hTe diverse transportation services of today's cities pose a significant opportunity for personalizing route planning user's experiences. To understand what people are taking into account in mobility decisions, we present a pair of human studies. First, we conducted interviews to explore what factors people consider to make decisions about their travel. Then, we designed, ran, and analyzed a survey to study which of those factors are crucial to model for understanding mobility decisions and behavior. Our analysis indicates that people's mobility decisions incorporate considerably more information than current applications support. Also, diferences in people's experience, personality, and requirements significantly impact their mode choice preferences. We close with a discussion about how these findings go beyond trip planning to other potential smart cities problems.
• Human-centered computing → Empirical studies in collaborative and social computing; User centered design; • Applied computing → Transportation.
hTe last decade has seen explosive growth in mobility oferings. Ridehail, rideshare, carshare, bikeshare and dockless scooter services all compliment the personal vehicle and existing public transit. To aggregate these oferings, companies (e.g., TripGo and Moovel) have been developing applications to enable trip planning and payment across these services. The diversity of oferings and population travel needs present a challenge for creating personalized IUI Workshops’19, March 20, 2019, Los Angeles, USA © 2019 Copyright for the individual papers by the papers’ authors. Copying permited for private and academic purposes. This volume is published and copyrighted by its editors. interfaces that help people make transportation decisions. An important component of generating good recommendation is understanding a person’s trip context so that a relevant mode from a set of alternatives can be suggested.
As an example, consider Jane who wants to commute to her ofifce and uses a travel application on her phone to figure out how to get there. The travel app suggests walking to the bus stop and taking the direct bus to ofice - chosen from all available route alternatives. This recommended trip has been personalized to Jane’s trip context. The app makes this suggestion by considering that the bus route is direct and has frequent service, the bus stop is nearby, the weather is pleasant, and that she values that she can work while traveling on the bus.
hTis paper studies what constitutes a person’s trip context that is useful in generating recommendations as in the example above. In this paper, we approach the problem of characterizing an individual’s trip context as follows: we leverage previous transportation research and psychological research of personality traits to generate an an initial set of factors; we conducted 20 traveler interviews to support the initial set of factors as well as generate new ones; we conducted a survey study with 235 participants from urban locations in the US to statistically validate factors generated in the previous steps. hTrough this approach, we identified a significantly expanded set of trip context factors grouped into two categories below: Static: Factors that do not change day to day – Network: Factors related to the transportation local transportation options – Personal: Factors related to an individual’s perception and experience with various transportation options – Personal traits: Factors related to an individual’s personality Dynamic context: Factors related to the trips purpose, the weather, and transportation network that vary from day to day. hTe rest of this paper is organized as follows. We begin by briefly summarizing prior work on understanding traveler context in Section 2. We describe our user interview methodology and results in Section 3. Then in Section 4, we present our survey and highlight factors of potential value for personalization. We close with a discussion of implications for smart cities in Section 5. 2
Researchers from diverse backgrounds have provided partial
answers to this question. Transportation mode choice research
typically focuses on travel time and cost with more complex models
including additional factors for income, auto ownership, and
purpose [
Our approach to understanding a traveler’s trip context begins by
supplementing existing social science research [
To elicit information relevant to mode choices, we developed a guide to conduct 30 minute, semi-structured interviews with participants. The guide contained probes detailing information of interest that were not be read aloud. The interview began with the participant signing an informed consent form that provided details about the overall project. The interviewer and the participant were instructed to not use the participant’s name or any other identifying information about them. Then, the participants were asked to provide details about themselves, including age, gender, occupation, as well as accessibility of various transportation at their home, ofice, and regular shopping location. The interviewer guided participants to describe their mobility routines including the locations they regularly visited and how flexible time to reach there was. Participants were asked to provide details about their usual travel (to work, shopping, social outing, other outings), available alternatives, and circumstances under which they would consider an alternative. The interviewer probed about how the participant would change their commute if they were recommended an alternative that would help congestion. Interviewer also asked about what effortful activity a participant did because it was good for the environment. Participants described their atitudes towards public transit. Additionally, participants also discussed the information they would like to know if they were recommended an alternative. The interview ended with a general open-ended conversation about travelling in Los Angeles and an appreciation for the participant’s time.
Using the guide, we interviewed 20 people (7 women, 13 men) in the age range 21-79 (mean 37:5, standard deviation 16:8) in Los Angeles. For their eforts, the participants were paid $40 using Amazon gift-cards or checks based on their preference. 3.2
Data Analysis and Results hTe interviews were recorded and transcribed. One author highlighted the content which containted information about mode choice in the transcripts. In a separate meeting and through an interactive process, the authors categorized and grouped factors. Through this process, we developed a comprehensive taxonomy of factors that underlie a person’s choice of mode for their commute. 3.2.1 Understanding mode choice behavior. Collectively, participants reported a variety of modes that they use for their weekly commute including private non-motorized modes - walking and biking, public transit options - bus and train, ride hailing - Uber/Lyft/Taxi , and shared rides - carpools, and private motorized modes - driving. Participants varied in terms of their commute with 8 people expressing a strict preference for driving regardless of where they were going. 2 people expressed a preference for working from home and driving for non-work related trips. 4 people expressed a clear preference for taking public transit, 3 for walking/biking, 2 for a ride service or carpool, and 1 for driving/biking. A few people undertook multi-modal trips that included taking a rideshare to a train station. 3.2.2 What underlies mode choice? Our analysis uncovered a large set of factors when they decided how to travel to their destination. hTese mode choice factors can be organized in a taxonomy as follows: (1) Static factors: This grouping contains factors that do not change from trip to trip and are a property of the transportation network that a person is embedded in due to where they reside and work, to their personal situation including education and income, as well as atributes of their personality. (a) Network factors: Many participants corroborated the factors transportation researchers use by highlighting the cost and time trade-of they must make: P 4 - It’s really time convenience and cost are the things that we would… and The first issue I think would be eficiency, the time of route. Others highlighted the accessibility of public transit as an important factor: P 12 - … and since it is so close to me, my nearest transit stop, like I said, is :2 miles away, it’s prety much a no-brainer . Participants also reported what value diferent modes provided to them: P 12 - (about public transit) Yeah. I do it because it’s stress-free. This is Los Angeles. It is just trafic 24/7. There is no quiet time on the roads anymore…, P 13 - While you’re on the bus or on the train you can do other stuf rather than just focusing on driving. Time spent driving is basically lost time., P 05 - I love my car because I really enjoy listening to the radio, NPR, or music and both. (b) Personal factors: Participants reported that the nature of their employment along with the flexibility it afords was a big factor in how they chose modes. P 07 - When I as working I would have taken almost any suggested mode of transportation. That would not have been an issue because I had a destination and I had an arrival destination approximate and I had a routine that could be duplicated, P 17 I create my own schedule so I try to always travel of peak actually. (c) Personality: As explored previously by other researcher, our analysis showed that people’s personality atributes such as openness, commitment etc. may afect their mode choice. Some participants expressed an openness to try new things P 25 - would be willing to try almost anything once, to be honest. That just may be my personal outlook on life and P 02 - ”When I was teaching I would sometimes drive and sometimes take the train and [I would out of conscience] try to take the train”. (2) Dynamic context: This grouping includes factors that may change from trip-trip. (a) Parking: P 4 - Yeah, I feel like Trader Joe’s there’s always that time when the parking lot’s packed and you can’t get a parking space and the store is sold out of stuf. (b) Weather: P 17 - Also, the weather here, it’s super hot a lot of the time and super sunny. I don’t want to be out walking around on the hot streets. (c) Trafic: P 7 - I would stay away from high trafic times. If I’m going to go I’m going to usually leave before four o’clock to be on my way home by five or six when trafic becomes a litle more congested. (d) Trip purpose: P 19 - Specifically to shopping. If I were going to a mall for some reason, which I don’t do particularly often, I wouldn’t want to be taking a bus there just because I wouldn’t want to be lugging around things that I’m buying on and of the bus. 3.2.3 Utility of personalized travel planning. Several of the participants talked about the cognitive load of planning a trip which especially afects the decision of choosing an alternative route: P 01 - I certainly see congestion over on the Sepulveda parallel to the 405. Very, very, heavy. Puts almost an extra hour on the bus trip, but I don’t really know what the solution is.. This suggests that a travel assistant that could personalize trips to a person’s context would be useful in travel planning. More importantly, participants expressed that they would be interested in diferent ways to make their trip if it was suggested without them having to invest time thinking: P 03 - If you were to tell me there was a diferent way, I would probably take it. If you were to say this is a beter way to travel or more eficient way, then yeah I would be open to that. and P 19 - So, as much information as you can take into account, I would be in favor of using that…I want as much intel as possible. 4
hTe traveler interviews above uncovered a wide range of factors that underlie the decision to take a specific mode for a given trip. However, the interviewed sample only had 20 people and was limited to the Los Angeles region. In a large population, all factors may not be significant in predicting mode choice and consequently, not all factors are equally valuable personalizing travel recommendations. To determine which factors mater for a larger population sample, we designed and ran a traveler behavior survey study. We used insights from previous research and traveler interviews to generate a large set of factors that may have a role in mode choice. hTe study had 112 questions that measured peoples’ mode choice travel behavior (via self-reports). It also recorded their mode choice context by measuring factors determined in the interview analysis and included public transportation availability and proximity, availability of other transportation modes, and evaluation of different modes as well as several dimensions of personality. 4.1
We deployed the travel behavior survey via Qualtrics 1 to 677 participants from urban areas in contiguous US. The participants were recruited by a local recruiting company that were tasked to find commuters in Los Angeles county. After eliminating responses that were completed in less than a reasonable threshold time, included repetitive responses, or contained contradictory responses, we obtained 235 responses (128 women & 106 men; age 18-79, mean 46:51, sd 13:5) for further analysis. The participants were paid $20 using Amazon gift-cards or checks based on their preference. These data were analyzed as described in the following sections. 4.2
We were interested in exploring how various static mode choice factors determined from previous research and traveler interviews impact mode choice behavior in our population sample. All questions were framed as multiple-choice questions. 4.2.1 Dependent variables. We define a participant’s mode choice behavior as the proportion of all local trips for which they use each mode. We measured a participants mode usage via self-reports for 8 modes: walking or wheelchair, bicycle, bus or shutle, train or subway or tram, ride services - Taxi, Uber, Lyft etc, carpools, private automobile, and motorcycle or scooter. Mode usage was measured on an ordinal scale:
1: 0-5%, 2: 5-25%, 3:25-50%, 4: 50-75%, 5: 75-95%, 6: 95-100%
Together, 8 ordinal measurements on the scale above (one for each
mode) characterize a participant’s mode choice behavior.
4.2.2 Independent variables. We measured participants on various
types of static factors: network factors, personal factors, and
personality using a variety of Likert and ordinal scales. An an example,
we measured accessibility of bus by having people report:
(1) Distance from a bus stop: How long does it take to walk to
the nearest bus stop/train station?
1: Don’t know, 2: No local service, 3: 31-60 minutes, 4: 21-30
minutes, 5: 11-20 minutes, 6: 6-10 minutes, 7: 5 minutes or less
(2) Frequency of bus or shutle: During rush hours, how often
do public transit services depart from your nearest stop/station?
1: Don’t know, 2: No local service, 3: Over an hour, 4: 31-60
minutes, 5: 21:30 minutes, 6: 11-20 minutes, 7: 6-10 minutes, 8:
Every five minutes or less.
4.2.3 Multi-variate, multiple linear regression. As there are several
dependent variables which together characterize mode choice
behavior, we performed multi-variate, multiple linear regression
analyses [
where Y is s a matrix of n observations on m dependent
variables; X is a model matrix with columns for p independent
variables, typically including an initial column of 1s for the regression
constant; B is a matrix of regression coeficients, one column for
1htps://www.qualtrics.com/
each independent variable; and E is a matrix of errors. We consider
the regression coeficients, their significance level, and standard
errors to understand the relationship between independent and
dependent variables. Additionally, for each independent variable we
consider type-II MANOVA’s test - Pillai’s statistic [
An example of this analysis is in Table 1. The column vector corresponds to mode usage dependent variable while the rows contain independent variables. The analysis shows that both distance and frequency are useful in diferentiating between various modes. hTis interpretation is based on the significant type-II MANOVA test reported in parenthesis underneath them in the table. Further, we see that closer the nearest bus stop is, the more likely it is that the person will walk. And, higher frequency of bus service greatly improves the usage of public transport. Both independent variables are negatively correlated with driving, as expected. In the following sections, we report on only the overall significance of factors due to space constraints. 4.3
Static factors are the aspects of the transportation network, the
individuals interaction with diferent modes, and their personality
that do not vary from day to day.
4.3.1 Network factors. Network factors concern the transportation
network around the home and workplace of the individual. Table
2 lists the factors by if they are significantly impact individuals
mode choice. In agreement with existing literature, the number of
transportation options available to an individual and the distance
and frequency of nearby transit were all important for
determining which modes people prefer. Other positive results that are less
obvious in the literature is that peoples’ mode choice is afected by
their beliefs concerning the possibility of cycling to their typical
destination. This belief was measured using a 7 point Likert
agreement to the statement - Cycling would be possible for at least part
of my typical trips/commute (considering fitness, time required, road
safety, space of bicycle on transit, and place to safely park a bike).
hTe Likert measurement can be considered a proxy for the how
accessible bicycling is in the participant’s transportation network.
4.3.2 Personal Factors. Personal factors pertain to the individual
and how they perceive and interact with the transportation
network they are embedded in. Personal factors include the time and
distance of peoples commutes, their previous experience with
different modes, and how they value diferent characteristics of
various modes. Table 3 lists the factors by if they supported as to being
important for transportation mode choice. In agreement with
previous research, the distance and time to work were both predictive
of overall mode preferences. Also, the familiarity with the public
transport system and knowledge of other options were important
for predicting mode preferences. Individuals considerations of the
health benefits, working en route, and parking also impacted their
mode choice. We were surprised to see no efect from peoples
atitudes toward sustainability, reliability and flexibility.
4.3.3 Personality Factors. Following recent work that suggests that
personality may be inferred from online behavior [
hTe majority of the dimensions of personality traits did not have support for predicting mode choice. We did not see an expected relationship between extroversion and public modes of transportation. Nor did we see any relationship between openness to new experiences and mode choice. The only Big 5 dimension that was predictive was conscientiousness (eficient v. easy-going). This was also supported in the responsibility dimensions with the only significant traits being anticipating the needs of others and taking others interest into account. Finally, the only supported dimensions of persuasion include reciprocity and consensus, but the findings for these were mixed with some versions of these traits showing no efect. 2http://ipip.ori.org/
Our traveler interview analysis suggests that dynamic factors such as purpose of the trip, weather, trafic etc. cause people to change mode preferences. Here we validate this finding as well as study how these factors afect mode preferences in a larger population sample. We extracted various scenarios from the traveler interviews that can have an impact on the mode choice for a trip. Our collection included: heavy trafic, heavy rain/snow fall, expecting alcohol consumption, accompanying child or infant, urgency, traveling while dark, planned grocery/shopping diversion, expecting parking problems, spare the air day, have an important appointment, hot weather, and pleasant weather. For each of these scenarios, we asked participants if they were more than 50% likely to select a mode to make a trip. The participant could pick multiple modes for each scenario. Additionally, mode usage (from previous sections) captures each participant does in the normal scenario. hTe scenarios are independent variables in our analysis and mode selection is the dependent variable. 4.5
We tabulated our data in a contingency table as shown in Figure 1. The horizontal axis has all the scenarios and the vertical axis has various modes. Each cell in the table represents the number of people who said that they will consider taking the corresponding mode under given scenario. As expected, most of the surveyed population drives in the normal scenario. This distribution changes as scenarios change. A Pearson’s χ 2 test on this contingency table was significant with a p-value < 0:005 suggesting that mode selection is impacted by dynamic factors. Weather has a big impact; we see that people consider shifting to manual modes of transport in pleasant weather while snowy/rainy and hot weather shift the distribution towards private motorized modes. Similarly, concerns about parking and expected alcohol consumption tends to shift the distribution toward public transit and mobility services. 5
Rapid worldwide urbanization coupled with new technologies from smart phones to dockless scooters to autonomous vehicles entering the marketplace creates a need for travel assistants. In this work, we present a framework for identifying what factors should drive the personalization of future traveler assistants. By first conducting open-ended qualitative interviews, we identify factors outside of existing literature. To test if these factor afect traveler behavior, we conducted a survey study whose results indicate the importance of static network, static personal, and static personality factors. Furthermore, we identified a set of dynamic factors (i.e., those that change from trip to trip) that impact mode choice. This study indicates that future travel assistants would benefit from personalization along a wider variety of dimensions than currently considered in the research community and the marketplace.
In addition to helping individuals navigate their environment,
understanding how people make transportation decisions could
have a significant impact on a critical societal problem.
Transportation is one of the largest consumers of energy in the world - in
the United States, it accounted for 29% of energy consumption in
20163. Many areas of urban transportation networks are
underutilized while other areas are congested. Congestion alone in the
United States wastes 6:9 billion hours and 3:1 billion gallons of fuel
per year [
hTis work was funded by the Advanced Research Projects AgencyEnergy (ARPA-e) under award number DE-AR0000612. The authors would also like to thank the anonymous reviwers of HUMANIZE19 whose detailed feedback greatly contributed to this paper.