Safety Science xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

Safety Science
journal homepage: www.elsevier.com/locate/safety

Evaluation of pedestrian safety margin at mid-block crosswalks in India

Chaudhari Avinasha, Shah Jitenb, Arkatkar Shriniwasa, , Joshi Gauranga, Parida Manoranjanc
⁎

a
Department of Civil Engineering, SVNIT, Surat, Gujarat 395007, India
b
Department of Civil Engineering, IITRAM, Ahmedabad, Gujarat 380026, India
c
Department of Civil Engineering, IIT, Roorkee, UK 247667, India

ARTICLE INFO ABSTRACT

Keywords: Amplified pedestrian-vehicle interaction especially at urban midblock crossings have ensued higher pedestrian
Safety margin fatalities and such situations further exacerbates under heterogeneous traffic conditions and therefore, it be-
Pedestrian comes indispensable to evaluate pedestrian safety especially at urban midblock crossings. To comprehend the
Gap acceptance factors influencing pedestrian safety, video-graphic survey was conducted at four urban midblock crossings
Mid-block
located in different parts of India. Data was extracted from the video and model was developed using Multiple
Rolling gap
Road crossing, vehicle speed, vehicle type
Linear Regression (MLR) Technique. The data includes mainly pedestrian demographics, vehicle characteristics
and roadway characteristics out of which, eleven variables are identified and analysed, which are found to be
significant. Regression model indicated that the PSM distinctly depends on the availability of gap in vehicular
flow, pedestrian crossing behaviour and speed of the approaching vehicle. Further, it is found that safety margin
increases with increase in pedestrian speed and vehicular gap, whereas the same decreases with rolling beha-
viour, while crossing the street. Elasticity analysis was also carried out among the different variables, wherein
the vehicle speed, pedestrian speed and vehicle gap were found to be the most influencing variables, whereas,
pedestrian age and platoon size had the least effect. It was also found that pedestrians are more likely to cross the
road by maintaining least marginal vehicular gap with respect to light vehicles compared to heavier vehicles,
highlighting that vehicle type influences pedestrian safety. Further, driver yielding behaviour significantly in-
fluences pedestrian safety. In addition, from the sensitivity analysis, it is corroborated that the safety margin
against to pedestrian speed and vehicle speed increases with increase in the vehicle gap size. Moreover, PSM
against waiting time decreases as the vehicle gap increases. The developed PSM model can enable engineers and
planners to evaluate safety at existing crossings at unprotected mid-blocks. The results of the PSM model may be
useful to design pedestrian facility and suggest appropriate remedial measures to enhance pedestrian safety.

1. Introduction motorised vehicle traffic, pedestrians’ vulnerability increases. Kumar

and Parida (2011) found that, 54 percentage of pedestrian-road acci-
Individual pedestrian judgment, regarding when and where to cross dents involved, during road crossing in urban areas. Hence, providing
the road is a very complex phenomenon, especially at prevailing un- safe and efficient pedestrian road crossing facilities is a long-established
protected and non-complying pedestrian behaviour conditions in India. goal in cities of India.
Normally, it is represented by various factors such as comfort, con- In developing countries like India, due to absence of foot over
venience, ease of crossing, and prevailing safety-level. To reach the bridge (FOB) or subways in general, pedestrians forced to cross at grade
destination on the other side of road, the pedestrian has to cross particularly, at mid-block. In addition, pedestrians ‘non-complying
somewhere at mid-block location. While crossing, the pedestrian be- walking behaviour leads to more number of conflicts at mid-block lo-
haviour changes dynamically based on the prevailing conditions. cations, which further leads to occurrence of severe accidents. A study
Several researchers had attempted to identify the factors influencing carried by Indian Institute of Technology (IIT) Delhi, India titled as
pedestrian behaviour including sudden decisions that affect pedestrian “Road safety in India: challenges and opportunities” states that 58% in
walking characteristics such as choice to accelerate or decelerate Mumbai and 47% in Delhi in urban areas are related to pedestrian’s
walking speed, stop or wait and where to cross a street (Ishaque and fatalities, which occur on road (Mohan et al., 2015). For this reason, it
Noland, 2008). Due to unprotected crossing of road and increase in is clear that pedestrian safety is a major issue for transport planners,

⁎
Corresponding author.
E-mail addresses: jitenshah@iitram.ac.in (S. Jiten), sarkatkar@ced.svnit.ac.in (A. Shriniwas), gj@ced.svnit.ac.in (J. Gaurang).

https://doi.org/10.1016/j.ssci.2018.12.009
Received 6 January 2018; Received in revised form 20 September 2018; Accepted 10 December 2018
0925-7535/ © 2018 Published by Elsevier Ltd.

Please cite this article as: Avinash, C., Safety Science, https://doi.org/10.1016/j.ssci.2018.12.009
C. Avinash et al. Safety Science xxx (xxxx) xxx–xxx

traffic engineers and policy makers. Hence, it is worth studying the road arrived at the same conflict point (Chu and Baltes, 2001). Negative
crossing behaviour of pedestrians under mixed traffic conditions at safety margins suggest that pedestrians accept gaps shorter than their
uncontrolled mid-block locations. Researchers have investigated pe- crossing time, and therefore, drivers would have to take evasive action
destrian road crossing behaviour, decision making process of pedes- to avoid a collision. However, negative safety margins correspond to an
trians as well as vehicular characteristics with the help of discrete accident and are not expected in a real crossing event always. Thus, a
choice models (Himanen and Kulmala, 1988; Sun et al., 2003; Lassarre safety margin is used as an indicator of relative safety, with a greater
et al., 2007) and log-normal models for investigating minimum gap at safety margin representing a safe gap or a safe crossing decision
mid-block (Yannis et al., 2010; Kadali and Vedagiri, 2012). (Zhuang and Wu, 2012). Recent studies found that pedestrians'
Further, field experiment in a naturalistic traffic environment was minimum PSM value while crossing the road depends on the vehicular
conducted to measure pedestrians’ characteristics such as crossing gap size; pedestrian speed, age and rolling gap (Kadali and Vedagiri,
speed for a given vehicle speed and also stopping distance and yielding 2013). Further, results from other studies showed that pedestrian be-
behaviour, when they are crossing streets. From these experiments, it havioural characteristics significantly affect the PSM value as well as
was observed that pedestrians’ performance change at different actual the probability of PVNC (Pedestrian Vehicle Non Conflict) behaviour.
speed levels and also under different weather conditions. The results The rolling behaviour of a pedestrian significantly decreased the pe-
show that pedestrians have accurate estimation of speed intervals, that destrian safety margin values (1 s) as well as the probability of PVNC
vary by weather conditions. When the speed of the ongoing vehicle (35%) with an approaching vehicle, while crossing the road. In addi-
exceeded the upper bound of the accurate interval, pedestrians are tion, the probability of PVNC increased with an increase in pedestrian
more likely to underestimate the vehicle speed, thereby increasing their platoon size. Pedestrians are more likely to maintain least margin va-
risk of incorrectly deciding to cross, when it is not safe to do so (Sun lues with small vehicles such as two-wheelers (2.06 s) and auto-rick-
et al., 2015). Also, a lognormal regression model was developed to shaw (2.69 s) under mixed traffic conditions (Kadali and Vedagiri,
examine pedestrian gap-acceptance behaviour. It was found that gap- 2016).
acceptance is better explained using the distance from the upcoming There are very limited studies carried out at unmarked and un-
vehicle, rather than its speed. Other significant effects included illegal protected mid-block locations in mixed traffic context with pedestrian
parking, presence of other pedestrians and incoming vehicles’ size. behavioural tactics. In the present study, PSM evaluated through pe-
Also, binary logistic regression model was developed to examine the destrian gap acceptance considering the effect of pedestrian beha-
effect of traffic gaps and other parameters on pedestrians' decisions to vioural tactics such as rolling gap (Brewer et al., 2006). On urban roads,
cross the street or not. The results reveal that this decision is sig- different types of vehicles move on the same road space, occupying any
nificantly affected by the distance from the incoming vehicles and the position on the road depending on availability of space at a given in-
waiting times of pedestrians (Yannis et al., 2013). Other studies have stant of time, without maintaining much lane-discipline. Due to mix of
employed three different methodologies for studying Pedestrian Gap different types of vehicles, absence of crossing facilities and traffic
Acceptance (PGA), the first model is deterministic and solely depends control system, traffic situation leads to noncompliant behaviour of
on the gap sizes that are accepted or rejected by pedestrians. The vehicle-driver as well pedestrians. It results in the hazardous pedestrian
second model is probabilistic and the probability of accepting a gap, crossing and also increased share of road crashes involving pedestrians.
which is calculated as a random variable from a distribution, that best With this background, the objective of this research is to evaluate the
fits the data. The third model uses a binary logit approach; multi-at- pedestrians' safety at unprotected unmarked mid-block locations in
tribute regression analyses are performed to capture the decision different cities of India.
making process of the pedestrian. The results show that PGA binary
logit models perform better than the other models. Moreover, addi- 1.1. Pedestrian safety margin (PSM)
tional insights are provided into the PGA behavior based on the data
(Sun et al., 2005). Safety Margin (SM) is defined as the difference between the time a
Most of the available studies are focused more on pedestrian gap pedestrian crossed the conflict point and the time the next vehicle ar-
acceptance behaviour, however the proposed study dealt with safety rived at the same conflict point (Chu and Baltes, 2001). Suppose a
margin concept based on real field data. Whereas, some studies have pedestrian reaches a conflict point at time T1, and the next vehicle
been carried out in virtual environment to examine the effect of pe- arrives a same conflict point at time T2, then the SM is T2–T1. It in-
destrian demographic and vehicular characteristics on decision making dicates that if the pedestrian is having a slower walking speed (T1 is
process, while crossing road (Oxley et al., 2005; Lobjois and Cavallo, more time) or the vehicle reaching faster (T2 is less time) it leads to
2007) and these results were correlated with pedestrian safety, though hazardous condition. Necessarily, it is the excess time when deducting
this is not fully realistic. Many studies often investigated the pedestrian the perceived time from available vehicular time gap during road
decision making process for road crossing, which depends on the dis- crossing. There are several studies using this concept of SM and they
tance of the vehicle or vehicular speed (Sun et al., 2003; Yannis et al., defined it in different ways (Chu and Baltes, 2001; Oxley et al., 2005;
2010). As a consequence, pedestrians select an unsuitable vehicular Lobjois and Cavallo, 2007; Tung et al., 2008; Guangxin and Keping,
gap, which leads to hazardous crossing. In this line, recent studies re- 2009).
ported that both types of data such as speed and distance of the vehicle The SM value extraction is a two-step process: in the first step, the
are really very important to make crossing decisions (Tung et al., 2008). pedestrian just crosses the conflict point. In the second step, the next
Studies were also carried on effect of type of vehicle on pedestrian oncoming vehicle just passes through the same conflict point, which is a
safety and it was found that the light-vehicles are more dangerous than virtual intersection of lines indicating the pedestrian's crossing and
heavy vehicles (Lefler and Gabler, 2004). Moreover, studies carried out vehicular movement paths. Therefore, the differences among these two
on pedestrian safety evaluation indicated, age as the main impact factor time steps are measured as SM, which is explained and well depicted in
(Chu and Baltes, 2001; Oxley et al., 2005; Holland and Hill, 2007; Fig. 1. The values are recorded for each of lanes, at a contemplated
Cavallo et al., 2009; Holland and Hill, 2010). A safety margin is defined conflict point during the vehicle-pedestrian interaction; out of these,
as the amount of time from the moment the pedestrian crosses the minimal values are taken in the model.
virtual conflict zone, until the next vehicle reaches the same zone
(Cavallo et al., 2009; Lobjois and Cavallo, 2007; Oxley et al., 2005; 2. Details of study locations and methodology:
Tung et al., 2008).
Safety Margin (SM) is defined as the difference between the time a The study is carried out at four mid-block sections on four-lane di-
pedestrian crossed the conflict point and the time the next vehicle vided roads located in the Western and Northern part of India, in three

2
C. Avinash et al. Safety Science xxx (xxxx) xxx–xxx

Fig. 1. Geometrical definitions associated with Safety margin. Where, T1 = Pedestrian accept the Gap; T2 = Pedestrian just crosses the conflict point; T3 = Next
oncoming vehicle in a vehicular path with Vehicle Gap; T4 = Next oncoming vehicle has just passed through the same conflict point as virtual line indicating the
pedestrian's crossing path and vehicular path.

Table 1
Details of study locations with traffic flow data.
Loc. No. Study Location Type Pedestrian Volume (Ped/h) Traffic flow (PCU/h) Avg. Traffic Speed (Km/h)

I Vile Parle, Mumbai Un-marked 818 3432 23.63

II Bandra, Mumbai Un-marked 711 2674 23.99
III Neelam Cinema Chandigarh Marked 944 3532 29.61
IV Ashdodia Ahmadabad Marked 510 3612 27.50

different cities. The details of the study locations are presented in comparison test was performed using Post Hoc technique(Student-
Table 1. Newman-Keuls) to check whether the extracted data set belongs to one
Video-graphic survey was conducted at selected marked/unmarked group or not. The descriptive statistics of the POST HOC test is pre-
crosswalk locations on a working day under clear weather conditions sented in Table 3. For performing post-hoc test, location wise coding
during mid-June, October and November 2015. Specifically, care was used is shown in the Table 4. The purpose of this test was to identify the
taken that, for the unprotected pedestrian condition, observations are group of extracted data (with no significant difference), which are put
made more rigorously on pedestrian-vehicular interaction, particularly together for developing a multiple regression model, one was used for
during time duration, when indication of upstream traffic signal, is calibrating and another was used for checking model validation and
green. To determine the vehicular speed from the recorded videos, a transformation.
trap length of known distance was marked at the entry and exit of study From Table 3 it has been observed that there is a significant dif-
locations along the transverse direction of the road. Two high-resolu- ference between the mean safety margin value for location IV with the
tion digital video cameras (frame rate: 29 frames per second; 55X rest of location, which may be attributed to higher vehicle volume
zoom) were deployed for collecting video-graphic data. One was kept at (Table1) and the vehicular gap available is lower than adequate. The
a vantage point of 5 m height from the ground level to capture move- differences in safety margin between the location averages about
ment of vehicles and pedestrians on the midblock road segments and 31.40% with location III has 60.13% higher value than the location IV.
the second one near the crosswalk to capture the pedestrian char- The test of variance was also performed on data extracted for SM values
acteristics as depicted in the Fig. 2. From the recorded videos, pedes- at different locations. The null hypothesis and alternative hypothesis for
trian crossing behaviour data was extracted at microscopic levels using SM values measured for different locations were considered to have
AVS video extractor software having an accuracy of 33 ms for further equal variances and unequal variances, respectively at confidence level
statistical analysis. of 95%. The result obtained from the statistical test is presented in the
The demographic composition of pedestrian relating to age and Table 4.
gender was identified by visual perception using video-graphic data. From the Table 4, it has been observed that the mean value of safety
The age band is classified in three groups as children (age < 15), margin (0.889) at location IV (Ashdodia Ahmedabad) differed sig-
younger pedestrians (age between 15 and 30), middle edge nificantly from mean values at other locations. Therefore, Locations-I, II
(30–50 years), elders (> 50 years). The extracted data comprised of and III are clustered in one group with minimum variation and the
pedestrian demographic characteristics such as gender and age groups, location-IV is treated as a separate group with wide variation in SM
platoon size, waiting time and SM value. Whereas, the vehicular traffic value. With these results, it was decided that the extracted data of lo-
characteristics extracted, are vehicle speed, vehicle type and available cations I, II and III are used for model formation, where as the extracted
gap sizes. Additionally, pedestrian behavioural tactics were also con- data of location IV is reserved for model validation. To check the nor-
sidered, including the effect of rolling-gap behaviour. mality of the extracted data of safety margin values; Q-Q plots for all
Further, referring the Fig. 1, SM values are calculated at each of the locations are plotted and presented in the Fig. 3.
lanes, at the potential conflict point during the interaction between the From the Fig. 3, it may be observed that the linearity of points be-
pedestrian and vehicle, out of these minimal values are taken for for- tween the observed value and expected normal value of safety margin
mation of model. Most of the related independent variables, which af- are nearly equal, means the data is normally distributed. Further, the
fect the safety margin, are considered for checking statistical sig- procedure of modeling starts with a correlation analysis to know the
nificance, which is presented in Table 2. correlation between the independent variables and SM as dependent
For the purpose of deriving significant difference in Safety margin variable. Bivariate (Pearson Correlation coefficient) analysis was also
values based on data extracted from selected locations, multiple performed using Statistical Package for the Social Sciences (SPSS 16.0).

3
C. Avinash et al. Safety Science xxx (xxxx) xxx–xxx

Location-I Location-II

Location-III Location-IV
Fig. 2. Details of the study locations.

Further, Multiple-Regression Model was also developed to predict adjusting themselves across different lanes in a traffic flow, which leads
minimum safety-margin value. Initially, eleven variables were con- to wrong judgment of vehicular gaps and hence, SM reduces. Further, if
sidered in the model; out of these, nine are identified to be the most pedestrians frequently use rolling gap, their speed reduces and these are
significant for the regression model for SM value at unmarked mid- very dangerous or unsafe. Longer waiting time implies increasing pe-
block locations. Correlations between the variables are depicted as destrian group/platoon size, reducing their speed, attempting small
follows in Table 5. gaps by accepting rolling over gap and forcing the approaching vehicle
Table 5 indicates that, speed of pedestrian is higher when they walk to reduce the vehicle speed and reduce SM. Effect of age on PSM at
as individual instead of group behaviour. Single pedestrian with larger unprotected mid-block crosswalk under mixed traffic conditions has
gap available crosses the street with a higher speed before approaching significant effect. The elder pedestrian experiences lower safety in their
the vehicle at conflict point with higher safety margin. It implies that, road crossing as compared to the rest, because elder pedestrian crosses
as speed of approaching vehicle increases, pedestrian increases crossing the road with lower walking speed taking higher crossing time and high
speed and tries to cover the crossing distance as early as possible risk. Driver yielding behaviour to pedestrian is observed with two-
without using the rolling behaviour. As platoon size or grouping be- wheeler by changing the path, reducing the speed and for Car, by re-
haviour increases, the SM value would be more, but due to increased ducing the speed and provide right of way to the pedestrian, whereas
platoon size or group behaviour the average pedestrian speed will re- the female pedestrians feel safer, when the approaching vehicles are of
duce and SM will reduce. Increased platoon size increases the use of heavy type. It is because the approaching speed of heavy vehicle is low
rolling gap. Due to increased waiting time, pedestrians are impatient compared to light vehicle. The descriptive statistics of different vari-
and start crossing the road with smaller available vehicular gap by ables considered for the model formation are given in above Table 2.

Table 2
Descriptive Statistics for the variables used in the Multi Linear Regression model.
Variable name Type Unit/Code Minimum Maximum Mean Std. Deviation N

Pedestrian Safety Margin Continuous Time in sec −2.98 9.98 1.213 0.539 4885
Pedestrian Speed Continuous m/sec 0.11 3.20 1.083 0.292 4885
Vehicular Gap size Continuous Time in sec 0.90 17.42 3.023 0.775 4885
Vehicle Speed Continuous Km/h 10.55 110.0 21.69 13.11 4885
Pedestrian waiting time Continuous Time in sec 0.00 170.0 8.961 14.36 4885
Gender Male 0 67.15%, By visual appearance
Female 1 32.85%
Age Child: < 15 0 5.63%
Young: 16–30 1 21.49%
Middle 31–55 2 61.08%
Elders: > 56 3 11.78%
Rolling behaviour Yes 1 22.70% Whether a pedestrian rolls over the available small vehicular gaps Pedestrian
No 0 77.30%
Pedestrian platoon Single 0 32.5%, Number of pedestrians in the group while crossing the road
Two 1 29.5%,
Three or more 3 38%
Type of vehicle Two wheeler 0 23.03%, The type of approaching vehicle corresponding to the accepted gap by pedestrian
Three wheeler 1 28.46%,
Car/SUV 2 41.7%,
Heavy vehicle 3 6.81%

4
C. Avinash et al. Safety Science xxx (xxxx) xxx–xxx

Table 3
Descriptive statistics of POST HOC (Student-Newman-Keuls) Test.
Loc No N Samples Mean Std. Deviation Std. Error 95% Confidence Interval for Mean Minimum Maximum

Lower Bound Upper Bound

I 2009 2.250 1.986 0.04430 2.1611 2.3348 −1.22 8.88

II 2913 2.210 2.084 0.03862 2.1327 2.2842 −1.00 8.22
III 744 2.070 2.247 0.08202 1.8934 2.2154 −1.10 8.80
IV 1097 0.889 1.479 0.04464 0.8019 0.9771 −1.20 6.94

N: Number of samples.

3. Multiple regression model 3.1. Model validation

Multiple regression model is useful to find the SM value maintained The pedestrian safety model was validated using total 1097 data
with an approaching type of vehicle, by pedestrian while crossing the points from unprotected mid-block crosswalk location-IV(Asthodia,
road. The safety margin values vary with pedestrian behaviour and Ahmadabad), which were pooled out due to significant difference in SM
ranges from negative values to positive values and was found to be best values. Based on the model results, significant variables were collected
fitted using Normal distribution. The Multiple regression model is the from the selected location for the model validation, the predicted values
most common modelling technique for predicting the response of the were calculated by substituting the values of variables in the obtained
linear continuous dependent variable for developing a relationship model and comparing with the observed values. From the developed
between independent variables and dependent variable. The model model, the correct success prediction was observed as 0.85 and root
framework adopted is given below: mean square error is 0.422, as presented in Table 8 and this indicates
that the difference between the observed mean value (1.213) and pre-
PSM = + 1 X1 + 2 X2 + 3 X3 + ………+ n Xn
dicated mean value (1.4034) is less.
0

PSM = Pedestrian Safety Margin Values; X1-n = explanatory variables; Further, the graph was plotted between observed and predicted PSM
β1-n = estimated parameters from the model; β0 = y-intercept of re- values a valid R2 value has been found as 0.663 and is shown in Fig. 4.
gression line. The results proved that the developed MLR model has better prediction
A stepwise regression was conducted for all the descriptive vari- capabilities for estimating the minimum safety margin for the pedes-
ables, with SM as dependent variable. The analysis aimed to find out trian road crossing behaviour at uncontrolled midblock crosswalk sec-
the most contributing factors among all the listed potential predictors tion in Indian metropolitan cities.
on PSM after considering partial correlations. The model fitting was According to Lewis scale of interpretation, estimation of accuracy
found to be statistically significant (Refer Table 6) using F-test with (Kenneth and Ronald, 1982) for any forecast with a MAPE value of less
selected independent variables. Table7 shows detailed regression ana- than 10% is to be considered highly accurate. 11–20% as good, 21–50%
lysis results along with coefficients, t-value and p-value and confidence as reasonable and 51% or more as inaccurate. The obtained value of
intervals. Total nine variables are found to be significant for modeling MAPE is 12.95% and hence, the predicted model is proven to be per-
SM as responding variable using multiple regression analysis. forming good.
The goodness-of-fit measure, coefficient of determination, that is R2
was found to be equal to 0.797 and adjusted R2 = 0.597, indicating
4. Discussions on different variable on PSM
59.70% variance as explained using nine predictors for this model. In
addition, all nine variables were found to be statistically significant at
4.1. Demographic
95% confidence level. The final model (Eq. (1)) then derived is given as
follows:
From the developed PSM model results, it is found that effect of age
PSM = 0.4551 + 1.162 PS + 0.0120 VS + 0.0929 VGAP on PSM at unprotected mid-block crosswalk under mixed traffic con-
+ 0.0337 AGE 0.005 WT + 0.0181 PPSIZE ditions is quite significant. When the adequate vehicular gap is not
available with high traffic flow, the male pedestrian crosses the road,
0.009 R_BEH 0.0209 VTYPE + 0.0252 DRYBEH (1)
taking risks with rolling behaviour by adjusting their position lane by
where VGAP = Vehicular Gap; PS- Pedestrian Speed m/s; VS = Vehicle lane with available vehicular gaps, whereas female pedestrian waits for
Speed Km/h; PPSIZE = Pedestrian Platoon size; WT = Waiting time; adequate available vehicular. Further female pedestrian feels safe while
R_BEH = Pedestrian rolling behaviour; AGE = Age of Pedestrian; crossing the road, when the approaching vehicle is a heavy vehicle. The
DRYBEH = Driver Yielding Behaviour VTYPE = Vehicle type elder pedestrian experienced lower safety in their road crossing as

Table 4
POST HOC (Student-Newman-Keuls) Test.

5
C. Avinash et al. Safety Science xxx (xxxx) xxx–xxx

Fig. 3. Location wise Q-Q plots for SM values.

Table 5
Correlations Between the variables.
S. N. Variable Positively Correlated Negatively Correlated

01 SM PS (r = 0.717), Age (r = 0.036), VG (r = 0.060), VS (r = 0.498), Dr Beh = 0.032) PPZ (r = −0.114), Rgap (r = −0.021), WT(r = −0.020).
VT(r = −0.012),
02 PS VS (r = 0.325), VG (r = 0.011), age (r = 0.002), Dr.beh (r = 0.018, p < 0.01) PPZ (r = −0.235), Rgap (r = −0.019), WT (r = 0.047) VT(r = 0.002)
03 VS Vgap (r = 0.073), age (r = 0.006), VT (r = 0.111), Dr.Beh (r = 0.031), PPZ R gap (r = −0.019), WT (r = −0.043).
(r = 0.46)
04 Age VT (r = 0.059), Dr Beh (r = 0.039) PPZ (r = −0.208), Rgap (r = −0.003), WT (r = −0.094).
05 WT PPZ (r = 0.096), Rgap (r = 0.034) VT (r = −0.077), Dr.beh (r = −0.001).
06 PPZ VT (r = 0.127) Rgap (r = −0.013), Dr Beh (r = −0.018)
07 Rgap VT (r = −0.033), Dr Beh (r = −0.018)

Note: SM-safety margin, PS-pedestrian speed, VG-Vehicle gap, WT-waiting time, VS-Vehicle speed, Rgap-Rolling gap, VT-Vehicle Type, PPZ- Pedestrian platoon size
Dr.Beh = Driving behaviour.

Table 6 consistent with results reported by researchers (Oxley et al., 2005;

F-test on Pedestrian safety margin model fitting results. Lobjois and Cavallo, 2007). Irrespective of the above discussion, it is
Model Sum of Squares df Mean Square F Sig. also found that gender has no significant effect on SM because of var-
iations in behavioural characteristics of pedestrians while crossing the
Regression 850.12 9 141.687 1208.20 0.00 road based on the approaching vehicle.
Residual 572.16 4876 0.117
Total 1422.282 4885

4.2. Effect of pedestrian rolling gap on PSM and speed

compared to the other age groups, as elder pedestrian crosses the road
with lower walking speed, taking higher crossing time, and hence al- In case of 4-lane divided road with high traffic volume, the pedes-
ways exposed to higher risks, as depicted in Fig. 5(c) with maximum trian used the rolling behaviour by accepting suitable gap lane by lane
variance against all types of vehicles. The results are found to be by reducing the speed, there by utilizing every possible chance, in-
dicating aggressive crossing behaviour, forcing the driver of the

6
C. Avinash et al. Safety Science xxx (xxxx) xxx–xxx

Table 7
Pedestrian safety margin model summary.
Variable Estimated coff. (β) Std. Error t Sig. 95.0% Confi. Interval for β VIF

Lower Bound Upper Bound

Constant −0.4551 0.054 −8.404 0.000 −0.561 −0.349

Pedestrian Speed in m/s 1.1621 0.018 63.151 0.000 1.126 1.198 1.207
vehicle speed in km/h 0.0120 0.000 29.84 0.000 0.011 0.013 1.155
Vehicle Gap 0.0929 0.006 5.120 0.000 0.020 0.045 1.029
Pedestrian age 0.0337 0.008 4.126 0.000 0.018 0.050 1.059
Waiting time −0.0005 0.000 −1.306 0.092 −0.001 0.000 1.057
Platoon Size 0.0181 0.006 3.287 0.001 0.007 0.029 1.014
Rolling behaviour −0.0093 0.033 −0.286 0.077 −0.073 0.054 1.033
Vehicle Type −0.0209 0.005 −3.847 0.000 −0.031 −0.010 1.055
Driver behaviour 0.0252 0.030 0.834 0.004 −0.034 0.084 1.014

The Significance at 95% confidence interval with P-value < 0.05; VIF: variation inflation factor.

Table 8 hence more crossing time. Earlier studies also reported that, elders have
Residual statistics by predicted model. more risk while crossing than younger pedestrians (Oxley et al., 2005).
Calibrated model Minimum Maximum Mean Std. deviation Age is positively correlated with SM and it indicated that younger as
well as Middle age pedestrian are generally safe due to reasonable
Predicated value 0.5526 9.1208 1.4034 0.4715 crossing speed.
Standard Predicated value −1.804 16.365 0.000 1.0000
Standard error of predicated 0.001 0.002 0.000 0.0000
value
Residual −0.0085 0.0102 0.000 0.0040
4.4. Effect of platoon size on pedestrian safety margin
Root mean square error 0.422
R2 for predicated model 0.85 Pedestrians perform individual noncompliant behaviour using
rolling gaps, because of longer waiting time at kerb due to high traffic
flow and unavailability of adequate vehicle gap. Here, pedestrians lead
5
to hazardous situation, while crossing. When pedestrians cross the road
4.5 in platoon size or group, they reveal significantly different behaviour.
4 In group, they use rolling gaps and attempt to accept smaller vehicular
Predicated PSM Value

3.5 gaps. In this condition, drivers have to stay alert by either reducing
their vehicle speeds and forced to stop their vehicles many times, of-
3
fering right of way for pedestrian movement. In such conditions, with
2.5 R² = 0.6638
increase in platoon size, there is negative effect on SM. The same result
2 was also supported by Leden (2002); reported that pedestrians’ risk
1.5 decreases with increase in platoon size.
1
0.5
4.5. Effect of vehicular gap size on pedestrian safety margin
0
0 1 2 3 4 5 6
Vehicular gap size has a positive effect on SM. Increased gap size
Observed PSM Value
increases the SM. The gap size is negatively correlated to waiting time.
Fig. 4. Validation of Safety margin model. Due to increased waiting time, an individual pedestrian prefers to
choose lower gap size or vehicular gap less than safe gap which leads to
approaching vehicle to either reduce the speed or halt the vehicle, wrong judgement of vehicular gaps and affects safety.
consequently. This behaviour is predominantly revealed by pedestrians
in group/platoon, while crossing the road. Fig. 5(a) and (b), depict that
4.6. Vehicular characteristics
rolling gap has negative influence on pedestrian safety margin. Results
show that the pedestrians’ unsafe crossing tendency increases with in-
In developing countries like India, maximum modal share is gen-
crease in use of rolling gaps, which results into significantly lower va-
erally of two-wheelers, then, car/three-wheelers and minimum of heavy
lues of SM. Consequently, the available gap size is ought to be very
vehicles, on urban roads. Pedestrians are more likely to cross the road
small, while pedestrians cross with rolling gap, but since crossing time
by maintaining marginal vehicular gaps, in front of small and light
is high during rolling behaviour the SM values are found to be in the
vehicles (two-wheelers and three-wheelers) as compared to heavy-ve-
range of −2.280 s to 1.22 s.
hicles (truck/bus), which is presented in Fig. 6(a–c). The pedestrians
are more cautious, when the approaching vehicle is a high-speed ve-
4.3. Effect of age on pedestrian safety margin hicle (car or two-wheeler). It is because of aggressive behaviour of
drivers in such situations.
Fig. 5(c) depicts observations that young and middle-aged pedes- Fig. 6(a) through (c) depict that female pedestrians feel safer, when
trians cross with more safety than child and elderly-age pedestrians, the approaching vehicles are heavy vehicles. It is because the ap-
which is because of reasonably higher crossing speed. On the contrary, proaching speed of vehicles is lower, when compared to light-vehicle
elders and children usually perceive more unsafe crossing than younger whereas male pedestrian takes a higher risk with lower safety. Further,
and middle age pedestrians due to lower crossing speed. Age difference mean SM depends on vehicle type and it is found to be lower for light-
variable play an important role in pedestrian safety as elders are more vehicle due to higher speeds, but significantly higher for heavy ve-
likely to have lower safety, because of the slower walking speed and hicles.

7
C. Avinash et al. Safety Science xxx (xxxx) xxx–xxx

2.8 2.0
2.6
1.8
2.4
2.2 1.6
2.0
1.8 1.4

Pedestrian Speed m/s

1.6
Safety of Margin

1.2
1.4
1.2 1.0
1.0
0.8
0.8
0.6 0.6
0.4
0.2 0.4

0.0
0.2
-0.2
-0.4 0.0
1 0 0 1
Type of Moment Type of movement

Note: 0 –Single without stopping 1-Rolling with stop in carriageway

(a) Effect of Pedestrian rolling gap on (b) Effect of Pedestrian rolling gap on
safety margin pedestrian speed

2.8
2.6
2.4
2.2
2.0
1.8
Safety of Margin

1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
-0.2
-0.4
0 3 2 1

Pedestrian age

Note: 0 –Child, 1- Young , 2- Middle age ,3-Elder

(c) Effect of Pedestrian age on safety margin

Fig. 5. (a) Effect of Pedestrian rolling gap on safety margin. (b) Effect of Pedestrian rolling gap on pedestrian speed. (c) Effect of Pedestrian age on safety margin.

5. Elasticity analyses according to the following formula (Eq. (2)):

e= ( Yi/ Xi)(Xi/Yi) = i (Xi/Yi) (2)
Elasticity (e) analysis was also carried out and e of the dependent
variable with independent variable, for each pedestrian was calculated The results obtained from elasticity analysis are presented in

2W 3W CAR HEAVY
2.500

2.000
Safety margin(s)

1.500

1.000

0.500

0.000
CHILD YOUNG MIDDLE ELDER

Fig. 6. (a–c) Safety margin for gender/age wise as a function of type of vehicle.

8
C. Avinash et al. Safety Science xxx (xxxx) xxx–xxx

Table 9
Parameter estimates, statistically significant and elasticity values in Multiple Regression model.
Independent variable Estimated coefficient (β) p-value < 0.05 Elasticity (e) Elasticity (e*)

Pedestrian Speed in m/s 1.1621 0.000 0.569 16.722

vehicle speed in km/h 0.0120 0.000 0.124 3.651
Vehicle Gap 0.0933 0.000 0.309 9.079
Pedestrian age 0.0337 0.000 0.028 0.832
Waiting time −0.0005 0.092 −0.002 −0.057
Platoon Size 0.0181 0.001 0.025 0.732
Rolling Behaviour −0.0093 0.077 0.000 −0.009
Vehicle Type −0.0209 0.000 −0.034 1.000
Driver behaviour 0.0252 0.004 0.005 0.142

ps-0.5 (m/s) ps-1(m/s) 20 kmph 40 kmph

5 ps-1.5 (m/s) ps-2 (m/s) 2.75 60 kmph 80 kmph
ps-2.5 (m/s) 2.55
4
SafetyMargin (s)

Safety Margin(s)
2.35
2.15
3
1.95
2 1.75
1.55
1 1.35
1.15
0 0.95
2.00 4.00 6.00 8.00 10.00 12.00 0.00 2.00 4.00 6.00 8.00 10.00
Vehicle gap (s) Vehicle gap(s)
Fig. 7. (a and b) Sensitivity of SM based on vehicle gap-pedestrian speed/vehicle speed.

Table 9. The relative effect (e*), as a normalization of the estimated values of an independent variable influencing on a particular set of
elasticity about the lowest elasticity, was also calculated to quantify dependent variable under given set of assumptions. This analysis gives
clearly, to what extent each of the independent variables, affect the notion about the impact of that variable on pedestrian safety margin,
dependent variable. while crossing the road using MLR analysis. It is found that vehicle gap,
From Table 9, it is observed that the pedestrian speed has maximum pedestrian speed, vehicle speed and waiting time are more influencing
effect on pedestrian safety. The pedestrian age depicts significant effect factors on pedestrian safety margin. During the high traffic volume
on safety; elder pedestrian feels more unsafe, while crossing the road as pedestrian waits at curb or median, waiting time increases; because of
compared to the rest, as they take longer crossing time to cross, with non-availability of adequate vehicular gap and non-compliance yielding
lower crossing speed. Pedestrian platoon size also has significant impact behaviour of the driver, pedestrians are impatient and tend to cross the
on elasticity in group behaviour. The pedestrian group feels safe to road in stage crossing or with rolling behaviour by lowering the
cross the road than an individual. The yielding behaviour of the driver crossing speed and leads to enhanced risk of collision. Studies have
has also significant effect on pedestrian safety. In developing countries, shown that there is an increase in the risk taking behaviour with an
driver yielding behaviour towards pedestrian is generally observed for increase in pedestrian waiting time at signalized intersections (Hamed,
two-wheelers by changing the path, reducing the speed, whereas car 2001). Sensitivity of pedestrian safety margin based on vehicle gap
drivers yield by reducing the speed and providing right-of-way to the accepted for different pedestrian and vehicle speeds are plotted as de-
pedestrian. The approaching vehicle-type also plays an essential role. picted in Fig. 7(a) and (b). Further, safety margin is quantified with
When a heavy vehicle approaches, the pedestrian feels safer than vehicle speed and waiting time for different values of vehicle gaps are
lighter vehicle, because of change in dynamic vehicular characteristics, presented in Fig. 7(a) and (b).
such as speed of vehicle. Age, platoon size and driver behaviour has As shown in Fig. 7(a), the slope of line plots, depicting variation of
relatively minimal effect on safety margin. Although, type of vehicle is SM with vehicle gap for different values of pedestrian speed, is rela-
usually an important factor for accepting the gaps (Yannis et al., 2010). tively flat and equally spaced, indicating that pedestrians maintains
In this study, it was also observed that the pedestrian’s safety against based on their speed, while crossing the street. The sensitivity of SM for
vehicle speed has a positive effect and it is because of the increase in larger vehicle gap size is observed to be increasing with increase in
pedestrian speed. In addition to this, waiting time and rolling gap be- pedestrian speed. Fig. 7(b) depict that slope of line plots indicating
haviour variables are found to have lesser effect on SM, but SM values variation of SM with vehicular gap for different values of vehicle speeds
are generally less than 2 s for range of these observations. Hence, these is steeper. It is observed that the sensitivity of the SM against vehicle
variables are also considered in the model, although they are found to gap size increases as the approaching vehicle speed increases.
be significant at 90% confidence level. Instinctively, it is known that The slope of line plots indicating variation of SM with vehicle
these variables entirely depend on pedestrian behaviour. In this case, speeds for different values of vehicular gaps is steeper as shown in
rolling behavior (stage-wise crossing across the lanes) occurs mainly, Fig. 8(a). The sensitivity of SM against vehicle speed increases as the
when traffic volume is higher. In addition, pedestrians in India (except vehicle gap increases and, on same lines, the sensitivity of SM (safety
female pedestrians) do not wait much patiently, till finding a suitable margin) against pedestrian speed increases as the vehicle gape size
safe gap for crossing road. Hence, these variables are cautiously in- increases. From Fig. 8(b), slope of line plots revealing variation of SM
cluded in the model. with waiting time for different values of vehicular gaps are flat and
sloping downwards. The sensitivity of SM against waiting time is ob-
6. Sensitivity analysis served to be decreasing with decrease in vehicle gap size. Although
slope of line is flat, it is extremely dangerous, as more waiting time
Sensitivity analysis is a technique used to determine how different always likely to result in smaller SM values. Moreover, range of SM

9
C. Avinash et al. Safety Science xxx (xxxx) xxx–xxx

vgap-4s vgap-8s
3 vgap-3.0s vgap-7.0s
2.8 2.1 vgap-10s vgap-6s
vgap-9.0s vgap-5.0s
2.6 2

Safety Margin(s)
Safety Margin(s)
2.4 1.9
2.2 1.8
2 1.7
1.8
1.6
1.6
1.4 1.5
1.2 1.4
1 1.3
0.00 20.00 40.00 60.00 80.00 100.00 0 20 40 60 80
Vehicle speed (kmph) Waiting time (s)

Fig. 8. (a and b) Sensitivity of SM based on vehicle speed/waiting time- Different vehicle gap.

values obtained with variation in waiting time also gives very small decreases. The developed PSM model would be useful to identify the
values (less than 2 s), which is notably unsafe. Hence, it is important to most critical situations (when the SM values are extremely low) during
consider this variable in developed model, although it is found to be the interaction between the pedestrian and vehicle on mid-block road
significant at 90% confidence level. crossings. Further, it can be advantageous to evaluate the existing pe-
destrian facilities at unprotected midblock location to improve the ex-
isting facilities, by suggesting the counter measures for the pedestrian
7. Conclusion
safety.
This study analyzed PSM at unprotected mid-block location under
The present study mainly highlighted on the behavioural char-
mixed-traffic condition. This study has some limitations; the age of
acteristics of individual pedestrian as observed from the video graphic
pedestrians was taken by visual appearance. But the individual pedes-
survey, that was carried out in the Western and the Northern parts of
trian age may improve the present model. The speed of the vehicle was
India. In developing countries like India, the road traffic environment is
considering within the length of video coverage section which is lim-
less adaptive for pedestrians’ needs and the pedestrians themselves are
ited, due to this the behavior of vehicular drivers are not predicted.
also non-compliant with traffic rules. For this purpose, four sections in
There is necessity to assess the driver behaviour with unprotected pe-
three different cities with varying geographical conditions are selected
destrian road crossing. The findings of the current study are limited to
to investigate pedestrian safety at mid-block crossing in an urban area.
only four-lane two-way (divided) carriageway.
In the present study, multiple- regression technique is used to analyse
pedestrian safety, while crossing the street. It was found that the pe-
Acknowledgement
destrians' minimum PSM values, while crossing the road depend on
parameters such as pedestrian speed, vehicle gap, vehicle speed, age,
The work described in this paper is supported by CSIR-CRRI Supra
platoon size, pedestrian rolling behaviour, vehicle type and driver
Institutional Network Project for Development of Indian Highway
yielding behaviour. The result of elasticity analysis shows that vehi-
Capacity Manual funded by Planning Commission, Government of India
cular gap size, vehicle speed, pedestrian speed and platoon size has the
under 12th five-year plan. Authors express deep sense of gratitude for
maximum effect on PSM. Further, it is also found that the pedestrian
the financial support.
rolling behaviour, waiting time, vehicle type and driver behaviour has
also reasonably significant effect. But, these variables are important due
Appendix A. Supplementary material
to unique behaviour by pedestrians especially under Indian conditions.
It was also found that there is no significant effect of gender on PSM,
Supplementary data to this article can be found online at https://
whereas age has a significant effect. Female pedestrians maintain
doi.org/10.1016/j.ssci.2018.12.009.
higher SM value than male for all approaching vehicle types, while
crossing the road. In general, pedestrian platoon size had a positive
References
effect on PSM, which is consistent with the results reported by Leden
(2002). Elderly and child category pedestrians are likely to face higher
Brewer, M.A., Fitzpatrick, K., Whitacre, J.A., Lord, D., 2006. Exploration of pedestrian
risks in comparison to young pedestrians. This is also confirmed based gap-acceptance behaviour at selected locations. Transp. Res. Rec. 1982 132–140.
on the SM values determined in the present study. Cavallo, V., Lobjois, R., Dommes, A., Vienne, F., 2009. Elderly pedestrians’ visual timing
Pedestrian platoon size gives a positive effect on SM. Sensitivity of strategies in a simulated street-crossing situation. In: Fifth international driving
symposium on human factors in driver assessment, training and vehicle design,
pedestrian SM was also analyzed on the basis of accepted gaps for Montana, USA.
different values of pedestrian speed, vehicle speed, and SM against Chu, X., Baltes, M.R., 2001. Pedestrian mid-block crossing difficulty. Technical Report.
waiting time and vehicle speeds for different values of accepted gaps. It pp. 1–79.
Guangxin, L., Keping, L., 2009. Pedestrian’s psychology on gap selecting when crossing
is observed that pedestrians are more likely to cross the road by
street. In: Proc., Int. Conf. on Transportation Engineering, ICTE, pp. 2138–2144.
maintaining minimum marginal vehicular gap for light vehicles (2 W Hamed, M.M., 2001. Analysis of pedestrian’s behaviour at pedestrian crossings. Saf. Sci.
and 3 W) in comparison to heavy vehicles (truck/bus). It is very well 38 (1), 63–82.
Himanen, V., Kulmala, R., 1988. An Application of logit models in analysing the beha-
brought out from the present study that the pedestrians are more cau-
viour of pedestrians and car drivers on pedestrian crossings. Accid. Anal. Prev. 20 (3),
tious, when the approaching vehicle is high-speed vehicle (car or 2 W), 187–197.
because of the aggressive behavior of driver. SM value increases with Holland, C., Hill, R., 2007. The effect of age, gender and driver status on pedestrians’
increase in the vehicle speed and hence, pedestrians tend to increase the intentions to cross the road in risky situations. Accid. Anal. Prev. 39, 224–237.
Holland, C., Hill, R., 2010. Gender differences in factors predicting unsafe crossing de-
crossing speed, indicating higher safety level. The magnitude of SM cisions in adult pedestrians across the lifespan: a simulation study. Accid. Anal. Prev.
against pedestrian speed increases, as the vehicle gap-size increases. 42, 1097–1106.
Similarly, the magnitude of SM increases against vehicle speed, as the Ishaque, M.M., Noland, R.B., 2008. Behavioural issues in pedestrian speed choice and
street crossing behaviour: a review. Transp. Rev.: A Trans. Tran Discipl. 28 (1),
vehicle gap increases. On the other hand, more dangerously, the mag- 61–85.
nitude of the SM against waiting time decreases, as the vehicle gap-size

10
C. Avinash et al. Safety Science xxx (xxxx) xxx–xxx

Kadali, B.R., Vedagiri, P., 2012. Pedestrians’ gap acceptance behaviour at uncontrolled Oxley, J., Fildes, B., Ihsen, E., Charlton, J., Day, R., 2005. Crossing roads safely: an ex-
mid-block location. In: Proc., Int. Conf. on 13th TRANSED, pp. 1–10. perimental study of age differences in gap selection by pedestrians. Accid. Anal. Prev.
Kadali, B.R., Vedagiri, P., 2013. Vedagiri ‘Pedestrians’ Safety Margin (PSM) for 37 (5), 962–971.
Unprotected Road Crossing’13th WCTR, July 15–18, Rio de Janeiro, Brazil. Sun, D., Ukkusuri, S.K., Benekohal, R.F., Waller, S.T., 2003. Modeling of Motorist
Kadali, B.R., Vedagiri, P., 2016. Proactive pedestrian safety evaluation at unprotected Pedestrian Interaction at Uncontrolled Mid-block Crosswalks. Transportation
mid-block crosswalk locations under mixed traffic conditions. Saf. Sci. 89, 94–105. Research Record, TRB Annual Meeting CD-ROM, Washington, DC.
Kenneth, D.L., Ronald, K.K., 1982. Advances in Business and Management Forecasting. Sun, D., Ukkauuri, S.V.S.K., Benekohal, R.F., Waller, S.T., 2005. Modeling of pedestrian
Emerald books, UK. gap acceptance for improving safety at uncontrolled mid-block crosswalks. Adv.
Kumar, P., Parida, M., 2011. Vulnerable road users in multi modal transport system for Transp. Stud.
Delhi. Journeys LTA 38–47. Sun, R., Zhuang, X., Changxu, Wu, Zho, G., Zang, K., 2015. The estimation of vehicle
Lassarre, S., Papadimitriou, E., Yannis, G., Golias, J., 2007. Measuring accident risk ex- speed and stopping distance by pedestrians crossing streets in a naturalistic traffic
posure for pedestrians in different micro-environments. Accid. Anal. Prev. 39, environment. Transp. Res. Part F.
1226–1238. Tung, Y., Liu, Y., Ou, Y., 2008. The pedestrian road-crossing behaviours between older
Leden, 2002. Pedestrian risk decrease with pedestrian flow. A case study based on data and younger age groups. Proceedings of the 9th Asia Pacific Industrial Engineering
from signalized intersection in Hamilton, Ontario. Accid. Anal. Perv. 34 (4), 457–464. and Management Systems Conference.
Lefler, D.E., Gabler, H.C., 2004. The fatality and injury risk of light truck impacts with Yannis G., Papadimitriour E., Akis Theofilatosl A., 2010. Pedestrian gap acceptance for
pedestrians in the United States. Accid. Anal. Prev. 36, 295–304. mid-block street crossing, In: 12th WCTR, Lisbon, Portugal, 1–11.
Lobjois, R., Cavallo, V., 2007. Age-related differences in street-crossing decisions: the Yannis, G., Papadimitriour, E., Akis Theofilatosl, A., 2013. Pedestrian gap acceptance for
effects of vehicle speed and time constraints on gap selection in an estim task. Accid. mid-block street crossing. Transp. Plan. Technol. 450–462.
Anal. Prev. 39, 934–943. Zhuang, X., Wu, C., 2012. The safety margin and perceived safety of pedestrians at un-
Mohan, Dinesh, Tiwari, Geetam, Bhalla, Kavi, 2015. Road Safety in India: Status Report. marked roadway. Transp. Res. Part F 15, 119–131.
Indian Institute of Technology, Delhi.

11