Accident Analysis and Prevention 71 (2014) 1521

Contents lists available at ScienceDirect

Accident Analysis and Prevention

journal homepage: www.elsevier.com/locate/aap

Effects of excessive speeding and falling asleep while driving on crash

injury severity in Ethiopia: A generalized ordered logit model analysis
Teferi Abegaz a, , Yemane Berhane b , Alemayehu Worku c ,
Abebe Assrat d , Abebayehu Assefa e
a
School of Public and Environmental Health, College of Medicine and Health Sciences, Hawassa University, Hawassa, Ethiopia
b
Addis Continental Institute of Public Health, Addis Ababa, Ethiopia
c
School of Public Health, College of Health Sciences, Addis Ababa University, Addis Ababa, Ethiopia
d
Ministry of Transport, Addis Ababa, Ethiopia
e
World Health Organization (WHO) Ethiopia, Addis Ababa, Ethiopia

a r t i c l e i n f o a b s t r a c t

Article history: The severity of injury from vehicle crash is a result of a complex interaction of factors related to drivers
Received 8 December 2013 behavior, vehicle characteristics, road geometric and environmental conditions. Knowing to what extent
Received in revised form 3 May 2014 each factor contributes to the severity of an injury is very important. The objective of the study was to
Accepted 3 May 2014
assess factors that contribute to crash injury severity in Ethiopia. Data was collected from June 2012 to
Available online 24 May 2014
July 2013 on one of the main and busiest highway of Ethiopia, which extends from the capital Addis
Ababa to Hawassa. During the study period a total of 819 road crashes was recorded and investigated by
Keywords:
trained crash detectors. A generalized ordered logit/partial proportional odds model was used to examine
Injury severity
Generalized ordered logit model
factors that might inuence the severity of crash injury. Model estimation result suggested that, alcohol
Signicant factors use (Coef. = 0.5565; p-value = 0.017), falling asleep while driving (Coef. = 1.3102; p-value = 0.000), driving
Highway safety at night time in the absence of street light (Coef. = 0.3920; p-value = 0.033), rainfall (Coef. = 0.9164; p-
value = 0.000) and being a minibus or vans (Coef. = 0.5065; p-value = 0.013) were found to be increased
crash injury severity. On the other hand, speeding was identied to have varying coefcients for different
injury levels, its highest effects on sever and fatal crashes. In this study risky driving behaviors (speeding,
alcohol use and sleep/fatigue) were a powerful predictor of crash injury severity. Therefore, better driver
licensing and road safety awareness campaign complimented with strict police enforcement can play a
pivotal role to improve road safety. Further effort needed as well to monitor speed control strategies like;
using the radar control and physical speed restraint measures (i.e., rumble strips).
2014 Published by Elsevier Ltd.

1. Introduction
Severity of an injury is the result of a complex interaction
between several factors representing driver and other road users,
road characteristics, vehicle defect and design, and environmen-
tal characteristics including weather condition and light (Mondal
et al., 2011; Thompson et al., 2013). Understanding to what extent
each factor contributes to the severity of crash related injury is one
of the most effective means to improve highway safety.
The human factor appears in the literature as being the most
common determinants of road safety. Demographic characteristics:
male, younger and older drivers are more likely to be involved in
Corresponding author. Tel.: +251 911361607.
E-mail addresses: teferiabegaz@gmail.com, tefabeg@yahoo.com (T. Abegaz),
yemaneberhane@gmail.com (Y. Berhane), alemayehuwy@yahoo.com (A. Worku),
abebeasrt@yahoo.com (A. Assrat), abebayehua@et.afro.who.int (A. Assefa).
death and severe injury crashes (Dissanayake and Lu, 2002; Helai
et al., 2008; Shope and Bingham, 2008; Haleem and Gan, 2013).
As indicated by many researchers the strongest association of
crash injury severity are related to behavioral factors, including:
alcohol/drug use (Kuruc et al., 2009; Tsui et al., 2010; Ponce
et al., 2011; Hels et al., 2013); speeding (Dissanayake and Lu,
2002; Afukaar, 2003); failure to wear seat belts (Munk et al.,
2008; Siskinda et al., 2011; Kashani et al., 2012); using mobile
phone while driving (Violanti, 1998) and sleep/fatigue (Radun and
Summala, 2004). Speeding is a critical safety concern, especially for
developing countries, where fatalities are more common among
pedestrians and users of two- and three-wheelers (Mohan, 2002).
Fatigue crashes are usually severe, as the driver makes no attempt
to limit the consequence (Radun and Summala, 2004). Such types
of crashes are linked with the nature of the road alignment; good
condition of the road network make the driver task easy and

http://dx.doi.org/10.1016/j.aap.2014.05.003
0001-4575/ 2014 Published by Elsevier Ltd.
16 T. Abegaz et al. / Accident Analysis and Prevention 71 (2014) 1521
monotonous, which demanded little effort as a result sleep/fatigue
will be induced (Summala, 1996; Rossi et al., 2011).
Studies have also suggest that road alignment (Hosseinpour
et al., 2014), street light condition (Wanvik, 2009; Mogaka et al.,
2011) and adverse weather conditions (Mondal et al., 2011) are
important predictors that affect the severity of vehicle crashes.
As stated by various researchers, rainy weather signicantly affect
crash injury severity (Mogaka et al., 2011; Mondal et al., 2011), wet
road reduce the tire traction efciency, leading to poor braking per-
formance (Cho et al., 2006). In addition, visibility obstruction due
to fog resulted a higher severity crashes (Abdel-Aty et al., 2011).
Vehicle type and trafc mix as well found to be important factors
and signicantly associated with injury severity (Chang and Wang,
2006). Commercial and public transportation operated in develop-
ing countries are inuenced by the owners to work excessively long
hours when exhausted (Mock et al., 1999). On the other hand, in
multi-vehicle crashes, heavy truck have better resistance to crash
impact and protect its occupants; however, results in more serious
injuries to the other vehicle(s) involved in the crash (Kockelman
and Kweon, 2002). Empirical evidence also showed, motorcycles
are a more dangerous mode of transportation than automobiles
due to the absence of structure that protect the riders during a
crash (Schlundt et al., 2004).
Most of these facts on crash related injury severity and asso-
ciated factors are obtained from developed countries. Evidence
is scarce from developing countries, especially in Africa (Lagarde,
2007). Thus, this study was conducted to explore the potential fac-
tors related to crash injury severity in Ethiopia.
2. Methods
2.1. Study setting
This study was conducted from June 2012 to July 2013, on one of
the main and busiest roads of Ethiopia, which extends South from
the capital Addis Ababa to Hawassa (Hawassa is the capital city of
Southern Nations Nationality and Peoples Regional State, located at
275 km South of Addis Ababa). This two-way and two-lane road has
an average width of 8 m and covers a total distance of 264 km. It is
part of the main route of the countrys import and export corridor
from the port of Djibouti and part of the Trans-African Highway
(an international road that stretched from Cairo to Cape Town).
Moreover, the road has a signicant economic importance since
many of the cash crops, oriculture farms, recreational areas and
tourist centers are located across the stretches of the road. Due
to these reasons, the road is considered as the most vital route in
terms of economy and trafc volume. The trafc mix on this road
is very diverse, shared by high speed vehicles, heavy trucks, animal
drowns carts, two-and three-wheelers and pedestrians. According
to the Ethiopian Road Authority report more than 20,000 vehicles
used the road daily (38.8% trucks, 24.3% cars, 21.4% buses and 15.5%
truck trailers) (ERA, 2012). Compared to other road networks in
the country, the Addis Ababa-Hawassa road has good features with
regard to its alignment (i.e., straight and less curved). Map of Addis
Ababa-Hawassa highway is illustrated in Fig. 1.
2.2. Statistical analysis
Crash injury severity often categorized universally as discrete
ordered categories such as: fatal injury, incapacitating injury, non-
incapacitating injury, possible injury, and property damage only.
However, this classication may slightly vary from place to place.
For example, in Ethiopia crash injury severity is classied into four
categories: fatal injury (death at the scene or up to one month fol-
lowing an incident); serious injury (victim hospitalized at least for
24 h); minor injury (victim treated at an outpatient service or hospi-
talized for less than 24 h), and property damage only (crash without
any human injury). Severity of the crash was determined according
to the level of injury to the worst injured road users. Accordingly,
we coded as: 1 = no injury, 2 = minor injury, 3 = serious injury and
4 = death. The contribution of driver characteristics (i.e., age and
vehicle ownership) and driver behavior (i.e., speeding, alcohol use,
phoning and sleep), collision type, infrastructure characteristics
(i.e., road alignment, posted trafc sign, speed limit, and section
type), vehicle type and environmental conditions (i.e., lighting and
weather) were assessed by the crash injury severity model.
Previously, numerous researchers implemented ordered dis-
crete outcome models; ordered logit and probit in injury severity
research (Odonnell and Connor, 1996; Abdel-Aty, 2003; Gray et al.,
2008; Pai and Saleh, 2008). However, these two models assume the
parameter estimates are constant across the severity levels, which
is not always fullled. This condition is said to be the proportional
odds or parallel-regression assumption. Violation of this assump-
tion leads to the formulation of an incorrect or mis-specied model.
It can be checked by using Brant test (Brant, 1990), a signicant
test statistic provides evidence that the parallel regression assump-
tion has been violated. Generalized ordered logit model relax the
proportional odds or parallel-regression assumption for some or
all variables. Eluru (2013) recommended generalized ordered logit
model as a best alternative of multinomial logit model for ordinal
response data. In recent years, this model adopted in the eld of
road safety (Wang and Abdel-Aty, 2008; Kaplan and Prato, 2012;
Mooradian et al., 2013). Generalized ordered logit model has the
form described below
 
  exp j Xi j
p (yi > j) = g Xi j =   , j = 1, 2, 3 (1)
1 + exp j Xi j
where Xi is a p 1 vector containing the values of crash i on the
full set of p explanatory variables, j is a p 1vector of regression
coefcients, j represents a cutoff point for the jth cumulative logit.
This model relaxes the parallel-line assumption for all of the covari-
ates which is not always true, and hence, increases the number of
parameters unnecessarily.
By assuming only few covariates violate the parallel-line
assumption, the unconstrained partial-proportional odds model
developed by (Peterson and Harrell, 1990) was applied. A
gamma parameterization of unconstrained partial-proportional
odds model with logit function can be specied as
  
  exp j Xi j + Ti j
P (Yi > j) = g Xi j =    (2)
1 exp j Xi j + Ti j 
where T is a q 1 vector, q p containing the values of crash i on that
subset of the p predictor variables for which the proportional odds
assumption not fullled.  j is a q 1 vector of regression coefcient
associated with q covariate in T , so that, T  j is the increment associ-
ated with the jth cumulative logit. Covered with a non-proportional
odds have one, k-2 and k-1 coefcients, where k is the number
of alternatives (in our case, k = 4). The overall contribution of these
variables on different injury severity category can be computed by
adding the gamma coefcients of the respected equation and the
beta coefcients.
Generalized ordered logit model can be tted by user-written
program geologit2 in Stata (Williams, 2006). Interpreting the
coefcients of intermediate categories for those unconstrained
variables requires special care, since the direction of the effect not
always determined by the sign of the estimate. Usually marginal
values are considered for interpretation. Model tness can be
assessed by using Pseudo R2 and Akaikes information criteria (AIC),
best-tted models has a smaller value of AIC (Stata, 2005).
T. Abegaz et al. / Accident Analysis and Prevention 71 (2014) 1521 17

Fig. 1. Map of Addis Ababa-Hawassa Road.

18 T. Abegaz et al. / Accident Analysis and Prevention 71 (2014) 1521
2.3. Sample size
For this study, the sample size was determined based on the
recommendation of Sudman (1976). He suggests that, 100 sub-
jects are needed for each major group or subgroup. Accordingly,
the sample size for each severity level was estimated by consider-
ing the proportion of the severity category. At least 100 samples
needed for the smaller proportion of the severity category. Based
on the Ethiopian Road Transport Authority, National Road Safety
Coordination Ofce report (NRSCO, 2006), the proportions of crash
injuries of varying severities in the studied road constituted 33.3%
property damage only, 12.2% minor injury, 26.3% serious injury and
28.2% fatal injuries. Thus, the smaller proportion was minor injury
crashes, in order to get 100 subjects for this severity category, we
consider a total of 820 crashes (i.e., 100/0.122 = 820).
2.4. Data collection and preparation
Relevant data on crash injury severity, driver related informa-
tion, road characteristics, vehicle type, weather and light conditions
were gathered by trained professional road crash investigators.
We used different data collection strategies; rst detailed incident
investigation carried out immediately at the scene of the crash, then
extra information collected by asking eyewitnesses and review-
ing medico-legal reports. Field supervision was done by two senior
trafc ofcers and the principal investigator. Consistency and com-
pleteness were checked in a daily base. The data were double
entered by two different data clerks using EpiData 3.1 statistical
software and cleaned accordingly.
3. Results
A total of 819 crashes were recorded during the study period. Of
which 424 (51.8%) were property damage only; 113 (13.8%) minor
injury crashes; 106 (12.9%) serious injury crashes and 176 (21.5%)
fatal crashes. Almost all drivers involved in the crashes were male
814 (99.4%). The distribution of severity of injury by background
characteristics are illustrated in Table 1.
The parameter estimate showed that, a variable violate the pro-
portional odds assumption had one beta coefcient, two gamma
coefcients and three alpha coefcients reecting the cutoff points.
Speeding (p-value = 0.00428) and a vehicle being a heavy truck (p-
value = 0.00065) were found to be violating the proportional odds
assumption. The gamma 2 and the beta coefcients for speeding
were (0.4205) and (0.7450), respectively. When we added these
two values we obtained the coefcient of speeding in the sec-
ond equation, which was (1.17). Likewise, when we added the
gamma 3 coefcient 0.5543 with the beta coefcient 0.7450 we
obtain the effect of speeding in the third equation, which was
1.2993. Similarly, the effect of heavy truck at various severity
categories is different. The parameter estimates in the second
and third equation were computed in a similar manner as stated
above, 0.3555 + (0.8157) = 0.4602 for the second equation and
0.6950 + (0.8157) = 0.1207 for the third equation. Parameter
estimates obtained from the generalized ordered logit model is
presented in Table 2.
3.1. Drivers risky behavior
Drivers behavior like excessive speeding, alcohol use and
sleep/fatigue were identied as signicant factors for crash injury
severity. Of these factors, sleep/fatigue is one of the potential
determinants; the result showed that, those drivers who slept
while driving ended up in a more serious injury category com-
paring with those alert drivers (Coef. = 1.3102; p-value = 0.000).
Similarly, driving after drinking alcohol was a signicant cause;
drunk drivers were more likely to sustain severe injuries comparing
with non drunk drivers (Coef. = 0.5565; p-value = 0.017). The effect
of speeding across the severity categories showed different magni-
tude, the rst panel of coefcients (i.e., property damage Vs minor
injury + serious injury + death) (Coef. = 0.7450; p-value = 0.000), the
second panel of coefcients (i.e., property damage + minor injury vs
serious injury + death) (Coef. = 1.17; p-value = 0.000) and the third
panel of coefcients (i.e., property damage + minor injury + serious
injury vs death) (Coef. = 1.2993; p-value = 0.000). The increasing
trends indicate that, the strongest effect being found in the highest
injury severity level (increasing the probability of serious and fatal
injuries). However, the result demonstrated that, those who drive
on a congested road network (higher density of road that causes
less spacing) were sustaining a less severe injury category (Coef.
1.2034 = ; p-value = 0.000) compared with free ow trafc.
3.2. Environmental conditions
Among those environmental related conditions, rainfall and
driving at night time in the absence of street light showed a sig-
nicant contribution to crash injury severity. Driving in a rainy
situation was identied to be signicant for crash injury sever-
ity comparing with driving on a clear or other weather conditions
(Coef. = 0.9164; p-value = 0.000). Similarly, night time driving in the
absence of street light was resulted a more serious injury outcome
(Coef. = 0.3920; p-value = 0.033) than day time driving and driving
at night time in the presence of street light.
3.3. Vehicle type
Of the vehicle types minibus or vans (Coef. = 0.5065; p-
value = 0.013) and two- or three-wheeled vehicles (bicycle,
motorcycle and Bajaj taxi) (Coef. = 0.9143; p-value = 0.004) were
increasing the probability of higher crash severity. On the other
hand, automobiles or station wagons or pickups were associated
with less severe crashes (Coef. = 0.4431; p-value = 0.037). Heavy
trucks traveled on a congested road network also showed protec-
tive effects on the severity, they were involved in less severe injury
crashes (Coef. = 0.8157; p-value = 0.000).
4. Discussion
This study assesses the potential predictors associated with
driver behavior, collision type, infrastructure characteristics, vehi-
cle type and environmental conditions using partial proportional
odds model. Of these factors, excessive speeding, alcohol use,
sleep/fatigue, driving at night time in the absence of street light,
rainfall, two-or three-wheelers and minibus or vans were found
to be increased crash injury severity. On the other hand, driving
in a congested road network and the vehicles being automo-
biles/station wagons/pickups were signicantly reducing crash
injury severity level.
One of the potential limitations of this study was lack of
adequate power to perform crash-specic statistical analysis
(single-versus multi-vehicle crash scenario), and hence we couldnt
addresses same safety issues, which repeatedly assessed as an
important factor contributing for crash injury severity. We couldnt
also assess the effect of seat belt use, since the seat belt law in
Ethiopia is mandatory for drivers only. Moreover, we incorporated
only those vehicles responsible for the crash in the statistical anal-
ysis, due to lack of information on the other vehicles, which were
not responsible for the causes of the crashes.
Our nding demonstrated that alcohol use increases the risk
of injury severity, consistent with previous studies (Kuruc et al.,
2009; Tsui et al., 2010; Ponce et al., 2011). The higher injury sever-
ity observed among sleepy/fatigue drivers are in accordance with
 T. Abegaz et al. / Accident Analysis and Prevention 71 (2014) 1521 19

Table 1
Distribution of injury severity by key explanatory variables on Addis Ababa-Hawassa highway from 2012 to 2013.

Variable Categories Injury severity frequencies

Property damage Minor 113 Serious 106 Death 176 Total 819
424 (51.8%) n (%) (13.8%) n (%) (12.9%) n (%) (21.5%) n (%) (100%) n (%)

Drivers age <25 years 58 (51.8%) 15 (13.2%) 16 (13.6%) 24 (21.4%) 113 (13.8%)
2534 years 214 (51%) 61 (14.5%) 54 (12.9%) 90 (21.5%) 419 (51.2%)
3544 years 95 (52.8%) 28 (15.6%) 19 (10.5%) 38 (21.1%) 180 (22%)
4554 years 38 (50%) 8 (10.5%) 11 (14.5%) 19 (25%) 76 (9.3%)
5564 years 13 (54.2%) 2 (8.3%) 5 (20.8%) 4 (16.7%) 24 (2.9%)
>64 years 5 (71.4%) 1 (14.3%) 1 (14.3%) 7 (0.8%
Driving behavior Alcohol use 28 (35%) 8 (10%) 14 (17.5%) 30 (37.5%) 80 (9.8%)
and condition Speeding 170 (39.5%) 51 (11.8%) 70 (16.2%) 140 (32.5%) 431 (52.6%)
Sleep/fatigue 20 (27.4%) 5 (6.8%) 13 (17.8) 35 (48%) 73 (8.9%)
Phoning 21 (39.6%) 8 (15.1%) 9 (17%) 15 (28.3%) 53 (6.5%)
Congested road 116 (78.9%) 14 (9.5%) 7 (4.8%) 10 (6.8%) 147 (17.9%)
Improper lane turning 54 (58.7%) 18 (19.5%) 11 (12%) 9 (9.8%) 92 (11.2%)
Safety belt Used 275 (57.7%) 70 (14.7%) 55 (11.5%) 77 (16.1%) 477 (58.2%)
Not used 75 (44.4%) 22 (13%) 24 (14.2%) 48 (28.4%) 169 (20.6%)
Unknown 74 (42.8%) 21 (12.1%) 27 (15.6%) 51 (29.5%) 173 (21.1%)
Driving experience 5 years 231 (50.5%) 68 (14.9%) 59 (12.9%) 99 (21.7%) 457 (55.8%)
610 years 148 (56.5%) 30 (11.5%) 32 (12.2%) 52 (19.8%) 262 (32%)
>10 years 45 (45%) 15 (15%) 15 (15%) 25 (25%) 100 (12.2%)
Number of vehicles One 123 (29.6%) 78 (18.8%) 76 (18.3%) 139 (33.4%) 416 (50.8%)
involved Two 273 (74.6%) 29 (7.9%) 29 (7.9%) 35 (9.6%) 366 (44.7%)
Three or more 28 (75.7%) 6 (16.2%) 1 (2.7%) 2 (5.4%) 37 (4.5%)
Vehicle type 2 and 3 wheelers 13 (29.5%) 5 (11.4%) 10 (22.7%) 16 (36.4%) 44 (5.4%)
Car, station wagon and pickup 85 (59%) 23 (16%) 16 (11.1%) 20 (13.9%) 144 (17.6%)
Minibus and van 41 (30.8%) 28 (21.1%) 27 (20.3%) 37 (27.8%) 133 (16.2%)
Medium and large bus 50 (54%) 14 (15.2%) 7 (7.6%) 21 (22.8%) 92 (11.2%)
Medium truck 72 (46.2%) 19 (12.2%) 25 (16%) 40 (25.6%) 156 (19%)
Heavy trucks 161 (67.1%) 20(8.3%) 19 (7.9%) 40 (16.7%) 240 (29.3%)
Others 7 (70%) 2(20%) - 1 (10%) 10 (1.2%)
Manner of collision Head-on 45 (47.9%) 11 (11.7%) 13 (13.8%) 25 (26.6%) 94 (11.5%)
Front-rear 170 (82.5%) 15 (7.3%) 11 (5.3%) 10(4.9%) 206 (25.2%)
Front-side 66 (68%) 12 (12.4%) 7 (7.2%) 12 (12.4%) 97 (11.8%)
Sideswipe 32 (76.2%) 3 (7.1%) 3 (7.1%) 4 (9.5%) 42 (5.1%)
Rollover 92 (58.6%) 16 (10.2%) 23 (14.6%) 26 (16.6%) 157 (19.2%)
Pedestrian collision 64 (29.1%) 52 (23.6%) 104 (47.3%) 220 (26.9%)
Animal collision 19 (65.5%) 4 (13.8%) 4 (13.8%) 2 (6.9%) 29 (3.5%)
Fixed object collision 10 (34.5%) 3 (10.3%) 8 (27.6%) 8 (27.6%) 29 (3.5%)
Vehicle maneuver Going street ahead 339 (48.9%) 99 (14.3%) 94 (13.6%) 161 (23.2%) 693 (84.6%)
Crossing the road 14 (48.3%) 3 (10.3%) 4 (13.8%) 8 (27.6%) 29 (3.5%)
Others 65(67%) 9 (9.3%) 10 (10.3%) 13 (13.4%) 97 (11.8%)
Technical status of Non-defective 399 (51.6%) 105 (13.6%) 101 (13%) 169 (21.8%) 774 (94.5%)
vehicle Defective 25 (55.6%) 8 (17.8%) 5 (11%) 7(15.6%) 45 (5.5%)
Road type One-way 67 (60.9%) 14(12.7%) 7 (6.4%) 22 (20%) 110 (13.4%)
Undivided two-way 140 (48%) 47 (16.2%) 41 (14.1%) 63 (21.6%) 291 (35.5%)
Divided two-way 210 (50.2%) 52 (12.4%) 59 (14.1%) 97 (23.2%) 418 (51%)
Road alignment Straight 387 (52%) 105 (14.1%) 97 (13%) 155 (20.8%) 744 (90.8%)
Curved 37 (49.3%) 8 (10.7%) 9 (12%) 21 (28%) 75 (9.2%)
Weather condition Clear 379 (57%) 95 (14.3%) 75 (11.3%) 116 (17.4%) 665 (81.2%)
Rainy 29 (25.9%) 14 (12.5%) 22 (19.6%) 47 (42%) 112 (13.7%)
Foggy 16 (38.1%) 4 (9.5%) 9 (21.4%) 13 (30.9%) 42 (5.1%)
Light condition Daylight 332 (57%) 72 (12.4%) 74 (12.7%) 104 (17.9%) 582 (71.1%)
Night with street light 34 (43.6%) 16 (20.5%) 10 (12.8%) 18(23%) 78 (9.5%)
Night without street light 57 (35.8%) 21 (13.2%) 26(16.4%) 55 (34.6%) 159 (19.4%)
Day of the week Weekday 316 (52.9%) 70 (11.7%) 76 (12.7%) 135 (22.6%) 597 (72.9%)
Weekend 108 (48.6%) 43 (19.4%) 30 (13.5%) 41 (18.5%) 222 (27.1%)

previous reports (Radun and Summala, 2004). The possible expla-
nation might be Addis Ababa-Hawassa road has good features with
regard to its alignment (i.e., straight and less curved); such types
of road network make the driver task simple and boring as a result
sleep/fatigue could be induced (Summala, 1996; Rossi et al., 2011).
As expected, high speed was another major factor associated with
increased injury severity in this study. In accordance with previ-
ous work (Afukaar, 2003), speeding is highly associated with fatal
injury due to high impact force that limit the human tolerance.
Mobile phone use while driving was not signicant in this study;
this may be due to the difculty of assessing the time span between
the crash and the conversation takes place. In our study driving in a
congested road network signicantly decrease crash injury sever-
ity, consistent with previous study, trafc congestion was found
to decrease the effect of the injury severity compared to free-ow
trafc conditions (Duncan et al., 1998). The possible explanation for
this may be as the road network is a two-way and two-lane road
with high trafc volume and diverse mix; drivers not respected the
posted speed limit and they are forced to drive to the lower speed.
Low speed reduces the severity of a crash.
The result showed that rainfall signicantly increase injury
severity, which is supported by different researchers (Junga et al.,
2010; Mogaka et al., 2011). Empirical evidence shows wet road
reduce the tire traction efciency, leading to poor braking perfor-
mance (Cho et al., 2006). Driving at night time in the absence of
street lights is one of the signicant predictors in our study, which
is consistent with other similar ndings (Clarke et al., 2006; Gray
et al., 2008; Mogaka et al., 2011). This might be due to inappropriate
20 T. Abegaz et al. / Accident Analysis and Prevention 71 (2014) 1521

Table 2
Model summary for injury severity on Addis AbabaHawassa Highway from 2012 to 2013 using generalized ordered logit model.

Variables Coefcients Standard error p-Value

Beta Alcohol use 0.5565 0.2332 0.017

Speeding 0.7450 0.1620 0.000
Sleep/fatigue 1.3102 0.2610 0.000
Congested road network 1.2034 0.2324 0.000
Night without street light 0.3920 0.1835 0.033
Rainy weather 0.9164 0.2045 0.000
2 & 3 Wheelers 0.9143 0.3135 0.004
Car (automobile station wagon & pickup) 0.4431 0.2124 0.037
Heavy trucks 0.8157 0.1970 0.000
Minibus & van 0.5065 0.2043 0.013
Gamma 2 Speeding 0.4205 0.1351 0.002
Heavy trucks 0.3555 0.1222 0.004
Gamma 3 Speeding 0.5543 0.2040 0.007
Heavy trucks 0.6950 0.1872 0.000
Alpha Constant 1 0.4375 0.1734 0.012
Constant 2 1.5217 0.1917 0.000
Constant 3 2.5249 0.2328 0.000
Summary statistics Number of observations 819
Log likelihood at convergence 862.64
AIC 1759.29
Pseudo R2 (Mc Fadden pseudo R-square) 0.1289

use of light, higher level of tiredness and associated slower reaction
and reduced hazard awareness.
In this study, Minibus or vans were more likely to be involved
in severe injury; these vehicles are commonly used for public
transportation service in Ethiopia. The possible explanation for
this may be, commercial and public road transport in developing
countries have revealed that transport owners, in favor of their
prots, frequently force their drivers to drive at higher speeds, to
work excessively long hours and to work when exhausted (Mock
et al., 1999). On the contrary, automobile station wagon and pick-
up reduce crash injury severity; such kinds of vehicles are used for
private trip and usually driven by the owner themselves, so they are
not at risk of severe crashes. Similarly, the study demonstrated that
heavy trucks traveled on a congested road network were involved
in a lower severity crashes, such vehicles use the rst 64 km of the
road network which is a part of the main route of the countrys
import and export corridor from the port of Djibouti (Addis Ababa-
Mojo network). This road network is characterized by high trafc
congestion, so drivers slow down their speed, as indicated by many
researchers low speed limits the seriousness of the crash outcome.
Road safety can be achieved by educating drivers, public trans-
port owners and the general public. Stricter enforcement of the
existing road safety laws is very important, especial emphasis given
for; speeding, alcohol, sleep/fatigue and driving for excessively long
hours. Speed limits are not respected by many of the drivers in
Ethiopia and even majority of police ofcers are not considered
speeding as an offense. However, Rwanda has lower levels of road
trafc injuries than many of its neighbors because its government
has been more stringent in enforcing laws on speeding (Watkins
and Sridhar, 2009). Amendments of the existing road safety laws
and incorporating the legal alcohol limit and implementing breath
test should also give more emphasis.
5. Conclusions
higher severity crashes. We also found that the probability of seri-
ous injury and death exceptionally increase with increasing speed.
Sleepy/fatigue driver as well increase crash injury severity due
to lack of control and delayed reaction time. Road environment
related factors: rainy weather (Coef. = 0.9164; p-value = 0.000),
driving at night time in the absence of street light (Coef. = 0.3920;
p-value = 0.033), also found to be signicant predictors of crash
injury severity. Vehicle type like minibus and vans commonly used
for public transportation service in Ethiopia and two and three-
wheeled motorized and non-motorized vehicles, resulted higher
severity crashes (Coef. = 0.5065; p-value = 0.013).
In general, this study exerts an important effort to under-
stand the effects of various interdependent factors on crash injury
severity. However, the work was limited to show variation in the
interaction of factors across different scenario of collision due to
the small sample. Therefore, we recommended further exploration
on this aspect, especially on single-vehicle versus multi-vehicle
crashes.
Based on these ndings more efcient interventions can be
developed to alleviate crash injury severity in Ethiopia. Preven-
tion strategies applied to reduce injuries and fatalities from trafc
collision should focus on driver risky behavior: speeding, alcohol
use and driving at night time or when drivers are sleep deprived.
Therefore, implementing better driver licensing and road safety
awareness campaign on safe driving practices can play a pivotal role
in road safety improvement. In addition, strict police enforcement
also applied for those frequent offenders. Further effort needed
as well to monitor speed control strategies like; using the radar
control and physical speed restraint measures (i.e., rumble strips).
The legal alcohol limit should be incorporated in the existing road
safety laws and implementing random breath test. The driving
hours should be restricted for those public transportation vehicles
and owners of such vehicles should also share responsibility for the
current unnecessary loss of life.

In this study partial proportional odds model was applied to

assess the signicant factors affecting crash injury severity. Model
estimation results suggested that risky driving behavior was a pow-
erful predictor of crash injury severity. Alcohol use (Coef. = 0.5565;
p-value = 0.017), falling asleep while driving (Coef. = 1.3102; p-
value = 0.000) and speeding (Coef. = 1.2993; p-value = 0.000) were
among those potential predictors identied by the model. Drivers
who had consumed alcohol had a higher risk of involvement in
Acknowledgments
This study was nanced by Hawassa University and the WHO
Ethiopia country ofce (Grant number: AFETH1206929). Special
thanks to the Addis Continental Institute of Public Health for spon-
soring the study. We also thank management bodies and trafc
ofcers involved in the data collection process.
 T. Abegaz et al. / Accident Analysis and Prevention 71 (2014) 1521
References
Abdel-Aty, M., Ekram, A.-A., Huang, H., Choi, K., 2011. A study on crashes related to
visibility obstruction due to fog and smoke. Accid. Anal. Prev. 43, 17301737.
Abdel-Aty, M.A., 2003. Analysis of driver injury severity levels at multiple locations
using ordered probit models. J. Saf. Res. 34 (5), 597603.
Afukaar, F., 2003. Speed control in developing countries: issues, challenges and
opportunities in reducing road trafc injuries. Int. J. Contr. Saf. Promot. 10,
7781.
Brant, R., 1990. Assessing proportionality in the proportional odds models for ordinal
logistic regression. Biometrics 46 (4), 11711178.
Chang, L.-Y., Wang, H.-W., 2006. Analysis of trafc injury severity: an application of
non-parametric classication tree techniques. Accid. Anal. Prev. 38, 10191027.
Cho, J., Lee, H., Yoo, W., 2006. A wet-road braking distance estimate utilizing
the hydroplaning analysis of patterned tire. Int. J. Numer. Methods Eng. 69,
14231445.
Clarke, D.D., Ward, P., Bartle, C., Truman, W., 2006. Young driver accidents in the UK:
the inuence of age, experience, and time of day. Accid. Anal. Prev. 38, 871878.
Dissanayake, S., Lu, J.J., 2002. Factors inuential in making an injury severity differ-
ence to older drivers involved in xed object-passenger car crashes. Accid. Anal.
Prev. 34, 609618.
Duncan, C.S., Khattak, A.J., Council, F.M., 1998. Applying the ordered probit model
to injury severity in truck-passenger car rear-end collisions. Transp. Res. Rec.
1635, 6371.
Eluru, N., 2013. Evaluating alternate discrete choice frameworks for modeling ordi-
nal discrete variables. Accid. Anal. Prev. 55, 111.
ERA, 2012. Ethiopian Road Asset. Ethiopian Roads Authority (ERA), Addis Abab,
Ethiopia, <http://www.era.gov.et/RoadAsset/Trafc>.
Gray, R.C., Quddus, M.A., Evans, A., 2008. Injury severity analysis of accidents involv-
ing young male drivers in Great Britain. J. Saf. Res. 39, 483495.
Haleem, K., Gan, A., 2013. Effect of drivers age and side of impact on crash severity
along urban freeways: a mixed logit approach. J. Saf. Res. 46, 6776.
Helai, H., Chor, C.H., Haque, M., 2008. Severity of driver injury & vehicle damage
in trafc crashes at intersections: a Bayesian hierarchical analysis. Accid. Anal.
Prev. 40, 4554.
Hels, T., Lyckegaard, A., Simonsen, K.W., Steentoft, A., Bernhoft, I.M., 2013. Risk of
severe driver injury by driving with psychoactive substances. Accid. Anal. Prev.
59, 346356.
Hosseinpour, M., Yahaya, A.S., Sadullah, A.F., 2014. Exploring the effects of roadway
characteristics on the frequency and severity of head-on crashes: case studies
from Malaysian federal roads. Accid. Anal. Prev. 62, 209222.
Junga, S., Qinb, X., Noyce, D.A., 2010. Rainfall effect on single-vehicle crash severities
using polychotomous response models. Accid. Anal. Prev. 42, 213224.
Kaplan, S., Prato, C.G., 2012. Risk factors associated with bus accident severity in the
united states: a generalized ordered logit model. J. Saf. Res. 43, 171180.
Kashani, A.T., Shariat-Mohaymany, A., Ranjbari, A., 2012. Analysis of factors associ-
ated with trafc injury severity on rural roads in Iran. J. Inj. Violence Res. 4 (1),
3641.
Kockelman, K.M., Kweon, Y.-J., 2002. Driver injury severity: an application of ordered
probit models. Accid. Anal. Prev. 34, 313321.
Kuruc, R., Sidlo, J., Valent, D., Mlynar, J., Murarikova, B., 2009. The responsibility of
alcohol impaired road users in fatal road trafc accidents. Bratisl. Lek. Listy 110
(12), 802806.
Lagarde, E., 2007. Road trafc injury is an escalating burden in Africa and deserves
proportionate research efforts. PLoS Med. 4 (6), e170.
Mock, C., Amegashi, J., Darteh, K., 1999. Role of commercial drivers in motor vehicle
related injuries in Ghana. Inj. Prev. 5, 268271.
Mogaka, E.O., Nganga, Z., Oundo, J., Omolo, J., Luman, E., 2011. Factors associated
with severity of road trafc injuries, Thika, Kenya. Pan Afr. Med. J. 8 (20).
Mohan, D., 2002. Road safety in less motorized environments: future concerns. Int.
J. Epidemiol. 31, 527532.
21
Mondal, P., Sharma, N., Kumar, A., Vijay, P., Bhangale, U.D., Tyagi, D., 2011. Are road
accidents affected by rainfall? A case study from a large Indian metropolitan
city. Br. J. Appl. Sci. Technol. 1 (2), 1626.
Mooradian, J., Ivan, J.N., Ravishanker, N., Hu, S., 2013. Analysis of driver and passenger
crash injury severity using partial proportional odds models. Accid. Anal. Prev.
58, 5358.
Munk, M.-D., Carboneau, D.M., Hardan, M., Ali, F.M., 2008. Seatbelt use in Qatar in
association with severe injuries and death in the pre-hospital setting. Prehospi-
tal Disaster Med. 23 (6), 547552.
NRSCO, 2006. Overview of the Road Safety Activities in Ethiopia. Ethiopian Road
Transport Authority, National Road Safety Coordination Ofce.
Odonnell, C.J., Connor, D.H., 1996. Predicting the severity of motor vehicle acci-
dent injuries using models of ordered multiple choice. Accid. Anal. Prev. 28 (6),
739753.
Pai, C.-W., Saleh, W., 2008. Exploring motorcyclist injury severity in approach-turn
collisions at t-junctions: focusing on the effects of drivers failure to yield and
junction control measures. Accid. Anal. Prev. 40, 479486.
Peterson, B., Harrell, J., 1990. Partial proportional odds models for ordinal response
variables. Appl. Stat. 39, 205217.
Ponce, J.D.C., MuNoz, D.R., Andreuccetti, G., Carvalho, D.G.a.D., Leyton, V., 2011.
Alcohol-related trafc accidents with fatal outcomes in the city of Sao Paulo.
Accid. Anal. Prev. 43, 782787.
Radun, I., Summala, H., 2004. Sleep-related fatal vehicle accidents: characteris-
tics of decisions made by multidisciplinary investigation teams. Sleep 27 (2),
224227.
Rossi, R., Gastaldi, M., Gecchele, G., 2011. Analysis of driver task-related fatigue using
driving simulator experiments. Proc. Soc. Behav. Sci. 20, 666675.
Schlundt, D., Warren, R., Miller, S., 2004. Reducing unintentional injuries on the
nations highways: a literature review. J. Health Care Poor Underserved 15,
7698.
Shope, J.T., Bingham, C.R., 2008. Teen driving: motor-vehicle crashes and factors that
contribute. Am. J. Prev. Med. 35 (3S), S261S271.
Siskinda, V., Steinhardta, D., Sheehana, M., Oconnorb, T., Hanksb, H., 2011. Risk
factors for fatal crashes in rural Australia. Accid. Anal. Prev. 43, 10821088.
Stata, StataCorp LP, 2005. Stata Statistical Software: Release 9. StataCorp LP, College
Station, TX. Trafc Engineering Manual, 2007. Florida Department of Transporta-
tion.
Sudman, S., 1976. Applied Sampling. Academic Press, New York, NY.
Summala, H., 1996. Transportation and trafc in the information society-monotony
and overload. In: Rantanen, J., Lehtinen, S., Huuhtanen, P., Hrm, M., Laiti-
nen, H., Lehtel, J. (Eds.), Work in the Information Society. Finnish Institute of
Occupational Health, Helsinki, pp. 108115.
Thompson, J.P., Baldock, M.R.J., Mathias, J.L., Wundersitz, L.N., 2013. An examination
of the environmental, driver and vehicle factors associated with the serious and
fatal crashes of older rural drivers. Accid. Anal. Prev. 50, 768775.
Tsui, K., Sze, N., So, F., Wong, S., 2010. Association between drink driving and severity
of crash injuries to road users. Hong Kong J. Emerg. Med. 17 (1), 3439.
Violanti, J.M., 1998. Cellular phones and fatal trafc collisions. Accid. Anal. Prev. 30,
519524.
Wang, X., Abdel-Aty, M., 2008. Analysis of left-turn crash injury severity by
conicting pattern using partial proportional odds models. Accid. Anal. Prev.
40, 16741682.
Wanvik, P.O., 2009. Effects of road lighting: an analysis based on Dutch accident
statistics 19872006. Accid. Anal. Prev. 41, 123128.
Watkins, K., Sridhar, D., 2009. Road trafc injuries: the hidden development crisis
make roads safe: the campaign for global road safety. In: A Policy Brieng for
the First Global Ministerial Conference on Road Safety, 1920 November 2009,
Moscow, London.
Williams, R., 2006. Generalized ordered logit/partial proportional odds models for
ordinal dependent variables. Stata J. 6 (1), 5882.