People

Transportation Research Record

2019, Vol. 2673(1) 524–531
Ó National Academy of Sciences:
Risk Assessment Model for Railway Transportation Research Board 2019
Article reuse guidelines:
Passengers on a Crowded Platform sagepub.com/journals-permissions
DOI: 10.1177/0361198118821925
journals.sagepub.com/home/trr

Shintaro Terabe1, Takayuki Kato1, Hideki Yaginuma1,

Nan Kang1, and Kosuke Tanaka1

Abstract
The purpose of this study is to develop a risk assessment measure to provide an understanding of the safety of railway station
platforms. We estimated the number of accidents on a platform in a year. It was influenced by the factors such as the design,
equipment, and the profile of the station users. Consequently, 16 factors were defined, such as the platform design and pas-
senger movement. Poisson regression and negative binomial regression models were employed to estimate and analyze the
number of accidents from a station database containing 158 platforms from 52 stations in Japan. The results show that the
number of accidents is related to the length of the narrow part of a platform, the width of the gap between the platform and
train, the curvature of the platform, passenger flow crossing, the number of trains passing and stopping, and the audio and
visual announcements concerning approaching trains. We expect that this result will allow railway companies to identify
weaknesses in station safety and subsequently set priorities for investments in safety. Furthermore, administrative authorities
can evaluate the safety performances of railway companies, and consider subsidies for investments in safety.

Trains are considered to be safer than other means of they do not result in an accident, such falls cause fear
transportation. However, passengers on train platforms and anxiety among the people involved.
sometimes encounter risks such as falling from the plat- We should invest in passenger safety by installing
forms or colliding with a moving train. Passengers who platform gates and other devices to alert passengers of
wish to walk a shorter distance to avoid a crowded plat- approaching trains. However, investing in these devices
form tend to walk near the platform edges. Many railway involves considerable costs. Considering the risks associ-
companies have been attempting to decrease the number ated with railway platforms, it is worthwhile for railway
of passengers walking on platform edges. However, their companies to understand the priorities for investment.
efforts up to this point have been futile at crowded urban Consequently, this study aims to provide a risk assess-
stations (Figure 1). ment that allows the comparison of different factors
According to a transportation safety white paper (1), affecting the number of passenger accidents on a railway
420 accidents occurred in Japan in 2016 that involved platform. Thus, the safety investment for each platform
being struck by a train, which resulted in 202 deaths. Of within a station can be prioritized.
those 420 accidents, 179 (resulting in 24 fatalities) The UK Rail Safety and Standards Board (RSSB)
involved falling from a platform and subsequent injury released a guidance document that aids railway compa-
or death as a result of colliding with a train. Therefore, nies in identifying safety performance indicators (SPIs)
to evaluate the Japanese railway system, the safety and that reveal the most appropriate operations for them to
security of these stations must be emphasized. In particu- perform, and to ensure their continued effectiveness. This
lar, factors such as the station design (platform width, document provides an overview and background for the
platform shape, presence of narrow parts, etc.), station SPIs, and also detailed step-by-step guidance on how to
equipment (automatic platform gates, warning devices
for approaching trains, etc.), train operation (operation
1
frequency, train speed, etc.), and users’ profile and influx Department of Civil Engineering, Tokyo University of Science, Yamazaki,
Noda, Japan
(crowding, passenger movement and behavior, etc.)
should be examined. These factors are associated with Corresponding Author:
falling from platforms onto the railway tracks. Even if Address correspondence to Shintaro Terabe: [email protected]
Terabe et al 525

Figure 1. Crowded urban platforms.

develop and manage them. However, it does not provide We obtained accident data from two different railway
SPIs (2). Fielding examined three programs, including companies. They allowed us to investigate not only the
the federal triennial reviews that monitor compliance typical characteristics of platform risks, but also the dif-
with planning and grant requirements (3). Although, ferent features that cannot be included in a general risk
safety was among one of the seven considered dimensions, assessment model. We estimated the number of passen-
it was not discussed in detail. Kecklund et al. also dis- ger accidents on a platform by applying the Poisson
cussed safety issues, focusing on the working environment model and regression model using a negative binomial
of train drivers in particular (4). Nelson et al. mentioned distribution.
the safety of level boarding, most commuter railways in
Japan enable passengers to board trains at the same level
as the platform, without a stairway (5). A few articles Data
such as the Transit Cooperative Research Program (6), Platform Accident Database
and Cothen et al. (7) and Muttram (8) discuss rail safety.
However, the discussion was limited to the regulatory Two anonymous railway companies, A and B, provided
procedures and policy issues only. Tennyson analyzed data on platform accidents on all of their company lines.
data from the National Transit Database from 1993 to The items contained in this database are shown in Table
1995 to review the relative safety of railways. According 1. These items include the accidents reported by the sta-
to this work, automated guideways led to no grade- tion staff on the platforms. The accidents were originally
crossing accidents, but led to a higher rate of station acci- classified into four types: falling, falling into gaps, mis-
dents. As a light railway covers fewer formal stations, steps, and others. Falling is defined as passengers falling
fewer station accidents have been recorded. A commuter from the platform onto the railway surface when no
railway involves more miles per passenger, and can reduce train is present, whereas falling into gaps is defined as
the frequency of station accidents per passenger-miles (9). passengers falling into the gap between the train and the
A study conducted by Hunter-Zaworski (10) was a platform. Missteps imply that passengers step off the
part of a larger research project that led to the develop- platform edge or their feet slip when boarding or alight-
ment of a manual (11) for practitioners to improve safety ing from a train. Others include falling onto the platform
at rail transit platform/train and platform/guideway or touching a running train.
interfaces. As a part of that effort, passenger injury inci-
dent data were collected from five rail transit systems,
and interviews were conducted with safety officers at rail
Selection of Stations, Platforms, and Accident Types
transit systems in the US and Canada. The data collected Approximately 500 platforms exist on the lines of the
showed that stairs and escalators, and general platform railway companies considered in this study, with acci-
tripping produced more injury incidents than the plat- dents reported on only some of the platforms. As it is
form/train and platform/guideway interfaces. difficult to investigate all of the stations in detail, it was
526 Transportation Research Record 2673(1)

Table 1. Available Items from Platform Accident Database

Group Item

When Date, day of the week, time

Where Line, station, platform number, train number, door number, gap
What Type of accident: falling, falling gaps, missteps, and others.
Who Gender, age, visually impaired, wheelchair, other disability
Why Drunk, sickness, smartphone use, reading book, trouble with another passenger

necessary to select appropriate stations to conduct the Crowded trains and platforms cause passengers to step
survey. We selected the top 30% stations with the largest off the edge of the platform or slip while boarding or
number of passengers from each railway company. For alighting from trains. At some stations, some passengers
the study30 stations, including 93 platforms were selected near the doors who are still onboard have to alight so as
from railway company A, and 22 stations, including 65 to give space to other passengers who want to alight at
platforms were selected from railway company B. In the station. The chances of stepping off the platform edge
total we surveyed 158 platforms from 52 stations. increase when crowded trains arrive at these stations.
We also selected accident types for this study. In the
database for railway company B, only a few accidents
classified as falling gaps, missteps, and others were
reported. We discussed the reason for fewer accidents of
falling gaps, missteps, and others with the railway com-
pany B. The reason was related to the information dis-
closure regulations. We had to take additional steps to
access the full accident database. To avoid a biased
result, we focused on falling accidents. The total number
of accidents was 418.

Summary of Accidents
Figure 2 shows the number of accidents summarized by
month. The total number of accidents decreases during
summer, and increases in April, November, and
December. Before aggregating the data, we predicted
that there would be many accidents in December owing Figure 2. Number of accidents classified by month.
to the year-end party season, and in April at the begin-
ning of the new fiscal year. In December, passengers can
be under the influence of alcohol owing to year-end par-
ties. In April, as many people begin to commute to new
companies and schools, they are unfamiliar with rail-
ways. Although the number of accidents in December is
high, it is not the highest. We discussed the reason for
the frequency of accidents being reported in November
with railway company A, however, the reason was still
unclear.
Figure 3 shows the number of accidents summarized
by the days in a week. The number of accidents is higher
on Fridays and Saturdays. This is caused by passengers
being inebriated rather than falling because of the crowd
caused by the commuting rush. Accidents occurred more
frequently on Mondays and Tuesdays for railway
company A compared to railway company B. Railway
company A has more commuter lines than railway com-
pany B. Figure 3. Number of accidents classified by day in a week.
Terabe et al 527

Table 2. Safety Evaluation Factors and Their Definitions

Safety factors Definition

Platform design
(1) Length of narrow part Length of a platform in which only one passenger can walk through [m]
(2) Width of gap between platform and train Width of gap between train and platform [cm]
(3) Platform shape Island-type or not
(4) Area of platform Usable platform area per number of tracks [m2]
(5) Platform curving in the middle Straight line, concave curve, or convex curve
Passengers flow
(6) Crowding on platform Number of daily passengers/area of platform
(7) Passenger flow crossing Number of crossing points of passenger flows
(8) Passenger flow outside of the white lines Platform length over which safety fences are not built
(9) Crowding at stairs and elevators Number of escalators and elevators
Train operations
(10) Number of trains passing and stopping Number of trains passing and stopping
(11) Visual announcements of approaching trains Number of LEDs warning of approaching trains
(12) Audio announcements of approaching trains Number of approach announcements, melody, and arrival announcement
(13) Clarity of indicators of train approach direction The direction of approaching trains is always the same or not
Passenger profile
(14) Number of drunken passengers (Not used in this research)
(15) Number of visually impaired passengers Number of facilities for visually impaired person near the station
(16) Number of elderly passengers (Not used in this research)

Risk Assessment Model Selection of Evaluation Factors. Accidents are likely to occur
under several conditions that depend on the design and
Modeling the Number of Passenger Accidents environment of a platform, and the passenger profile.
Although the number of accidents on the platform From the list of all typical evaluation factors presented
appears to be strongly correlated with the number of in the previous research, we selected several factors and
passengers, other factors concerning the platform also considered their contribution to platform safety.
exist. It is interesting to model the relationship between We focused on the selection of evaluation factors that
the number of accidents and the platform characteristics. facilitate and simplify the evaluation procedure for rail-
We performed our own on-site survey to collect informa- way companies. This was achieved through a discussion
tion on these characteristics, and estimated a model to with the railway companies and general passengers. The
predict the number of accidents using the platform fac- principles of selection were to allow quantitative evalua-
tors. The model shows the factors of a platform that tions, to use observable items from on-site surveys or
increase the number of accidents. Therefore, railway existing statistics, to allow the data to be collected easily,
companies can evaluate accident reductions when they to manage the number of factors, and to facilitate the
improve their platforms. Thus, this model is useful for comparison of one station to another. We concluded that
enhancing platform safety. four major factors exist: platform design, passenger flow,
train operation, and passenger profile. Each major factor
encompasses between three and five minor factors (Table
Selection of Evaluation Factors
2). The categorization based on the concepts and evalua-
Focusing on Platform Safety. This research focuses on safety tion factors was presented in the previous section.
in the case of accidental falling from a platform.
Evaluation factors were applied from a list presented in
a study published by Yamada et al. (12). It is noteworthy
On-Site Survey to Measure Evaluation Factors
that a significant risk is present around the stairs and On-Site Survey. An initial field survey was conducted in
escalators. Passengers may fall on the stairs or their shoe August 2015, and a supplemental survey was carried out
may become stuck on an escalator. If the stairs and esca- in March 2018. The investigator visited the platforms
lators are crowded, the above mentioned problems will and observed the safety factors shown as factors (1) to
increase. As accidents related to stairs and escalators are (13) in Table 2. The investigator paced the platforms or
not directly related to trains, we did not consider these used a ruler to measure the lengths for factors (1), (2),
factors here. This must be examined separately as a (4), and (8). He observed the platforms to record factors
pedestrian behavior problem or mechanical risk, rather (3), (5), (7), and (9)–(13). The number of daily passengers
than associated with platform safety. for factor (6) was calculated from the statistics listed on
528 Transportation Research Record 2673(1)

Table 3. Station Layout Arrangement

Island platform Side platform

Explanatory Variables

Island platform with opposite approach 1 0 0

Island platform with same approach 0 1 0

the companies’ websites and in the metropolitan transit island platform faces two rail tracks, whereas the side
census. The number of facilities for factor (15) was platform faces only one rail track.
counted manually on the map around the stations. The The ‘‘island platform with same approach dummy’’
safety factors of (14) and (16) were not used in this parameter is positive and statistically significant. The
research because the appropriate data were not available. island-shaped platform in which trains enter from the
same direction on both sides causes more accidents than
the other shaped platforms. We attributed this to the
Poisson Regression presence of several transferring passengers between the
We estimated the number of accidental fallings using the normal and express trains at these island platforms with
Poisson regression model. In general, we can assume that same approach. This also includes the case in which
the number of accidents that occur follows the Poisson many passengers transfer from a train that has termi-
distribution. Explanatory variables were chosen from nated at this platform onto another train that travels
Table 2. The result of the Poisson regression is shown in beyond the station at these types of platform.
Table 4. All of the variables are considered in model 1. Both the ‘‘platform curving is concave’’ and ‘‘platform
curving is convex’’ parameters had negative signs, imply-
ing that these curves cause less accidents. However, this
Factors Related to Platform Structure. The parameter for the
is contrary to our intuition. This is because the absolute
‘‘length of narrow part’’ is negative. This implies that
z value was small, and was therefore not statistically sig-
accidents are less likely to occur as the length of the nar-
nificant even at a 5% significance level.
row part becomes longer. However, this counteracts the
hypothesis because the absolute z value was small and it
was not significant at the 5% significance level. The z Factors Related to Passenger Flow. As the sign for the ‘‘log
test statistic used as the null hypothesis determined that (number of passengers)’’ is positive, the number of acci-
an individual predictor’s regression coefficient is zero dents increases as the number of passengers increases.
given that the rest of the predictors are in the model. The The parameter for ‘‘passenger flow crossing’’ is nega-
sign for the ‘‘width of gap between platform and train’’ is tive. This implies that the greater the number of intersec-
positive. This implies that if the gap between the plat- tions at the entrance and exit of the station, and at the
form and train is large, a falling accident will tend to line connecting the elevator facility and each door of the
occur. The sign for the ‘‘area of platform’’ is negative. train, the lower the number of accidents that will occur.
This implies that the larger the effective area of the plat- This is contrary to our intuition as the absolute z value
form, the lower the number of accidents. was small, and was therefore not significant at the 5%
The station layout arrangement (Table 3) affects significance level. As the parameter for the ‘‘Passenger
the number of accidental fallings. The parameter is nega- flow outside white line’’ is positive, accidents tend to
tive for the ‘‘island platform with opposite approach occur fences and platform doors have not been installed.
dummy.’’ This means that an island-shaped platform in However, as the absolute value for the parameter esti-
which trains enter from opposite directions on both sides mate is small, the influence is considered to be small. For
cause fewer falling accidents in comparison to a side example, the number of accidents will be 0.963 times
platform. This results was contrary to our intuition, and greater if the fence and the platform door are not main-
the estimated parameter was not statistically significant. tained above 10 m.
We thought that the passengers might fall from the other The ‘‘number of elevators’’ is used to calculate the
side of the platform after alighting from a train. In addi- ‘‘passenger flow crossing.’’ As the correlation between
tion, the passengers on the side platform seem to experi- the ‘‘number of elevators’’ and ‘‘passenger flow crossing’’
ence reduced risks compared to the island platform. The is high, it was not included in the explanatory variables.
 Table 4. Parameter Estimates by Poisson Regression and Negative Binomial Distribution Regression

Model 3: Model 4:
Model 1: Model 2: All variables negative Selected variables negative
All variables Poisson Selected variables binomial distribution binomial distribution
Explanatory variables regression model Poisson regression model regression model regression model

Intercept 25.081*** 24.662*** 25.610*** 24.498***

Log (number of passengers) 5.594 3 1022 1.071 3 1022
Length of narrow part 25.09 3 1022 21.179 3 1022* 21.08 3 1022*
22
Width of gap between platform and train 1.05 3 1022*** 1.13 3 10 *** 1.155 3 1022*** 1.00 3 1022***
Area of platform 23.389 3 1022 24.241 3 1022 24.82 3 1022
Platform curving is concave 22.833 3 1022 22.70 3 1022** 24.200 3 1022
Platform curving is convex 23.177 3 1022 24.00 3 1022* 22.145 3 1022
Passenger flow crossing 21.466 3 1022 1.057 3 1022*** 2.145 3 1022 2.033 3 1022*
Passenger flow outside white line 3.746 3 1022*** 3.561 3 1022* 3.575 3 1022** 3.62 3 1022**
Number of trains passing and stopping 1.401 3 1022 1.968 3 1022* 21.405 3 1022
Visual announcement of approaching train 6.031 3 1022 1.057 3 1022 1.282 3 1022
22
Audio announcement of approaching train 3.342 3 1022** 2.681 3 10 *** 3.990 3 1022* 3.527 3 1022*
Island platform with opposite approach 23.836 3 1022 23.998 3 1022 25.10 3 1022
Island platform with same approach 9.747 3 1022*** 1.122 3 1022*** 8.229 3 1022*** 7.862 3 1022***
Number of elderly passengers 4.039 3 1022 23.634 3 1022
Railway company A dummy 9.631 3 1022*** 7.96 3 1022*** 1.186*** 1.345***
Goodness of fit
AIC 1275.7 1267.6 1206.7 1200.1

Note: Statistical significance ***: 0.1%, **:1%, *:5%.

529
530 Transportation Research Record 2673(1)

Factors Related to Train Service. As the sign for the ‘‘number distribution of the Poisson distribution and the gamma
of trains passing and stopping’’ is positive, the higher the distribution. It can be said that the negative binomial
frequency with which trains enter the platform, the more distribution regression model loosens the highly restric-
likely it is that fall accidents will occur. This may be tive assumption that the variance is equal to the mean
attributed to many passengers boarding trains in which created by the Poisson model. A negative binomial distri-
the train entry frequencies are high. bution has a parameter called the ‘‘size parameter’’
As ‘‘visual announcement of approaching train’’ has a unlike the Poisson distribution. In a generalized linear
positive sign, accidents are more likely to occur with model using a negative binomial distribution, the size
more indicators informing passengers about approaching parameter can be estimated simultaneously as the para-
trains. This is contrary to our intuition. However, the meter estimation of the explanatory variable, and subse-
absolute z value was small and it was not significant at quently the actual data variation can be expressed.
the 5% significance level. As ‘‘audio announcement of Using a negative binomial distribution, we also cre-
approaching train’’ has a positive sign, accidents tend to ated a negative binomial distribution all variable model,
occur more frequently with an increase in the number of that is, a negative binomial distribution variable selection
times the broadcasting sounds informing passengers model with variable selection using the stepwise method.
about approaching/departing trains. The parameter estimation result is not significantly
different from the Poisson regression discussed above.
The ‘‘length of narrow part’’ that was insignificant in the
Factors Related to User Attributes. The ‘‘number of drunken
previous modeling was included in the variable selection
passengers’’ and the ‘‘number of visually impaired pas-
using the stepwise method. Two variables regarding the
sengers’’ were excluded from the explanatory variables
platform curving and the ‘‘number of trains passing and
because an appropriate source of data or statistics could
stopping’’ were excluded. In addition, as the AIC repre-
not be traced. As the sign for the parameter for the
senting the fitness of the model has a smaller value in the
‘‘number of elderly passengers’’ is positive, accidents are
negative binomial distribution regression, it can be said
more likely to occur as the number of elderly users
that the goodness of fit is improved.
increases. However, the absolute z value was small and it
was not significant at the 5% significance level.
Furthermore, as ‘‘railway company A dummy’’ was a Comparison of Regression Models to the Actual
significant variable, factors other than explanatory vari- Number of Accidents
ables were considered to have affected the number of
accidents occurring on company A and company B lines. Figure 4 shows the comparison of the negative binomial
Next, model 2 was estimated using selected variables. The and Poisson models with the actual number of accidents.
stepwise method was used for the variable selection. Ten vari- Since we count the number of platforms the number of
ables were considered statistically significant. Additionally, accidents for 1 year, the same platform but in a different
the estimated values for the parameters for ten variables did year is regarded as an individual platform.
not change significantly from model 1. The Akaike informa- The total number of accidents is zero on 542 plat-
tion criterion (AIC) was also declined; therefore, we con- forms. However, it is 507 in the Poisson model and 479
cluded that model 2 is better than model 1. in the negative binomial distribution model. The reason
for these discrepancies in the estimated number of acci-
dents is that the fractions generated when estimating the
Negative Binomial Distribution Regression Model number of accidents from the model were rounded up.
Next, a regression model for the number of falling acci- The models estimate the number of accidents as being
dents using a negative binomial distribution was gener- lower than the actual number at the zero-accident plat-
ated. The Poisson distribution is generally used to forms, and as being higher at the one-accident platforms.
represent the occurrence probability of a rare phenom- This means that these models underestimate at safer plat-
enon. The parameter for the Poisson distribution is the forms and overestimate at less safe platforms. Good
average number of occurrences of an event per unit time, insights can be obtained from these models.
and ‘‘variance is equal to average’’ is assumed. However,
the data used in this study is over-dispersed with large
Conclusion
variances with respect to the average, and does not sat-
isfy the assumption that "the variance is equal to the We developed a risk assessment model for train passen-
average" of the Poisson distribution. gers on a crowded platform. The risk investigated related
Therefore, we used a negative binomial distribution to the number of platform accidents such as falling from
that can represent the overdispersion of count data. The the platform to the railway surface. The estimated model
negative binomial distribution is an infinite mixture reflected the passengers’ behavior at a crowded platform,
Terabe et al 531

analysis and interpretation of results: T. Kato, S. Terabe, H.

Yaginuma, N. Kang, K. Tanaka; draft manuscript preparation:
S. Terabe, T. Kato. All authors reviewed the results and
approved the final version of the manuscript.

References
1. Cabinet Office, Government of Japan, White Paper on
Traffic Safety in Japan, 2017.
2. Rail Safety and Standards Board. Measuring Safety Per-
formance, 2011. www.rssb.co.uk/sitecollectiondocuments/
pdf/reports/research/T852_guide_final.pdf. Accessed July
27, 2013.
3. Fielding, G. J. Transit Performance Evaluation in the
U.S.A., Transportation Research Part A, Vol. 26A, No. 6,
1992, pp. 483–491.
4. Kecklund, L., M. Ingre, G. Kecklund, M. Söderström, T.
Akerstedt, E. Lindberg, A. Jansson, E. Olsson, B. Sand-
blad, and P. Almqvist, The TRAIN–Project: Railway
Figure 4. Comparison of negative binomial and Poisson models Safety and the Train Driver Information Environment and
with the actual number of accidents. Work Situation – A Summary of the Main Results, Pre-
sented at 2. Signalling Safety, London, UK, 2001.
and revealed that the risk increased as the passenger flow 5. Nelson, D. O., E. Martinez, and S. Bonina. Commuter
Rail: Balancing Freight Clearance with Level Boarding
outside the white lines increased, and/or when the gap
and Americans with Disabilities Act Accessibility Require-
between the platform and train was wider. This indicated ments. Transportation Research Record: Journal of the
that the passengers tended to walk close to the edge when Transportation Research Board, 2012. 2275: 120–129.
they wished to walk a shorter distance to avoid the crowd. 6. Transit Cooperative Research Program (TCRP) Research
Subsequently, the possibility of passengers falling into the Results Digest 85, Rail Passenger Safety: Equipment and
gap between the train and the platform becomes higher. Technologies. onlinepubs.trb.org/onlinepubs/tcrp/
The station layout arrangement and conflicts in passenger tcrp_rrd_85.pdf. Accessed July 27, 2013.
flows also affect the risk of accidents occurring. 7. Cothen, G. C. Jr., C. F. Schulte, J. D. Horn, and D. C.
These results will enable transit operators and station Tyrell. Consensus Rulemaking at the Federal Railroad
designers to compare the number of accidents that will Administration: All Aboard for Railway Safety Measures.
occur before and after a safety investment for a platform TR News, January–February 2005, Transportation
is established. It also helps to determine the type of a Research Board of the National Academies, Washington,
D.C., pp. 8–14.
safety investment considering the current condition of
8. Muttram, R. I. UK Railway Restructuring and the Impact
each railway platform.
on the Safety Performance of Heavy Rail Network. Japan
As most commuter railways in Japan enable passen- Railway & Transport Review, Vol. 34, 2003, pp. 4–11.
gers to board trains at the same level as the platform, it is 9. Tennyson, E. L. Rail Transit Safety Analysis: 1993, 1994,
not possible to directly translate this study to non-level and 1995. Transportation Research Record: Journal of the
platform boarding stations on commuter railways and Transportation Research Board, 1998. 1623: 112–117.
light railway. Generally, trains are 1–2 cm higher than 10. Hunter-Zaworski, K. Analysis of Passenger Incident Data
the platform level and this study did not consider this dif- from Five Rail Transit Systems. Safety, Vol. 3, No. 4, 2017,
ference in the level. However, it will become possible to p. 21. http://doi.org/10.3390/safety3030021.
apply this study after considering the difference in levels. 11. Hunter-Zaworski, K., D. Anderson, U. Rutenberg, E.
Tomaszewska, and J. McConnell. Manual to Improve Rail
Transit Safety at Platform/Vehicle and Platform/Guideway
Acknowledgments Interfaces. Transit Cooperative Research Program (TCRP)
We would like to thank Editage (www.editage.jp) for English Research Report 189, 2017.
language editing. We would also like to thank the anonymous 12. Yamada, S., S. Terabe, and M. Kasai, Safety Performance
reviewers for their valuable comments to improve the quality of Measures for Railway Stations, Presented at 93rd Annual
the paper. Meeting of the Transportation Research Board, Washing-
ton, D.C., 2014.
Author Contributions
The Task Force on Transit Safety and Security (AP018T) peer-
The authors confirm contribution to the paper as follows: study reviewed this paper (19-02278).
conception and design: S. Terabe; data collection: T. Kato;