. (2008) Factor 1: Desire to help improve air quality

Factor 2: Desire for productivity and reliability

Factor 3: Sensitivity to time

Factor 4: Sensitivity to safety and privacy

Factor 5: Need for fixed schedules

Factor 6: Sensitivity to stress and comfort

Factor 7: Willingness to use transit

This paper Factor 1: Need for reliable and stress-free commute

Factor 2: Need for privacy and comfort

Factor 3: Dynamic work schedule and complexity of trips

Factor 4: Tolerance to walking and waiting

Factor 5: Attitude to importance of public transportation

Factor 6: Perceived safety of travel ambience

Transportation (2011) 38:643–661 653

123

The probability that the trip maker i will choose public transportation is given by:

Pi PTð Þ¼ P a þ b0Si þ c0Ti þ u0Fi þ ei;PT [ ei;PA � �

; ð7Þ

or

Pi PTð Þ¼ P ei;PT � ei;PA [ � a þ b0Si þ c0Ti þ u0Fið Þ � �

ð8Þ

The error terms ei,PT and ei,PA are assumed to be independent and identically distributed, with a Gumbel distribution (Ben-Akiva and Lerman 1985). This assumption results in the

following functional form for the probability of trip maker i choosing public transportation over private automobile:

Pi PTð Þ¼ exp a þ b0Si þ c0Ti þ u0Fið Þ

1 þ exp a þ b0Si þ c0Ti þ u0Fið Þ ð9Þ

It follows that the probability of trip maker i choosing private automobile is:

Pi PAð Þ¼ 1

1 þ exp a þ b0Si þ c0Ti þ u0Fið Þ ð10Þ

The parameters a, b, c, and u are estimated using a log-likelihood maximization approach (Lehmann et al. 1998), where the probabilities of the actual choices c made by each trip maker in the sample are multiplied to obtain the likelihood function. The loga-

rithm of this function is then maximized with respect to a, b, c, and u to obtain the parameter estimates. This methodology is summarized below:

L ¼ YN

i¼1 PiðcÞ; c 2 PA; PTf g ð11Þ

In L ¼ XN

i¼1 In Pi cð Þð Þ; c 2 PA; PTf g ð12Þ

maxa;b;c; and u In L ¼ XN

i¼1 In Pi cð Þð Þ

! ; c 2 PA; PTf g ð13Þ

Two separate logistic regressions were estimated for the purpose of this paper. The

dependent variable, reflecting the choice of public transit, had a value of 1 if the person

chose public transportation, and 0 if the trip maker chose the private automobile. The first

regression used the trip maker’s socioeconomic attributes (Si) along with travel times and costs of public transportation and private automobile modes (Ti). The second regression added the factor scores (Fi) to the first regression. The results of the model estimation are discussed in the next section.

It must be noted that a sequential estimation process where the factor scores are gen-

erated using a separate model and are subsequently incorporated into a discrete choice

model, results in inconsistent and inefficient estimates of the parameters a, b, c, and u. Consistent estimates of the parameters can be obtained using a simultaneous estimation

approach, as proposed by Bolduc et al. (2008) or Ben-Akiva et al. (2002). This paper does

not correct the parameter estimates obtained from the sequential estimation process. The

authors will conduct a simultaneous estimation of parameters in a future research paper.

Further, because the data were weighted to ensure that the sample proportions matched the

observed proportions from ACS data for key household and person-level attributes, a logit

estimation based on weights may result in inefficient estimates of a, b, c, and u.

654 Transportation (2011) 38:643–661

123

Model results

Table 5 shows the results from the two model specifications discussed above. A log-

likelihood test comparing the two models clearly bears out that underlying dimensions of

travel perception and behavior have a key role to play in determining the choice of public

transportation. Details of the test are as follows:

v2 ¼�2 � LLModel 1 � LLModel 2ð Þ ð14Þ

where, LLModel 1 and LLModel 2 represent the log-likelihood values for Models 1 and 2,

respectively, as detailed in Table 5.

v2 ¼�2 � �177:795 þ 117:177ð Þ¼ 121:236 ð15Þ The critical value of the v2 statistic with six restrictions, for the six factor variables, at

the 95% confidence level is 12.59, which is much smaller than the v2 test statistic com- puted in (15). Therefore, we can safely reject the hypothesis that the factor score variables

are not significant in explaining public transportation choice.

Both models indicated a strong effect of the ‘‘surplus’’ of vehicles over workers on

transit choice. Specifically, both models indicated that as the number of available vehicles

exceeds the number of workers in the household, the probability of choosing transit falls

steeply. Both Model 1 and Model 2 indicated a strong preference for transit among city-to-

city commuters. This is a fairly intuitive finding, because of the high parking costs within

the city, as well as better familiarity with the use of transit services for the city commuters.

However, Model 1 indicates that the suburb-to-city commuters have a higher probability

of taking transit, all else being equal, than city-to-city commuters. This is a counter-

intuitive finding given the competitiveness of public transit within the city. The inclusion

of factor scores in Model 2 appears to rectify this issue, with the preference towards transit

being much lower for suburb-to-city commuters than for city-to-city commuters.

There were significant differences in the sensitivities to level of service variables

implied by Models 1 and 2. Model 1 had statistically insignificant coefficients for transit

in-vehicle time, transit fare and auto operating costs at the 90% confidence level. More

importantly, this model implied an extremely high sensitivity to out-of-vehicle time as

compared to the sensitivity to in-vehicle time. The model implied that, ceteris paribus, a minute of out-of-vehicle time was as onerous as 7.5 min of in-vehicle time.

Model 2, on the other hand, had a statistically significant coefficient for in-vehicle time

at the 90% confidence level. Although the transit fare and auto operating costs continued to

be insignificant, the level of significance was markedly better than for Model 1. The second

model also indicated a more reasonable sensitivity to out-of-vehicle time relative to in-

vehicle time, implying that, ceteris paribus, a minute of out-of-vehicle time was as onerous as roughly 1.9 min of in-vehicle time.

Model 2 had a slightly higher estimate of implied value of time for commute trips as

compared to Model 1. The Bureau of Labor Statistics reported average weekly wages that

ranged between a low of $789 for McHenry County and a high of $1,197 for Lake County

(Bureau of Labor Statistics 2010). For a typical 40-h work week, these numbers translate to

$19.7 and $29.9 per hour, respectively. The values of time implied by Models 1 and 2 were

between one-third and one-half of the hourly wage numbers, which seemed reasonable.

A study of the implied elasticity of transit choice probability to changes in level of

service measures from the two models yielded several interesting observations. The

elasticities were computed for every trip maker in the estimation sample using the values

Transportation (2011) 38:643–661 655

123

T a

b le

5 B

in a ry

lo g is

ti c

re g re

ss io

n re

su lt

s

D e p e n d e n t

v a ri

a b le

: T

ra n

si t

= 1

if tr

a n

si t

c h o se

n ;

0 if

a u to

c h o se

n

M o

d e l

1 :

L e v

e l

o f

se rv

ic e

a n

d so

c io

e c o

n o

m ic

v a ri

a b

le s

o n

ly M

o d e l

2 :

L e v

e l

o f

se rv

ic e ,

so c io

e c o

n o

m ic

v a ri

a b

le s,

a n

d fa

c to

r sc

o re

s

C o

e ff

. S

td .

e rr

o r

z- st

a t

C o e ff

. S

td .

e rr

o r

z- st

a t

T ra

n si

t c o

n st

a n

t -

2 .0

7 6

7 0

.9 6 2

- 2

.1 6

0 -

3 .8

2 0

0 1

.2 0

6 -

3 .1

7 0

S o

c io

e c o

n o m

ic v

a ri

a b

le s

N u

m b

e r

o f

v e h

ic le

s in

H H

- n

u m

b e r

o f

w o

rk e rs

in H

H -

0 .8

3 4

3 0

.1 7 9

- 4

.6 7

0 -

0 .8

1 7

3 0

.2 2

2 -

3 .6

9 0

C it

y -t

o -c

it y

c o m

m u te

r (1

= y

e s,

0 =

n o

) 1

.9 7 6

0 0

.8 5 0

2 .3

2 0

2 .3

7 3

2 1

.0 0

3 2

.3 7 0

S u

b u

rb -t

o -c

it y

c o m

m u

te r

(1 =

y e s,

0 =

n o

) 2

.0 3 1

5 1

.0 7 6

1 .8

9 0

1 .2

7 8

3 1

.2 9

6 0

.9 9 0

L e v

e l

o f

se rv

ic e

v a ri

a b

le s

T ra

n si

t in

-v e h ic

le ti

m e

(m in

) -

0 .0

1 2

6 0

.0 1 4

- 0

.8 9

0 -

0 .0

3 2

5 0

.0 1

8 -

1 .7

9 0

T ra

n si

t o

u t-

o f-

v e h

ic le

ti m

e (m

in )

- 0

.0 9

4 1

0 .0

1 8

- 5

.2 9

0 -

0 .0

6 0

1 0

.0 1

9 -

3 .2

2 0

T ra

n si

t fa

re (c

e n ts

) -

0 .0

0 0

9 0

.0 0 2

- 0

.4 5

0 -

0 .0

0 1

7 0

.0 0

2 -

0 .7

2 0

A u

to tr

a v

e l

ti m

e (m

in )

0 .0

2 5

5 0

.0 1 2

2 .2

2 0

0 .0

2 7

2 0

.0 1

4 1

.9 2 0

A u

to o

p e ra

ti n

g c o

st (c

e n

ts )

0 .0

0 0

5 0

.0 0 3

0 .1

8 0

0 .0

0 2

7 0

.0 0

3 0

.8 3 0

F a c to

r sc

o re

s

1 .

N e e d

fo r

re li

a b le

a n

d st

re ss

-f re

e c o

m m

u te

– –

– 0

.7 6

5 1

0 .1

7 2

4 .4

4 0

2 .

N e e d

fo r

p ri

v a c y

a n

d c o

m fo

rt –

– –

- 0

.8 1 0

5 0

.1 5

5 -

5 .2

2 0

3 .

D y n a m

ic w

o rk

sc h e d u le

a n d

c o m

p le

x it

y o f

tr ip

s –

– –

- 0

.6 9 0

4 0

.1 6

9 -4

.0 9

0

4 .

T o le

ra n c e

to w

a lk

in g

a n d

w a it

in g

– –

– 0 .6

0 9 4

0 .1

8 3

3 .3

2 0

5 .

A tt

it u

d e

to im

p o

rt a n

c e

o f

p u

b li

c tr

a n

sp o

rt a ti

o n

– –

– 1

.0 9

6 2

0 .1

9 5

5 .6

3 0

6 .

P e rc

e iv

e d

sa fe

ty o

f tr

a v

e l

a m

b ie

n c e

– –

– -

0 .3

2 0

5 0

.1 6

3 -

1 .9

6 0

M o

d e l

d ia

g n

o st

ic s

N u

m b

e r

o f

o b

se rv

a ti

o n

s (N

) 8

6 8

8 6

8

N u

m b

e r

o f

v a ri

a b

le s

8 1

4

In it

ia l

lo g

li k

e li

h o o

d (c

o n

st a n

ts o

n ly

) -

2 5

7 .9

9 8

- 2

5 7

.9 9

8

F in

a l

lo g

li k

e li

h o o

d a t

c o

n v

e rg

e n

c e

- 1

7 7

.7 9

5 -

1 1

7 .1

7 7

656 Transportation (2011) 38:643–661

123

T a

b le

5 c o

n ti

n u

e d

D e p e n d e n t

v a ri

a b le

: T

ra n

si t

= 1

if tr

a n

si t

c h o se

n ;

0 if

a u to

c h o se

n

M o

d e l

1 :

L e v

e l

o f

se rv

ic e

a n

d so

c io

e c o

n o

m ic

v a ri

a b

le s

o n

ly M

o d e l

2 :

L e v

e l

o f

se rv

ic e ,

so c io

e c o

n o

m ic

v a ri

a b

le s,

a n

d fa

c to

r sc

o re

s

C o

e ff

. S

td .

e rr

o r

z- st

a t

C o e ff

. S

td .

e rr

o r

z- st

a t

P se

u d

o R

-s q

u a re

d w

.r .t

. c o

n st

a n

ts o

n ly

m o

d e l

0 .3

1 0

9 0

.5 4

5 8

T im

e a n d

c o st

se n si

ti v it

ie s

R a ti

o o

f tr

a n

si t

o u

t- o f-

v e h

ic le

to in

-v e h

ic le

ti m

e se

n si

ti v

it y

7 .4

9 1

.8 5

Im p

li e d

V a lu

e o

f T

im e

(d o

ll a rs

p e r

h )

8 .6

8 1

1 .6

1

Transportation (2011) 38:643–661 657

123

of the level of service variables and the predicted probability of transit choice. A weighted

average of these individual elasticities was then computed for the entire data sample. These

elasticities are shown in Table 6, along with the weighted mean values of the level of

service variables.

Model 1 indicated a very low own-elasticity of transit choice to transit in-vehicle time

as compared to the cross-elasticity to auto travel time. This implies that all else being

equal, a 1% increase in transit in-vehicle time will have a much smaller impact on the

choice of transit than a 1% decrease in auto travel times. Model 2, however, appears to

show a higher own-elasticity to transit in-vehicle time compared to the cross-elasticity to

auto travel time. While the relative magnitudes of transit and auto travel time elasticities

are not exactly known, one would expect the own-elasticity of transit choice probability to

transit in-vehicle time to be higher than the cross-elasticity of transit choice probability to

auto in-vehicle time. In this sense, Model 2 appears to be more intuitive than Model 1.

From a policy standpoint too, using Model 1 instead of Model 2 can lead one to estimate a

much lower change in transit ridership due to service cuts or conversion from express to

local service.

Similar observations can also be made in relation to the elasticity of transit choice to

transit fares and costs. Model 2 indicates that the fare elasticities, while still very much in

the inelastic zone, are much higher than those implied by Model 1. Using Model 1 instead

of Model 2 for evaluating changes to fare policy can accordingly imply smaller changes in

projected ridership.

Overall, it appears that explicitly accounting for traveler attitudes through the inclusion

of factor scores improves the intuitiveness of the model. In addition, there is a marked

improvement in goodness-of-fit measures for Model 2 as compared to Model 1 (see

Table 5). The pseudo R-squared measure with respect to the constants-only model, increased from 0.31 for Model 1 to 0.55 for Model 2, indicating an improvement of over

75%.

The coefficients of the factor variables in Model 2 were all significant at the 90%

confidence level, and had intuitive signs. Factor 5, which captures the trip maker’s attitude

toward transit and its importance to society, had the most positive coefficient. Factor 1,

indicating the need for a reliable, stress-free, and productive commute, had the next highest

coefficient. Tolerance to walking and waiting, represented by Factor 4, also had a strong

positive sign.

Among the factors that reduced the likelihood of taking transit, the need for privacy and

comfort represented by Factor 2 had the highest coefficient. Factor 3, which represents the

extent of dynamism in the work schedule and the complexity of trip-making in terms of the

Table 6 Elasticity of transit choice to transit and auto level of service

Level-of-service variable Weighted sample mean

Model 1 Model 2

Elasticity Std. error z-stat Elasticity Std. error z-stat

Transit in-vehicle time (min) 43 -0.4819 0.543 -0.890 -1.2456 0.697 -1.790

Transit out-of-vehicle time (min)

34 -2.9278 0.563 -5.200 -1.8706 0.587 -3.190

Transit fare (cents) 252 -0.1928 0.426 -0.450 -0.3704 0.519 -0.710

Auto travel time (min) 42 0.9024 0.404 2.230 0.9623 0.498 1.930

Auto operating cost (cents) 206 0.0830 0.463 0.180 0.4780 0.578 0.830

658 Transportation (2011) 38:643–661

123

number of intermediate stops and need to pursue activities at multiple locations, had the

next most negative impact. Perceived safety of the trip maker’s ambience had the lowest

negative effect on transit choice.

To better quantify the effect of changes in attitudes on transit choice, elasticities were

computed for each factor score. The elasticities of factor scores are presented in decreasing

order of magnitude in Table 7. Since the factor scores are standardized, their means across

the sample are close to 0 and their standard deviations close to 1.

Table 7 reinforces the observations made previously on the relative impact of various

factors. Trip makers whose family and friends ride transit regularly or who have a positive

perception toward transit themselves, appear to be the most likely to choose transit. Despite

the seemingly obvious nature of this statement, it points toward an interesting ‘‘networking

effect’’ for transit choice. It also appears that trip makers who place a premium on reliable,

stress-free and productive commute are more likely to use transit. This provides an insight

into the key messages that transit operators could consider in their marketing campaigns.

Finally, although models with factor scores as explanatory variables are not intended for

forecasting purposes, the evaluation of elasticities of transit market share to travel times

and fares suggests that ignoring the impact of attitudes can lead one to estimate a much

lower change in transit ridership due to service cuts or fare changes.

Summary and conclusions

This paper demonstrated the importance of understanding trip maker attitudes and

accounting for their impact on the choice of public transportation to work. An important

caveat is in order here. Just as attitudes toward travel affect the daily mode choice

behavior, the choice of public transportation could in turn affect attitudes in the longer

term. We recognize the feedback between attitudes and behavior but such an analysis

would require a more elaborate experimental set up than what was developed for the RTA

Attitudinal Survey.

A factor analysis of the 23 statements from the RTA Survey indicated the presence of

six broad constructs. These included the need for reliable and stress-free commute, need

for privacy and comfort, the complexity of trip-making behavior, tolerance to waiting and

walking, general attitude toward public transportation, and finally, the perceived safety of

the travel environment. Reliability of travel, avoidance of stress, privacy and comfort, and

sensitivity to safety were the factors that were common to this study as well as similar

studies in the past (Proussaloglou et al. 2001; Lieberman et al. 2001; Shiftan et al. 2008).

Table 7 Elasticity of transit choice to factor scores

Factor Effect on transit choice

Weighted mean

Elasticity Std. error z-stat

5. Attitude to importance of public transportation : 0.033 0.196 0.045 4.38

1. Need for reliable and stress-free commute 0.019 0.058 0.021 2.83

4. Tolerance to walking and waiting 0.009 0.037 0.017 2.20

6. Perceived safety of travel ambience ; 0.012 -0.016 0.012 -1.40

3. Dynamic work schedule and complexity of trips

0.010 -0.051 0.019 -2.61

2. Need for privacy and comfort 0.024 -0.095 0.027 -3.52

Transportation (2011) 38:643–661 659

123

Comparison of binary logistic regressions for the choice of public transit with and

without attitudinal constructs yielded several insights. First, the explicit treatment of

factors generated more intuitive estimates of the importance of level of service charac-

teristics. The model with factor scores provided a more reasonable ratio of sensitivities to

out-of-vehicle and in-vehicle times. Second, ignoring the impact of attitudes could lead one

to estimate a much lower change in transit ridership due to service cuts or conversion from

express to local service. Third, the model with factor scores generated higher elasticities to

transit fares. Using the model without factor scores for evaluating fare policy changes can

accordingly suggest a lower expected change in ridership. Finally, the addition of factor

scores improved the goodness-of-fit of the choice model by 75%.

The choice model with factor scores helped identify a pecking order of important

determinants of transit choice. The way transit is perceived by the trip makers themselves

or by their immediate friends and family could have a major impact on the choice of public

transportation. From a policy standpoint, this could mean that an incentive program that

subsidizes current transit riders for inducing their friends to ride transit could be effective

in increasing transit ridership. The authors are cognizant of the operational and techno-

logical challenges of such a program. A small-scale experiment to study the feasibility of

such a program could be very helpful.

Finally, another important finding is that traveler preferences and behavior are affected

by the need for a reliable, stress-free and productive commute. These are attributes where

transit often has an advantage over the automobile especially in metropolitan areas where

fixed route transit service competes with autos facing congested highway conditions.

Marketing campaigns should therefore emphasize all three of these attributes to appeal to

commuters, stress the competitive advantages of transit, and affect their mode choice

behavior.

Acknowledgments The authors would like to thank RTA for its financial and managerial support on the Market Analysis project, which is the source of the content and data presented in this paper. The authors also thank CMAP for providing highway and transit network skim data from the regional travel demand model. The authors acknowledge the efforts of Laurie Wargelin of Abt SRBI, and Greg Spitz and Margaret Campbell of RSG in administering the CATI and on-line surveys, respectively.

References

Ben-Akiva, M., Lerman, S.: Discrete Choice Analysis: Theory and Applications to Travel Demand. The MIT Press, Cambridge, MA (1985)

Ben-Akiva, M., Walker, J., Bernardino, A., Gopinath, D., Morikawa, T., Polydoropoulou, A.: Integration of choice and latent variable models. In: Mahmassani, H. (ed.) In Perpetual Motion: Travel Behaviour Research Opportunities and Application Challenges, pp. 431–470. Elsevier Science, Pergamon (2002)

Bolduc, D., Boucher, N., Alvarez-Daziano, R.: Hybrid choice modeling of new technologies for car choice in Canada. Transp. Res. Rec. J. Transp. Res. Board 2082, 63–71 (2008)

Bureau of Labor Statistics: Quarterly census of employment and wages, fourth quarter 2009. http://stats. bls.gov/news.release/pdf/cewqtr.pdf (2010). Accessed 31 July 2010

Domencich, T.A., McFadden, D.: Urban Travel Demand: A Behavioral Analysis. American Elsevier Publishing Company, Inc., New York (1975)

Golob, T.F.: Joint models of attitudes and behavior in evaluation of the San Diego I-15 congestion pricing project. Transp. Res. A 35, 495–514 (2001)

Johansson, M.V., Heldt, T., Johansson, P.: The effects of attitudes and personality traits on mode choice. Transp. Res. A 40(40), 507–525 (2006)

Kim, J.O., Mueller, C.W.: Introduction to Factor Analysis: What It Is and How to Do It. Sage Publications, London (1991)

660 Transportation (2011) 38:643–661

123

Kuppam, A.R., Pendyala, R.M., Rahman, S.: Analysis of the role of traveler attitudes and perceptions in explaining mode-choice behavior. Transp. Res. Rec. J. Transp. Res. Board 1676, 68–76 (1999)

Lehmann, D.R., Gupta, S., Steckel, J.H.: Marketing Research. Addison-Wesley, New York (1998) Lieberman, W., Schumacher, D., Hoffman, A., Wornum, C.: Creating a new century of transit opportunity:

strategic planning for transit. Transp. Res. Rec. J. Transp. Res. Board 1747, 60–67 (2001) Outwater, M.L., Castleberry, S., Shiftan, Y., Ben-Akiva, M., Zhou, Y.S., Kuppam, A.: Attitudinal market

segmentation approach to mode choice and ridership forecasting. Transp. Res. Rec. J. Transp. Res. Board 1854, 32–42 (2003)

Proussaloglou, K.E., Vaidya, R., Haskell, K., Ben-Akiva, M.: An attitudinal market segmentation approach to commuter mode choice and transit service design. In: Proceedings of the 2001 Annual Transpor- tation Research Board Meeting, Washington, DC (2001)

Proussaloglou, K.E., Koppelman, F.: Use of travelers’ attitudes in rail service design. Transp. Res. Rec. J. Transp. Res. Board 1221, 42–50 (1989)

Recker, W.W., Golob, T.F.: An attitudinal modal choice model. Transp. Res. 10, 299–310 (1976) RTA, CTA, Metra, Pace: 2007—The year of decision, regional transportation strategic plan, final report. Mov-

ingbeyondcongestion.org, http://www.movingbeyondcongestion.org/downloads/RegTransportStratPlan_ FinalReplrez_031207.pdf (2007). Accessed 28 July 2010

Shiftan, Y., Outwater, M.L., Zhou, Y.S.: Transit market research using structural equation modeling and attitudinal market segmentation. Transp. Policy 15(3), 186–195 (2008)

Spear, B.D.: Generalized attribute variable for models of mode choice behavior. Transp. Res. Rec. J. Transp. Res. Board 592, 6–11 (1976)

Stopher, P.R.: A probability model for travel mode choice for the journey to work. Highw. Res. Rec. 283, 57–65 (1969)

Stopher, P.R.: Choice of mode travel for the journey to work. Unpublished Ph.D. Dissertation, University College, London (1967)

Train, K., McFadden, D., Goett, A.: The incorporation of attitudes in econometric models of consumer choice. Working Paper, Cambridge Systematics (1986)

Zhou, Y.S., Viswanathan, K., Popuri, Y., Proussaloglou, K.: Transit district customers in San Mateo County, California: who, why, where, and how. Transp. Res. Rec. J. Transp. Res. Board 1887, 183–192 (2004)

Author Biographies

Yasasvi Popuri is a Senior Associate in the Chicago office of Cambridge Systematics with interest and experience in the fields of Discrete Choice Modeling, Marketing Research, and Transportation Demand Forecasting. He is a Young Member of the Transportation Research Board (TRB) Committee on Transportation Demand Forecasting.

Kimon Proussaloglou is a Principal in the Chicago office of Cambridge Systematics specializing in urban and intercity travel demand analysis and forecasting. Over the past two decades he has used Discrete Choice Modeling and Marketing Research techniques to advise US Federal and State transportation agencies on important policy issues.

Cemal Ayvalik is an Associate in the Chicago office of Cambridge Systematics with experience in the fields of Marketing Research, Geographic Information Systems and Transportation Demand Forecasting. He has worked on several major transit market analysis projects in Chicago over the past decade.

Frank Koppelman , professor emeritus of civil and environmental engineering at Northwestern University and founding principal of Midwest System Sciences, has been active in academic and professional education, research and consulting in travel behavior for more than 35 years. Dr. Koppelman is the first recipient of the Lifetime Achievement Award of the International Association for Travel Behavior Research.

Aimee Lee is the Division Manager for Strategic Planning and Policy at the Regional Transportation Authority (RTA) in Chicago. She served as the Project Manager for RTA on the recently concluded Market Analysis study.

Transportation (2011) 38:643–661 661

123

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.