You are on page 1of 25

IATSS Research

THE CHARACTERISTICS OF ROAD TRAFFIC CRASH SEVERITY ON PAVED AND


UNPAVED ROADS
--Manuscript Draft--

Manuscript Number: IATSS-D-20-00046

Article Type: Original Research Paper

Keywords: Paved road; unpaved road; crashes; mixed logit model; risk factors; Ghana

Abstract: Road traffic crashes has become a major problem for most countries including Ghana.
Though crashes on unpaved roads are low compared to the paved roads, there is a
disproportionately high incidence of traffic deaths and injuries on unpaved roads. The
study was to examine the characteristics and risk factors associated with road traffic
crashes on both paved and unpaved roads and identify potential measures to control
them. Police reported crash data for the period 2014-2016 in Ghana were aggregated
and analyzed. Factors influencing the severity of crashes on paved and unpaved roads
were determined by firstly using descriptive statistics and then developing two separate
mixed logit models for crashes on both road types. Relatively, unpaved roads were
found to be riskier for commuters as higher proportion of fatal crashes occurred on
them. Risk factors including gender and age of the driver, vehicle type, collision type
and road surface condition were established to be significant in the models. In order for
road safety interventions to succeed, road managers must appreciate and make
considerable political and economic commitments to address the issues presented by
these different types of facilities.

Powered by Editorial Manager® and ProduXion Manager® from Aries Systems Corporation
Highlights

Highlights

 Paved and unpaved roads crashes were studied using two separate mixed logit models.

 Significantly higher proportion of all crashes occurred on paved roads.

 Comparatively, unpaved roads were riskier for travelers than paved roads.

 Risk factors found included vehicle type, collision type and road surface condition.

 Road managers should understand the peculiar challenges presented by the facilities.
Manuscript (without Author Details) Click here to view linked References

Abstract

Road traffic crashes has become a major problem for most countries including Ghana. Though

crashes on unpaved roads are low compared to the paved roads, there is a disproportionately high

incidence of traffic deaths and injuries on unpaved roads. The study was to examine the

characteristics and risk factors associated with road traffic crashes on both paved and unpaved

roads and identify potential measures to control them. Police reported crash data for the period

2014-2016 in Ghana were aggregated and analyzed. Factors influencing the severity of crashes

on paved and unpaved roads were determined by firstly using descriptive statistics and then

developing two separate mixed logit models for crashes on both road types. Relatively, unpaved

roads were found to be riskier for commuters as higher proportion of fatal crashes occurred on

them. Risk factors including gender and age of the driver, vehicle type, collision type and road

surface condition were established to be significant in the models. In order for road safety

interventions to succeed, road managers must appreciate and make considerable political and

economic commitments to address the issues presented by these different types of facilities.

Keywords: Paved road; unpaved road; crashes; mixed logit model; risk factors; Ghana

1
1. Introduction

The transportation sector of any country is an important component of the economy impacting on

development and the welfare of the populace. When transport systems are efficient, reliable and

safe, they provide economic and social opportunities and benefits that result in positive

multiplier effects such as better accessibility to markets, employment, education and health and

enable the citizenry to save time, cost and suffering. According to the Ministry of Roads and

Highways (MRH), Ghana’s road transport infrastructure had increased to 14,367 km of trunk

roads, 48,100 km of feeder roads and 16,600 km of urban roads as at the end of the year 2017.

The Ghana Infrastructure Report Card by the Ghana Institution of Engineering [1] scored the

roads infrastructure a grade of D3 and determined that less than 30% of the entire national road

networks have paved surfaces.

The skill and technique required for driving on unpaved and paved roads are different.

For example, the coefficient of friction on gravel surfaces is much lower than on paved surfaces

[2]. Therefore, a longer stopping distance than on asphalt or concrete pavements under similar

conditions is usually needed. Again, surface conditions of gravel roads are likely to change with

time, space, and frequency and quality of maintenance work [2]. In addition to the surface

condition, there are other physical features of unpaved roads such as sharp horizontal and/or

vertical curves, narrow carriageways, narrow or no shoulder and narrow bridges that can have

influence on safety.

In Ghana, the available statistics from the Building and Road Research Institute of the

Council for Scientific and Industrial Research indicate that, on the average, about 2,000 people

are killed in road traffic crashes annually [3]. A possible limitation of the database is the level of

2
under-reporting of crashes. The extent of under-reporting of road traffic crashes has been studied

in Ghana [4]. However, it has not been accounted for in this study.

Although several research works (e.g., [5-8]) have studied the characteristics and risk

factors influencing road traffic crashes in Ghana, no studies have examined crashes on paved

versus unpaved roads. Managers of these roads need information on strategies to employ to

enhance the safe operations on the roads. The objective of this research work is to study the

characteristics of paved and unpaved road crashes, identify factors that influence these crashes

and recommend measures to forestall them. A paved road is defined as one with a hard smooth

surface, most commonly asphalt or concrete. Unpaved roads generally have gravel or dirt

surfaces [9].

2. Material and Methods

2.1. Data description

In this study, the severity of road traffic crashes on paved and unpaved roads in Ghana were

analyzed. The traffic crash data used for the study spanned over the three-year period between

2014 and 2016. The data were extracted from the national road traffic crash database maintained

by the Building and Road Research Institute of the Council for Scientific and Industrial

Research. The database was compiled from police reports. The definition of crash severity

follows that incorporated in the police traffic crash reporting form. The crash severity variable is

classified into four categories: fatal crash, serious crash, minor crash and property damage only.

The other variables considered have been grouped into four main sections: driver, vehicle,

roadway and environmental characteristics.

3
Over the study period, a total of 36,519 crashes occurred in Ghana. The frequency and

percentage distribution of all the variables considered for the study are presented in Table 1.

From the data, significantly higher proportion (97%) of all crashes occurred on paved roads

while only 3% occurred on unpaved roads. The proportion of fatal crashes was greater on the

unpaved roads as compared with the paved roads (29.27% vs. 14.60%) and the proportion of

young drivers (<26) involvement in crashes was higher on unpaved roads than for paved roads

(18.95 vs. 12.45). In general, unpaved roads were more unforgiving as higher proportion of

drivers who committed an error were involved in fatal crashes. Cars (53.10%) were most

frequent to be involved in crashes on paved roads. This was followed by buses (18.82%),

motorcycles (12.95%) and heavy goods vehicle (HGV) (11.85%). Again, for unpaved roads,

cars (37.30%) were mostly involved in crashes though the percentage was lower than for paved

roads. This was followed by motorcycles, HGVs and buses representing 25.02%, 18.31% and

12.18%, respectively. Noticeably, the proportions of motorcycles and HGVs involvement in

crashes was higher on unpaved roads. Most (73.30%) of the crashes occurring on paved roads

were multi-vehicle crashes whereas on the unpaved roads, they were single vehicle collisions

(51.10%).

4
Table 1: Frequency and percentage distribution of all the variables considered in the study
Paved road data Unpaved road data Combined data
Variable Frequency % Frequency % Frequency %
Total 35470 97.00 1049 3.00 36519 100
Crash severity
Fatal crash 5177 14.60 307 29.27 5484 15.02
Serious crash 8676 24.46 291 27.27 8967 24.55
Minor crash 8271 23.32 237 22.59 8508 23.30
Property damage only 13346 37.63 214 20.40 13560 37.13
Driver gender
Male 33273 96.27 1016 98.45 34289 96.33
Female 1290 3.73 16 1.55 1306 3.67
Driver’s age
Less than 26 4284 12.45 194 18.95 4478 12.63
26–60 29214 84.89 816 79.69 30030 84.74
Above 60 916 2.66 14 1.37 930 2.62
Driver error
At fault 23669 68.45 820 79.53 24489 68.77
Not at fault 10908 31.55 211 20.47 11119 31.23
Vehicle type
Car 18656 53.10 389 37.30 19045 52.65
HGV 4165 11.85 191 18.31 4356 12.04
Bus 6613 18.82 127 12.18 6740 18.63
Motorcycle 4550 12.95 261 25.02 4811 13.30
Bicycle 560 1.59 32 3.07 592 1.64
Other 589 1.68 43 4.12 632 1.75
Vehicle maneuver
Going ahead 29681 85.39 913 88.30 30594 85.47
Turn 1452 4.18 10 0.97 1462 4.08
Overtaking 1061 3.05 15 1.45 1076 3.01
Parked 1449 4.17 30 2.90 1479 4.13
Other 1116 3.21 66 6.38 1182 3.30
Vehicle defect
Yes 3900 11.27 123 11.92 4023 11.29
None 30707 88.73 909 88.08 31616 88.71
Number of vehicles
Single vehicle 9469 26.70 536 51.10 10005 27.40
Multi-vehicle 26001 73.30 513 48.90 26514 72.60

5
Table 1: Continued
Paved road data Unpaved road data Combined data
Variable Frequency % Frequency % Frequency %
Road description
Other 3587 10.11 296 28.22 3883 10.63
Straight and flat 31883 89.89 753 71.78 32636 89.37
Road width (meters)
Less than 7 20228 57.03 773 73.69 21001 57.51
7-14 14510 40.91 263 25.07 14773 40.45
Above 14 732 2.06 13 1.24 745 2.04
Surface condition
Dry 35170 99.15 1002 95.52 36172 99.05
Wet 300 0.85 47 4.48 347 0.95
Location type
Junction 9262 26.11 76 7.24 9338 25.57
Not at junction 26208 73.89 973 92.76 27181 74.43
Collision type
Head on 4280 12.07 211 20.11 4491 12.30
Rear end 11329 31.94 104 9.91 11433 31.31
Right angle 2757 7.77 20 1.91 2777 7.60
Side swipe 5808 16.37 137 13.06 5945 16.28
Ran off road 2896 8.16 245 23.36 3141 8.60
Hit object 8400 23.68 332 31.65 8732 23.91
Time of crash
Day time 22533 63.53 765 72.93 23298 63.80
Night time 12937 36.47 284 27.07 13221 36.20
Day of week
Weekday 24914 70.24 737 70.26 25651 70.24
Weekend 10556 29.76 312 29.74 10868 29.76
Weather
Other 5961 16.81 174 16.59 6135 16.80
Clear 29509 83.19 875 83.41 30384 83.20
Settlement type
Urban 24507 69.09 274 26.12 24781 67.86
Rural 916 2.58 56 5.34 972 2.66
Village 10047 28.33 719 68.54 10766 29.48

6
2.2. Mixed logit model specification

The traffic crashes severity outcome at any of the road types (paved or unpaved) can be modelled

as multi-category variable using multinomial logit model [10, 11]. However, the multinomial

logit model assumed that the impact of the explanatory variables on the crash severity outcome is

fixed across individual vehicle. In order to account for heterogeneity across vehicles, the mixed

logit also known as random parameter model was implemented. Following previous study [12,

13], the mixed logit model is formulated by first defining a function that determine the vehicle

crash severity as:

Sin  β i Xin   in (1)

where S in is a severity function determining the crash severity category i in crash n, X in

represents a vector of explanatory variables that influence crash severity category i in crash n, β i

represents a vector of estimable model parameters for vehicle crash severity category i, and  i

represents an error term. If the error term is assumed to follow generalized extreme value

distribution, then the standard multinomial logit model can be expressed in probability form as:

exp β i X in 
Pin  (2)
 exp β i X in 
i I

where Pin represents the probability that a vehicle n sustains crash severity outcome i and I

represents the set of all crash severity outcome (i.e., fatal, serious, minor and property damage

only). To account for individual vehicle heterogeneity, a mixed logit model, which is a

generalization of Equation (2), is used and it is defined as:

exp β i X in 
Pin    exp β i Xin  f β |  d (3)

i I

7
where f  |   is a density function of  and which accounts for the individual vehicle

heterogeneity whilst  denotes a vector of parameters (mean and standard deviation) describing

the density function. Some of the elements of  may be fixed and some may be randomly

distributed. But if all the elements in are fixed then the mixed logit model will reduces to the

standard multinomial logit defined in Equation 2 [11]. If the standard deviation in  is

significantly different from zero, then the associated parameter is considered to be random. The

mixed logit probability for a specific individual vehicle crash severity outcome are calculated by

integrating the conditional probability over the distribution of the random variable  [12].

The parameters of the mixed logit model are estimated using a Monte Carlo simulation-

based maximum likely approach with logit probabilities approximated by 200 random draw

based on Halton sequence from a specified distribution. The most commonly used distributions

include normal, lognormal, triangular, Weibull, and uniform [11]. However, previous literature

have shown that the normal distribution is more appropriate for crash severity data [12, 14]. The

impact of each explanatory variable on the crash severity outcome in the mixed logit model was

interpreted by calculating the average marginal effect usually expressed in percentage as:

P P X  1  Pin X ink  0
 Xin  in ink  100% (4)
ink Pin X ink  0

where Pin X ink  1 is the predicted probability of specific to each crash severity outcome i for

individual vehicle n calculated when the kth explanatory variable, X ink is equal to 1. The

marginal effect is estimated by averaging the simulation-based marginal effect over all

observations.

8
3. Empirical Results

Factors influencing the severity of vehicle crashes on paved and unpaved roads were determined

by developing two separate mixed logit models for crashes on both road types. Following the

existing literature on mixed logit modelling [12], all the parameters in each model were first

specified as random. The significance of the standard deviation associated with each random

parameter were determined. The parameters were then considered to be fixed if the

corresponding standard deviation of each parameter distribution is not significant at 5% alpha

level. The estimated parameters of mixed logit model for crashes on paved and unpaved roads

are summarized in Tables 2 and 3, respectively. For both models, all the parameters retained are

significant at 5% alpha level and their goodness-of-fit test statistics suggest that the models are

satisfactory. The impact of each explanatory variable on the crash severity were determined by

calculating the associated marginal effect. The marginal effects for paved and unpaved crashes

models are presented in Tables 4 and 5, respectively. The interpretation of the models is done

under the following sections: driver, vehicle, roadway and environmental characteristics.

3.2.1 Driver characteristics

The results of the mixed logit models suggest that gender of the driver significantly influence the

severity of crashes on paved roads but not on unpaved roads. That is, males were associated with

more severe crashes compared to females on paved roads. According to the marginal effect, the

probability for a male driver to be involved in fatal crashes increased by 3% whilst the likelihood

of involving in non-fatal crashes decrease by 0.07–2.2%. With respect to drivers age, drivers

with age 26-60 years were more likely to be involved in fatal crashes on paved roads. The

marginal effect indicates that the probability for involving in fatal crashes increased by 2% for

such drivers. When the driver is at fault on paved roads, the crash severity was more likely to

9
result into fatalities. The probability for at fault drivers to be involved in fatal crashes increased

by 0.6%.

3.2.2 Vehicle characteristics

Vehicle type was found to influence crash severity on both road surface types. For instance,

heavy goods vehicles were found to increase the likelihood of fatal crashes on both roads. The

marginal effect indicates that the probability for heavy goods vehicle to be involved in fatal

crashes increased by 10% on paved roads and 12% on unpaved roads. Buses were more likely to

be associated with severe crashes on paved roads whiles for unpaved roads, they were more

likely to be associated with serious or minor crashes. The probability for a bus to be involved in

more severe (fatal/serious) crashes increased by 1-7% and less severe crashes (minor crashes/

prosperity damage only) decrease by 0.1-7% on paved roads. In the case of unpaved roads, the

probability for a bus to be involved in fatal crashes decrease by 10% whilst that of serious or

minor crash increase by 6-9% but that of property damage decrease by 5%. Motorcycle was

associated with more likelihood to be involved in more severe crashes on paved and unpaved

roads. According to the marginal effect, the probability for the crash severity to be classified as

more severe increased by 1-18% on paved roads and 1-13% on unpaved roads. The parameter for

minor crashes on paved roads was normally distributed with mean of 1.718 and standard

deviation of 2.826. This shows that minor crash severity associated with motorcycle has

heterogeneous effect on paved roads.

10
Table 2: Parameters of mixed logit model for crashes on paved road data*
Fatal Hospitalized Minor Injury
Variable Coefficient P-value Coefficient P-value Coefficient P-value
Driver characteristics
Male driver 0.639 <0.001 0.469 <0.001 0.272 <0.001
26-60 years 0.140 0.004
At fault driver 0.251 <0.001 0.236 <0.001
Vehicle characteristics
Heavy goods vehicle 0.825 <0.001
Bus 0.724 <0.001 0.429 <0.001 0.304 <0.001
Motorcycle (standard deviation)* 3.765 <0.001 3.452 <0.001 1.718 (2.826) <0.001 (<0.001)
Bicycle 3.125 <0.001 2.181 <0.001
Other vehicle type 0.924 <0.001 0.309 0.007
Turn 0.293 <0.001
Overtaking 0.541 <0.001 0.283 0.004 0.286 0.002
Parked -0.263 0.004 -0.230 0.004
Other vehicle action -0.301 0.007
Defected vehicle -0.279 <0.001 -0.133 0.003
Multi-vehicle -2.124 <0.001 -1.967 <0.001 -1.357 <0.001
Roadway characteristic
Straight and flat -0.743 <0.001 -0.460 <0.001 -0.397 <0.001
7 – 14 meters 0.639 <0.001 0.255 <0.001 0.174 <0.001
Above 14 meters -0.716 <0.001
Junction -0.822 <0.001 -0.530 <0.001 -0.325 <0.001
Rear end -2.445 <0.001 -1.775 <0.001 -1.191 <0.001
Right angle -1.720 <0.001 -1.048 <0.001 -0.481 <0.001
Side swipe -2.567 <0.001 -1.569 <0.001 -1.061 <0.001
Ran off road (standard deviation)* -3.695 (1.818) <0.001 (<0.001) -2.455 <0.001 -1.394 <0.001
Hit object -1.066 <0.001 -0.804 <0.001 -0.694 <0.001
Environmental characteristics
Night time 0.198 <0.001

11
Weekend 0.241 <0.001 0.086 0.007
Clear weather -0.347 <0.001
Rural 0.684 <0.001 0.593 <0.001 0.562 <0.001
Constant 1.521 <0.001 1.861 <0.001 1.542 <0.001
Goodness of fit statistics
Number of observations 137084
Restricted Log likelihood (constant only) -47276
Log-likelihood at convergence -39024
#
Values in the parenthesis represent the standard deviation of the random parameter distribution
0
Table 3: Parameters of mixed logit model for crashes on unpaved road data
Fatal Hospitalized Minor
Variable Coefficient P-value Coefficient P-value Coefficient P-value
Vehicle characteristics
Heavy goods vehicle 0.593 0.002
Bus 0.535 0.026 0.693 0.002
Motorcycle 2.551 <0.001 2.929 <0.001 2.608 <0.001
Other vehicle type 1.259 <0.001
Defected vehicle -0.501 0.049
Multi-vehicle -0.911 0.001 -0.667 <0.001
Roadway characteristics
Junction -0.903 0.005 -0.783 0.016
Rear end -0.804 0.005 -1.249 <0.001
Side swipe -1.626 <0.001 -1.440 <0.001
Ran off road -1.609 <0.001 -0.838 <0.001
Hit object -0.569 0.043
Dry road surface 1.330 <0.001 0.428 0.001
Goodness-of-fit Statistics
Number of observations 4148
Restricted Log likelihood (constant only) -1451
Log-likelihood at convergence -1304
1

12
Bicycles were more likely to be involved in fatal or serious crashes on paved roads. The

probability for bicycle to be involved in fatal crashes increased by 27% on paved roads whilst the

probability for non-fatal crashes decreased by 2–16%. Other types of vehicles (e.g. tractors) have

high likelihood of involving in fatal crashes on both road types. The probability for other types

of vehicles to be involved in fatal crashes increased by 10% on paved roads and 29% on unpaved

roads. Vehicle maneuver significantly influence the crash severity of vehicle when involve in a

crash on paved roads. The results suggest that turning vehicles were more likely to be involved

in serious crashes and the probability for such crashes increased by 4%. Overtaken vehicles were

more likely to be associated with more severe crashes. The probability for overtaken vehicles to

be involved in fatal crashes increased by 4% whilst that of non-fatal crashes decreased by 0.04–

3% on paved road. Parked vehicles were less likely to cause minor crashes on paved roads and

the probability for less severe crashes decrease by 3% or less. Other maneuvering actions were

less likely to cause fatal crashes on paved roads and the probability associated with that

decreased by 4%. Defected vehicles were found to be significant factor for injury severity of

crashes occurring on paved and unpaved roads. According to the marginal effects, the probability

for fatal crashes due to defected vehicle decreased by 4% on paved roads and 12% on unpaved

roads whilst the probability for nonfatal crashes increased by 0.2–4% on paved roads and 0.3 –

7% on unpaved roads. Multi-vehicle crashes were associated with less probability for fatal

crashes on paved and unpaved roads. The results showed that the probability for fatal crashes

decreased by 2% on paved roads and 15% on unpaved roads.

13
Table 4: Average marginal effect (%) of mixed logit model for crashes on paved road
Variable Fatal Hospitalized Injury Damage
Driver characteristics
Male driver 3.09 -2.24 -0.78 -0.07
26-60 years 2.17 -1.93 -0.24 -0.01
At fault driver 0.64 -0.06 -0.55 -0.02
Vehicle characteristics
Heavy goods vehicle 9.75 -4.42 -2.67 -2.66
Bus 6.64 0.84 -0.06 -7.42
Motorcycle 18.26 11.01 1.02 -30.28
Bicycle 27.43 -8.88 -16.38 -2.17
Other vehicle types 10.13 -7.79 -2.23 -0.11
Turn -3.53 3.68 -0.14 -0.01
Overtaking 3.64 -3.17 -0.43 -0.04
Parked 3.16 -2.80 -0.36 0.00
Other vehicle action -3.71 3.21 0.49 0.02
Defected vehicle -3.73 3.53 0.19 0.02
Multi-vehicle -3.22 1.76 1.37 0.10
Roadway characteristics
Straight and flat -3.77 3.17 0.55 0.05
7 – 14 meters 5.40 -4.58 -0.78 -0.05
Above 14 meters 0.87 0.19 -1.06 0.00
Junction -4.05 3.15 0.84 0.06
Rear end -11.19 8.02 2.62 0.55
Right angle -15.00 9.64 3.79 1.57
Side swipe -25.91 13.97 5.06 6.88
Ran off road -22.63 -15.83 4.17 34.29
Hit object -6.22 -9.61 -2.16 17.99
Environmental characteristics
Night time 2.81 -2.47 -0.32 -0.01
Weekend 2.48 -2.00 -0.45 -0.02
Clear weather -4.35 3.80 0.53 0.02
Rural 3.55 7.32 2.22 -13.08

14
Table 5: Average marginal effect (%) of mixed logit model for crashes on unpaved road
Variable Fatal Hospitalized Injury Damage only
Vehicle characteristics
Heavy goods vehicle 12.08 -5.25 -3.67 -3.16
Bus -9.85 5.97 8.50 -4.62
Motorcycle 0.58 12.64 3.42 -16.63
Other vehicle type 28.75 -16.38 -11.76 -0.61
Defected vehicle -11.90 6.70 4.94 0.26
Multi-vehicle -14.98 16.36 -2.03 0.66
Roadway characteristics
Junction -10.04 -5.78 14.76 1.06
Rear end -6.19 -16.47 21.13 1.52
Side swipe -19.31 -11.22 28.49 2.05
Ran off road -9.38 -4.02 12.50 0.90
Hit object -1.42 0.06 1.27 0.09
Dry road surface 21.11 -5.95 -14.14 -1.02

3.2.3 Roadway characteristics

Straight and flat paved roads significantly decrease crash severity. According to the marginal

effect, the probability for fatal crashes on straight and flat paved road decrease by 4% whilst that

of non-fatal crashes increased by 0.1–3%. The width of road was found to be a significant factor

for injury severity occurring on paved road. The results indicate that crashes occurring on wider

paved roads significantly increase crash severity. The probability for fatal crashes increased by

5% on paved roads having width size of 7–14 meters. However, property damage only crashes

decreased when the road width exceeds 14 meters. Location type was found to be associated with

the severity of crashes on paved and unpaved roads. That is, crashes occurring at junctions on

paved roads decrease the probability of sustaining fatal injury by 10% and increase the

probability for non-fatal injuries by 0.1–3%. For crashes occurring at junctions on unpaved

roads, the probability of sustaining fatal and serious injuries decreased by 10% and 6%,

respectively and increase less severe crashes by 1–15%. Collision type significantly influence the

severity of crashes on paved and unpaved roads. For instance, rear end crashes were associated

15
with 11% less likely of sustaining fatal injury and 1-8% more likely of sustaining non-fatal

injury on paved roads whilst for unpaved roads the probability for more severe injury decreased

by 6–16% and that of less severe injury increased by 2–21%. Right angle crashes were

associated with less likelihood to result into more severe crashes on paved roads. The probability

of being involved in fatal crashes during right angle collisions decrease by 15% but the

probability increase by 1–10% for non-fatal crashes. Side swipe collisions on paved roads

decrease the probability of fatal crashes by 26% and increased non-fatal crashes by 5–14%

whiles the crashes occurring on unpaved roads decreased fatal and serious injury by 11–19% and

increased less severe crashes by 2–28%. Ran off road crashes were associated with less

likelihood for fatal and serious crashes irrespective of road surface type. However, the parameter

for fatal crashes on paved roads was normally distributed with mean of -3.695 and standard

deviation of 1.818. According to the marginal effects, ran off road crashes on paved roads

decreased the probability for more severe crashes by 16–23% and the probability for less severe

crashes increased by 4–34%. For unpaved roads, ran off road crashes decreased the probability

for more severe crashes by 4–9% and the probability for less severe crashes increased by 1–13%.

Vehicle crashes involving road side objects significantly influences the crash severity. According

to the average marginal effect, the probability for more severe crashes decreased by 2–10% and

that of property damage only crashes increased by 18% on paved roads. For unpaved roads, the

probability for fatal crashes decreased by 1% and that of non-fatal crashes increased by 1% or

less. Road surface condition significantly influence crash severity on unpaved roads but not

paved road. Compared to unpaved road with wet surface, dry surface unpaved roads were found

to be significantly associated with high injury severity. The probability of being involved in fatal

crashes increased by 21% and that of non-fatal crashes decreased by 1–14%.

16
3.2.4 Environmental characteristics

The analysis shows that none of the variables relating to environmental characteristics were

found to influence crash severity on unpaved roads. The model results suggest that night-time

crashes significantly increase the likelihood of involving in fatal crashes. According to the

marginal effect, the probability of involving in fatal crashes during night-time increased by 2.8%.

With respect to day of week, the results showed that weekend crash were associated with fatal

and serious crashes on paved roads. That is, the probability of being involved in fatal crashes

increased by 2.5% whilst that of non-fatal injury decreased by 0.02–2% for weekend crashes.

Fatal crashes significantly decrease on paved roads during clear weather condition. The

probability of being involved in fatal crashes on paved road during clear weather condition

significantly decrease by 4.4% whilst that of non-fatal crashes increase by 0.02–4% compared to

adverse weather condition. Crashes occurring on rural areas were significantly associated with

more severe crashes. That is the probability of sustaining more severe injury increased by 2–7 %

when involved in crashes on paved rural roads. On the other hand, the probability of being

involved in properties damage only crashes decrease by 13%.

4.0 Discussion

According to Wu, et al. [15], drivers are more cautious when driving on unpaved roads compared

to paved roads. This, and the high level of traffic exposure (vehicle kilometer travelled) may

explain the high proportion of crashes on the paved as against the unpaved roads. It was

observed that the likelihood of male drivers to be involved in fatal crashes increased by 3%,

drivers aged 26-60 years increased by 2% and probability of driver at fault involving in fatal

crashes increased by 0.6%. The findings of the study agree with existing literature [16-18] which

17
concluded that adolescents and younger adults are most prone to risk taking behavior,

particularly young males.

When it comes to vehicle characteristics, heavy goods vehicles were found to increase the

likelihood of fatal crashes on both roads. The probability of heavy goods vehicles to be involved

in fatal crashes increased by 10% on paved roads and 12% on unpaved roads. Ackaah [19] found

that excessive loading by goods vehicles, inappropriate use of goods vehicle for passenger

transport, poor road side parking without warning signs, driver fatigue and poor night visibility

are the key contributory risk factors associated with these crashes. The probability of motorcycles

to be associated with fatal crashes was significant on both paved and unpaved roads. It increased by

1-18% on paved roads and 1-13% on unpaved roads. The higher percentage increment on paved than

unpaved roads can be attributed to higher speeds on the paved roads. Cyclists were also associated

with higher fatal crashes on the paved roads because they are vulnerable road users in the high speed

mix traffic. Mostly, even roads in urban areas hardly have exclusive lanes for bicycles. The more

death of motorcyclists could be as a result of non-helmet use as previous studies [20, 21] established

that motorcycle riders and passengers without helmet are more prone to fatal crashes. The

probability for other vehicle types (e.g., tractors) to be involved in fatal crashes increased by

10% on paved roads and 29% on unpaved roads.

With roadway characteristics; almost all of the factors understudy (location and collision

types) were significant on both paved and unpaved roads. The results from hit object and side

swipe collisions on unpaved roads were consistent with the findings of Liu and Dissanayake

[22], undoubtedly due to the narrow width of such roads. However, road surface condition,

specifically, dry road surface significantly influence crash severity on unpaved roads but not on

paved road. When compared with wet road surface on the unpaved roads, dry surfaces were

found to be significantly associated with fatal injury severity. Thus, the probability increased by

18
21% in fatal crashes and decreased by 1-14% in non-fatal crashes. With respect to road surface

type, Liu and Dissanayake [22] also found crashes on slippery road surface likely decreasing in

high injury crashes and explained that this might be due to drivers being more careful when

driving on slippery road surfaces as compared to dry road surfaces. Again, surface conditions on

gravel or dirt roads change with time and distance. The high number of fatal crashes on the dry

unpaved roads may, partly, be explained by dust which blur the vision of drivers, particularly,

when the vehicle is following another vehicle or there is an on-coming vehicle.

With environmental characteristics; none of the variables (time, day, weather, people

settings) used in this study were found to significantly influence crash severity on unpaved roads

but rather only on paved roads. Consistent with other studies, fatal night-time crashes reduce

with good road lighting (Ackaah & Adonteng, 2011; Owens and Sivak, 1993). The results

showed that weekend crash were associated with fatal and serious crashes on paved roads.

Weekends in Ghana are characterized by social gatherings such as funerals and weddings which

are packed with drinking and have long been associated with drunken driving (Ackaah &

Adonteng, 2011). This study also found crashes on paved rural roads to be significantly

associated with severe injuries. Excessive vehicular speeds on the rural highways, coupled with

the delay and poor handling of crash victims before being transported to medical centers have

greatly contributed to the high number of severe traffic injuries on the non-urban highways

(Afukaar, et al., 2003). Generally, this study and other retrospective studies are mostly limited by

incomplete documentation of some variables and also the inability to retrieve missing data.

19
4.1 Conclusions

Unpaved roads represent a significant proportion of the public roadway system in the country.

Road managers should understand the types of safety challenges presented by these facilities as

well as the types of safety interventions required to address these challenges. This study assessed

the characteristics of paved and unpaved road crashes, determined factors influencing the

severity of these crashes by developing two separate mixed logit models for crashes on both road

types. Overall, unpaved roads were more unforgiving as higher proportion of drivers who

committed an error were involved in fatal crashes.

Cars (53.10%) were most frequent to be involved in crashes on paved roads followed by

buses (18.82), motorcycles (12.95%) and heavy goods vehicles (11.85%) whereas on unpaved

roads, it was cars (37.30%), motorcycles (25.02%), heavy goods vehicles (18.31%) and buses

(12.18%). From the two separate mixed logit model for crashes on both road types, factors

including gender, age of the driver, vehicle type, collision type and road surface condition were

established to influence road traffic crashes on the different road types.

In order for road safety interventions to succeed, governments should make considerable

political and economic commitments to address the problem. Measures to deal with the problem

should target both the unpaved and paved roads. For the deaths on the paved roads, focus should

be on cars, buses and motorcycles whereas on the unpaved roads, focus should be on cars,

motorcycles and heavy goods vehicles.

Declarations of interest: none

20
References

[1] Ghana Institution of Engineering, Ghana infrastructure report card. . Accra, Ghana: Ghana

Instition of Engineering, 2016.

[2] S. Dissanayake and L. Liu, "Speed limit-related issues on gravel roads," Kansas Department

of Transportation2009.

[3] W. Ackaah, J. Larley, and J. Larbi, "Road Traffic Crashes in Ghana - Statistics 2018," CSIR-

Building and Road Research Institute, Kumasi, Ghana2020.

[4] M. Salifu and W. Ackaah, "Under-reporting of road traffic crash data in Ghana,"

International Journal of Injury Control and Safety, vol. 19, pp. 331-339, 2012.

[5] W. Ackaah and D. O. Adonteng, "Analysis of fatal road traffic crashes in Ghana,"

International Journal of Injury Control and Safety Promotion, vol. 18, pp. 21-27, 2011.

[6] E. N. Aidoo, R. Amoh-Gyimah, and W. Ackaah, "The effect of road and environmental

characteristics on pedestrian hit-and-run accidents in Ghana," Accident Analysis and

Prevention, vol. 53, pp. 23-27, 2013.

[7] E. N. Aidoo and W. Ackaah, "A generalized ordered logit analysis of risk factors associated

with driver injury severity," Journal of Public Health: From Theory to Practice, pp. 1-7,

2019.

[8] F. K. Afukaar, P. Antwi, and S. Ofosu-Amaah, "Pattern of road traffic injuries in Ghana:

implications for control," Injury Control and Safety Promotion, pp. 69-76, 2003.

[9] G. M. Denning and C. A. Jennissen, "All-terrain vehicle fatalities on paved roads, unpaved

roads, and off-road: evidence for informed roadway safety warnings and legislation," Traffic

Injury Prevention vol. 17, pp. 406-412, 2016.

21
[10] F. Ye and D. Lord, "Comparing three commonly used crash severity models on sample

size requirements: multinomial logit, ordered probit and mixed logit models," Analytic

methods in accident research, vol. 1, pp. 72-85, 2014.

[11] K. E. Train, Discrete choice methods with simulation. Cambridge University Press, New

York, 2003.

[12] Q. Wu, G. Zhang, X. Zhu, X. C. Liu, and R. Tarefder, "Analysis of driver injury severity

in single-vehicle crashes on rural and urban roadways," Accident Analysis and Prevention,

vol. 94, pp. 35-45, 2016.

[13] S. Washington, M. Karlaftis, and F. Mannering, Statistical and econometric methods for

transportation data analysis, 2ed. ChapmanandHall/CRC, Boca Raton,FL., 2011.

[14] F. Chen and S. Chen, "Injury severities of truck drivers in single- andmulti-vehicle

accidents on rural highways," Accident Analysis and Prevention, vol. 43, pp. 1677-1688,

2011.

[15] Q. Wu, F. Chen, G. Zhang, X. C. Liu, H. Wang, and S. M. Bogus, "Mixed logit model-

based driver injury severity investigations in single-and multi-vehicle crashes on rural two-

lane highways," Accident Analysis & Prevention, vol. 72, pp. 105-115, 2014.

[16] M. E. Aitken, C. Graham, J. B. Killingsworth, S. H. Mullins, D. Parnell, and R. M. Dick,

"All-terrain vehicle injury in children: strategies for prevention," Injury Prevention, vol. 10,

pp. 303-307, 2004.

[17] J. Chen, M.-j. Kresnow, T. R. Simon, and A. Dellinger, "Injury-prevention counseling

and behavior among US children: results from the second Injury Control and Risk Survey,"

Pediatrics, vol. 119, pp. e958-e965, 2007.

22
[18] A. Grummon, C. Heaney, W. Dellinger, and J. Wilkins, "What influences youth to

operate all-terrain vehicles safely?," Health education research, vol. 29, pp. 533-546, 2014.

[19] W. Ackaah, "Road traffic crashes involving goods vehicles in Ghana," Journal of the

Ghana Institution of Engineers, vol. 6&7, pp. 25-30, 2009.

[20] G. M. Denning, K. K. Harland, D. G. Ellis, and C. A. Jennissen, "More fatal all-terrain

vehicle crashes occur on the roadway than off: increased risk-taking characterises roadway

fatalities," Injury prevention, vol. 19, pp. 250-256, 2013.

[21] G. M. Denning and C. A. Jennissen, "All-terrain vehicle fatalities on paved roads,

unpaved roads, and off-road: evidence for informed roadway safety warnings and

legislation," Traffic injury prevention, vol. 17, pp. 406-412, 2016.

[22] L. Liu and S. Dissanayake, "Factors affecting crash severity on gravel roads," Journal of

Transportation Safety & Security, vol. 1, pp. 254-267, 2009.

23

You might also like