SOCIAL MEDIA, TRAVEL SURVEYS AND METRO DISRUPTIONS
Anastasia M. Pnevmatikou, Ph.D, Researcher
Matthew G. Karlaftis†, Ph.D, Associate Professor
Department of Transportation Planning and Engineering, School of Civil Engineering, National
Technical University of Athens
1.
SOCIAL MEDIA, TRAVEL SURVEYS AND METRO DISRUPTIONS
Collecting data during unplanned or even planned metro disruptions on altered travel
behaviour patterns, is a major priority for operators in order to evaluate current
demand management strategies and minimize the impacts to traffic and the
transportation system.
Public transport users are not ‘mobility static’; they use the web to get information on
actual travel times, they check on the smartphone the arrival of their bus, they even
consult transport related websites to get information on potential disruptions on the
network. Social media (such as Facebook, LinkedIn, Google+, Twitter, YouTube,
Flickr, MySpace, Instagram, FourSquare, Pinterest) are booming in popularity for
generations Y and Z, as they tend to be more technologically oriented compared to
previous generations in most developed countries. The possibilities offered by the
social media and web in collecting data for transport surveys related to disruptions on
public transport networks is discussed in this paper. The purpose of this study is to
analyse the factors that influence travel patterns during public transport strikes, using
data that comes from a web-based survey of Athens, Greece, designed to study
metro travelers’ response to a programmed metro closure.
The motivation for this research came through the need for various data collection
methods to support ongoing research activities on public transport network
disruptions as a result of strikes of operational employees. As these initiatives are not
under the umbrella of a research program, the method of data collection needed to
be made with the minimum cost.
Key research questions for this paper include: what role does age, income, and
frequency of mode use play in how travelers respond to network closures? What are
the limitations and bias of using social networks to conduct transport surveys?
Findings may prove useful in understanding changes in Public Transport user
choices and patterns during service disruptions, and in better planning the ‘return’ of
users to PT following closures.
1
© AET 2014 and contributors
2. BACKGROUND
Immediately following a metro disruption or even well in advance of a programmed
metro closure, people often use TV media and online social media to get information
on the disruption and the alternative ways to travel. As there is usually a time gap
between the actual occurrence of the disruption and the investigation of the problem,
there is considerable difficulty of collecting in situ reliable travel data.
Travel data collected by travelers affected during the disruption may be alternatively
used to support research. Due to recent development in social media technologies
and increasing use of these technologies in daily life, there is a clear need to explore
the use of these tools during metro closures.
Past work on metro disruptions has focused on emergency management and
analyses of travel behaviour. Most previous studies investigated the impact of
network disruptions (due to a transit strike, bridge closures, or special events) on
highway congestion (Lo and Hall, 2006; Zhu et al., 2011; Blumstein and Miller, 1983).
Although there is growing interest in research on travel behaviour during disruptions,
little research exists on the role of social media to collect data related to metro
disruptions. Most studies have explored the use of social media for emergency
management during disruptions (Harazzeen, 2011; Outlook Research Limited, 2012,
Pender, 2013). Over the past two decades, many papers have been published on the
impact of ICT on travel behavior (Wang and Law, 2007) and on the use of social
media to collect passenger feedback, aid transport planning or provide news
(Mokhtarian et al., 2006; Colins et al., 2012, Evans-Cowley and Griffin, 2012), hence
there is almost no research specifically examining the use of social media in
collecting travel data during a real metro disruption.
Note that just recently, transport operators in Europe have began to employ social
media applications (such as Facebook, LinkedIn, and Twitter) to provide transport
information to users as up-to-date service information, travel alerts, alternative travel
routes or modes (Pnevmatikou et al., 2014).
3. SURVEY DESIGN, DISSEMINATION AND DATA COLLECTION
We adopted a web-based survey approach to collect information from public
transport users. For regular internet users, the Web has been found to be useful
means of conducting research. The travel survey data was collected between
2
© AET 2014 and contributors
November 27th, 2011 and January 27th, 2012, during a series of planned strikes in
the Athens metro system.
The final sample contained 1944 questionnaires (1083 car non owners (NCO) plus
861 car owners (CO)). The survey was designed as a typical conjoint choice type
experiment which intentionally did not consider the presence of a no-choice option
because the purpose is to analyze travel patterns under repeated strikes where the
available options were limited. A Stated Preference survey was undertaken, in an
effort to obtain information on traveler preferences with respect to a hypothetical
metro service disruption. Survey participants were asked to select preferred choices
among hypothetical scenarios of alternative trip and mode options. Attributes and
attribute levels for each scenario are presented in Table 1. Four, three-level attributes
were used to describe the “bus” option: a) in-vehicle time, b) out-of-vehicle time, c)
bus fare and d) number of transfers. The attributes of the “car” option were: a) invehicle time, b) out-of-vehicle time and c) total operating cost. Finally, the attributes
for the “taxi” option were: a) in-vehicle time, b) out-of-vehicle time and c) taxi charge.
A fractional factorial orthogonal design was used to reduce the number of choice
scenarios to 27 SP choices; the design was subsequently divided into three groups
(blocks) of 9 choices for each SP questionnaire to reduce respondent burden.
TABLE 1: Definition of Attributes and Attribute Levels in the Stated Preference Design
Variables
In-vehicle-travel time (min)
Total travel cost (euro)
Out-of-vehicle travel time (min)
Number of transfers
Travel by Bus
25
40
50
1.20
1.40
2.00
10
13
18
0
1
2
Travel by Car
15
30
40
3.00
5.00
8.00
8
15
20
0
0
0
Travel by Taxi
10
25
35
3.00
7.00
12.00
3
5
7
0
0
0
The available options considered were buses, private cars, and taxis. Alternatives of
either canceling the trip or shifting the departure time were not offered to the
respondents, as the closure of the metro system was programmed for long periods
and therefore, at least for commuters, such actions would be meaningless.
The questionnaire was constructed using the commercial survey software
kwiksurveys (www.kwiksurveys.com) and was publicly available on several transportrelated websites. The questionnaire was also disseminated to personal contacts and
3
© AET 2014 and contributors
mailing lists via email, facebook, and ‘what’s up’ application. Table 2 presents the
questionnaire’s dissemination strategy.
Table 2: Dissemination strategy
No of days needed to collect Number of
the sample
completed)
Personal contacts
2 months
147
Mailing lists
2 months
114
Travel related websites
1 month
1500
Facebook, what’s up
2 months
183
Total
1944
sample
(fully
Response rate by facebook, what’s up application and mail was about 50%. A
significant number of questionnaires of about 78% of our total sample came from
respondents who visited various transport related websites via their Smartphones
(android, iOS, windows phones), or their desktop computers, to get information
regarding potential closure of the metro network.
Tables 3 and 4 present sample traveler characteristics of car-owners and car nonowners.
Table 3 Sample characteristics-travelers owning a private vehicle
4
© AET 2014 and contributors
Variable
Male
Type
Dummy
Dummy
Statistics
F(1)=495
F(0)=366
F(1)=276
(%)
(57%)
(43%)
(32%)
Age18-25
Age25-35
Dummy
F(1)=345
(40%)
Age35-45
Dummy
F(1)=156
(18.1%)
Age45-55
Dummy
F(1)=64
(7.4%)
Age55+
Dummy
F(1)=20
(2.3%)
Work
Dummy
Low_Income
Dummy
F(1)=528
F(0)=333
F(1)=375
(61%)
(39%)
(44%)
Med_Income
Dummy
F(1)=291
(34%)
High_Income
Dummy
F(1)=195
(22%)
Subway Users
Dummy
F(1)=644
F(0)=217
(75%)
(25%)
Usual Travel Time to
work/School
Categorical
Flexible working
Dummy
F(1)=328
F(2)=275
F(3)=156
F(4)=102
F(1)=418
F(0)=443
(38%)
(32%)
(18%)
(12%)
(49%)
(51%)
Description
=1 if male
=0 if female
=1 if respondent’s age >=18 and
<=25
=0 if not
=1 if respondent’s age >=25 and
<=35
=0 if not
=1 if respondent’s age >=35 and
<=45
=0 if not
=1 if respondent’s age >=45 and
<=55
=0 if not
=1 if respondent’s age >55
=0 if not
=1 if working
=0 if not
=1 if <800 euros
=0 if not
=1 if <800-1500 euros
=0 if not
=1 if >1500 euros
=0 if not
=1 if they use subway at least 12 times a week or more
=0 if they use subway less than
once a week
=1 if 5-30 minutes
=2 if 31-45 minutes
=3 if 46-60 minutes
=4 if >60minutes
=1 if they have flexible working
hours
=0 if they do not have flexible
working hours
Table 4 Sample characteristics-travelers not-owning a private vehicle
Variable
Male
Type
Dummy
Dummy
Statistics
F(1)= 427
F(0)= 656
F(1)=650
(%)
(39%)
(61%)
(60%)
Age18-25
Age25-35
Dummy
F(1)=320
(29%)
Age35-45
Dummy
F(1)=87
(8%)
© AET 2014 and contributors
5
Description
=1 if male
=0 if female
=1 if respondent’s age
>=18 and <=25
=0 if not
=1 if respondent’s age
>=25 and <=35
=0 if not
=1 if respondent’s age
>=35 and <=45
=0 if not
Age45-55
Dummy
F(1)=17
(2%)
Age55+
Dummy
F(1)=9
(1%)
Work
Dummy
Low _Income
Dummy
F(1)=368
F(0)=715
F(1)=806
(34%)
(66%)
(74%)
Medium_Income
Dummy
F(1)=239
(22%)
High_Income
Dummy
F(1)=38
(4%)
Subway users
Dummy
F(1)=904
F(0)=179
(83%)
(17%)
Usual Travel Time to
Work/school
Categorical
Flexible working
Binary
F(1)=300
F(2)=316
F(3)=281
F(4)=186
F(1)=580
F(0)=503
(28%)
(37%)
(33%)
(22%)
(54%)
(46%)
=1 if respondent’s age
>=45 and <=55
=0 if not
=1 if respondent’s age
>55
=0 if not
=1 if working
=0 if not working
=1 if <800 euros
=0 if not
=1 if <800-1500 euros
=0 if not
=1 if >1500 euros
=0 if not
=1 if they use subway at
least 1-2 times a week or
more
=0 if they use subway
less than once a week
=1 if 5-30 minutes
=2 if 31-45 minutes
=3 if 46-60 minutes
=4 if >60minutes
=1 if they have flexible
working hours
=0 if they do not have
flexible working hours
As expected limited questionnaires were collected by middle-aged travelers.
The majority of our sample is between 18 and 35 years old. This is a common
problem with web surveys and surveys distributed using social media
applications, which mainly suffer from serious coverage problems. The reason
is that people with no internet access or unable to use computers are not able
to participate to web surveys. However, this shortcoming is later alleviated in
further research (not analysed in this paper) by jointly using SP and RP data,
forming in this way a representative sample for all ages of metro travelers.
4. MODEL ESTIMATION
Analysis was undertaken depending on the car ownership for the
respondents, since the car was not offered as an alternative option to
travelers who reported not owning a private vehicle. For each dataset, we
tested a Multinomial Logit model, to model choices among alternative mode.
Model estimation was done using the NLOGIT software package (v5.0). The
mathematical framework for Logit Models is discussed in detail in Washington
et al. (2010). Table 5 presents the results of the fully specified model
incorporating trip related and traveler related variables using Logit model.
© AET 2014 and contributors
6
In the MNL model, most parameters are significant and with the expected
signs, except for ‘age’ variable for ‘bus’ alternative and age group 45-55 for
‘car’ alternative. This is expected as age group 45-55 is not represented
enough in our sample, which is a limitation of using social media and web to
collect travel data.
As expected, results indicate that ‘individuals’ with higher travel times derive
higher utility from public transport than from a car or taxi, probably because of
the effects of costs.
Table 5 Fully-Specified MNL, MNP and HEV results for car owners and car non owners
Utility parameter name
Logitb
Logita
Model
coefficient
t-stat
coefficient
t-stat
0.959
3.47
0.787
2.52
0.391
7.57
BUS
Constant Bus
Gender: Male
Age:18-35
n/s
-0.693
-2.34
Age:35-45
n/s
-0.711
-2.34
Age:45-55
-0.878
-2.55
-0.878
-2.55
Income: High
-0.289
-2.79
-0.460
-3.36
0.526
8.20
n/s
Income: Low
Usual Travel time :46-60 mins
0.276
2.57
0.149
2.16
Usual Travel time :>=60 mins
Use FRT modes at least
once a week
CAR
Constant Car
0.766
5.89
0.194
2.48
0.337
3.64
1.419
5.00
-
-
Gender: Male
0.216
2.88
-
-
Age:18-35
0.713
2.74
-
-
Age:35-45
0.573
2.16
-
-
-
-
-
-
n/s
Age:45-55
Trip purpose: Work
Use FRT at least once a week
Flexible working hours
In-vehicle time
Cost
Out-of-vehicle-time
Number of transfers
Null Log-Likelihood
Final log-likelihood
 Likelihood ratio test
Rho-square (ρ2)
-0.239
-0.497
-2.12
-5.62
n/s
-0.041
-25.34
-0.220
-25.55
-0.041
-9.44
-0.255
-8.08
-7829.10
-6537.45
-2583.31
0.165
n/s
n/s
-0.039
-20.99
-0.299
-37.90
-0.039
-5.36
-0.193
-6.29
-6097.18
-4891.29
-2411.79
0.198
Not significant at 10% level
a
N=861 respondents; sample size for MNL model refers to individuals (each providing 9 responses), no of
observations is 7749.
b
N=1083 respondents; sample size for MNL, MNP, HEV model refers to individuals
(each providing 9 responses), no of observations is 9747.
© AET 2014 and contributors
7
The positive sign for ‘male’ suggests that male car owners spend more time
driving than traveling on the bus during a metro closure. The age variable is
alternative specific; car-owners between 18 and 35 appear to drive more often
during a closure than other age groups. Travelers who usually travel more
than 45 minutes (Table 5) are more attracted to the bus during closures. As
we can see from Table 5, the coefficient of ‘usual travel time (>45 minutes)’ is
positive and highly significant for bus users. This finding is reasonable since
travelers, and particularly commuters, usually drive for shorter distances
during subway closures.
The coefficient for ‘transfer’ for car owners is 6 times higher than that of travel
time (either in-vehicle or out-of-vehicle time), indicating that car owners are
more likely to object to additional boarding on different modes during a
subway closure compared to travelling longer or paying more. For non-car
owners though, travel time is less significant than transfer and cost during a
subway closure. Non-car owners have a lower value for time than car-owners,
and value more the cost of public transport than car-owners. These findings
indicate that non-car owners derive the highest benefit from a reduction in bus
fare during a subway closure. Travelers who use metro regularly would use
bus in the event of a programmed closure of the subway network, while
travelers who usually travel by modes other than subway, would use the car.
Commuters who own a car seem to be more likely to drive in the event of a
closure than other travelers.
Low income travelers who own a car tend to use bus more during metro
closures, while the income variable was found to be non-significant for car
users. This is expected, as low income travelers usually prefer public
transportation modes during closures for longer distances due to financial
constraints. Flexibility of working hours was found to be statistically significant
only for car owners, indicating that travelers who are flexible with arrival and
departure time are less likely to choose car-related modes during a subway
closure, while travelers with inflexible hours are restricted to using a car in the
event of a metro closure.
As results indicate ‘age’ variable is not statistically significant for all age
groups and this is an important issue of heterogeneity of the data collection
method used in this survey. Nevertheless, there are gender differentials in
Information and Communication Technologies’ (ICT) uptake, as well as in
mobility practices. In addition, not everyone has access to ICT applications
© AET 2014 and contributors
8
and platforms, can afford such service (e.g. internet utility bills) or acquire
devices required to run them (Pnevmatikou et al., 2014).
5. MODEL ESTIMATION
This study offers an analysis of traveler responses to a programmed metro
closure due to personnel strike using social media and web to collect travel
data. A Multinomial Logit model was built to better understand the choice of
model for travelers during a strike. Socio-demographic variables (age, income,
gender, flexibility in working hours) and trip-related variable (purpose, usual
travel time) were among the variables discussed. Results indicated that
travelers who are regular metro travelers and have therefore been more
affected by network disruptions, are less likely to shift to the car as a result of
that disruption. Younger travelers (age <35 years) are more likely to change
their travel patterns. Regular metro travelers are more likely to use other
public transportation alternatives rather than shifting to the car during a
programmed closure.
The travel patterns during a subway closure depend on their individual
socioeconomic and trip related characteristics. Our research shows that those
travelers who are flexible with arrival and departure times at their destination,
would travel by public transport during a closure. For travelers who are not
flexible in terms of time our research indicates that they would consider using
their private vehicle during a closure.
Web and Social media was found to be a convenient way of data collect
during metro disruptions. However, there is a wide range of barriers related to
the use of social media and the web to collect travel data. One limitation of
this study is the relative small size of travelers aged over 45 years old. Special
groups of vulnerable users, like unemployed people, immigrants or even
people with reduced mobility may be socially excluded from using the wide
range of social media applications and may require extra facilities for public or
private transport and improved personal skills to use these platforms. Further
research should be aimed at collecting larger data sets, possibly relying on
sources other than social networks. Since not everyone is trained to use
social media and web and not everyone has access to internet, adequate
training provided by the transport operators is needed pre-disruption in order
to prepare travelers and provide them with adequate knowledge on how to
use these devices during such events to get information on alternative modes
of travel.
© AET 2014 and contributors
9
The results of our questionnaire can shed light on traveler experiences during
a closure and on the strategies people adopt when experiencing a disruption.
BIBLIOGRAPHY
Accent, 2010. Review of Stated Preference and Willingness to Pay Methods.
Available at:
http://webarchive.nationalarchives.gov.uk/+/http://www.competitioncommission.org.uk/our_role/analysis/summary_and_report_combined.pdf
Blumstein A. and H. Miller, (1983). Making do: The Effects of a Mass Transit
Strike on Travel Behavior, Transportation, Vol. 11, Springer, 1983, pp. 361382.
Bregman, S. (2012). Uses of Social Media in Public Transportation - A
Synthesis of Transit Practice. Transit Cooperative Research Program.
Transportation Research Board. Brighton, Massachusetts, USA. Available at:
http://onlinepubs.trb.org/onlinepubs/tcrp/tcrp_syn_99.pdf
Collins, C. , Hasan, S. and S. Ukkusuri, (2012). A Novel Transit Riders'
Satisfaction Metric: Riders' Sentiments Measured from Online Social Media
Data. Transportation Research Board 91st Annual Meeting. Washington, DC.
Harazzeen, S., (2011). Effects of information on decisions of passengers
during service disruptions. Civil Engineering. London, UK, University College.
MSc.
Washington S., Karlaftis M.G., and F.L. Mannering, (2010). Statistical and
Econometric Methods for Transportation Data Analysis, 2ndedition. Chapman
& Hall/CRC, Boca Raton, FL.
Mokhtarian, P.L, Salomon I., and S. Handy, (2006). The Impacts of ICT on
Leisure Activities and Travel: A Conceptual Exploration. Transportation, 33,
pp.263-289.
Lo S. and R. Hall R. Effects of the Los Angeles Transit Strike on Highway
Congestion, Transportation Research Part A, Vol. 40, Elsevier, 2006, pp. 903917.
Evans-Cowley J. S, and G. Griffin, (2012). Micro-participation: Community
Engagement in Transportation Planning with Social Media. Transportation
Research Board 91st Annual Meeting. Washington, DC.
Outlook Research Limited, (2012). Short and Tweet – How passengers want
social media during disruption. London, UK, for Passenger Focus and Abelio,
UK.
© AET 2014 and contributors
10
B. Pender, G., Currie A., Delbosc N., Shiwakoti, (2013). Social Media
Utilisation during Unplanned Passenger Rail Disruption – What’s not to ‘Like’?
Australasian Transport Forum 2013 Proceedings, 2-4 October 2013,
Brisbane, Australia.
Pnevmatikou AM, Milioti CP and MG Karlaftis, (2014), Economic recession
and the impact of ict on integrated mobility during disruptive events, OPT-I, An
International Conference on Engineering and Applied Sciences Optimization,
Kos Island, Greece, 4-6 June 2014.
Wang, D. and F.Y.T Law, (2007). Impacts of Information and Communication
Technologies (ICT) on Time Use and Travel Behavior: A Structural Equations
Analysis. Transportation, 34, 513-527.
Zhu S., Tilahun N., He X. and D. Levinson (2011). Travel Impacts and
Adjustment Strategies of the Collapse and the Reopening of I-35W Bridge,
2011, in D. Levinson, H. Liu and M. Bell, eds, `Network Reliability in Practice',
Springer, pp. 5-20.
© AET 2014 and contributors
11