Accident Analysis and Prevention 53 (2013) 23 27

Contents lists available at SciVerse ScienceDirect

Accident Analysis and Prevention

j ourna l h o mepage: www.elsev ier .co

The eff risacciden

Eric Nim ahCSIR-Building a

a r t i c l

Article history:Received 15 AReceived in re12 December 2Accepted 14 D

Keywords:PedestrianHit-and-runRoad accidentLogistic regresGhana

ied as. Madestfect oedestd Roanary idennctioedest

insta

1. Introduction

Death aglobal phenAccording tworldwide

In addititrafc acciddown by veaccident hainjury, disa2011). Mostligence of b(2000), bothtrafc regucaution. Rotributing fasubgroup ofcontinue toin Ghana, aDerry et al.,are killed on

CorresponKumasi, Ghana

E-mail add

of pedestrians is the leading cause of fatalities among urban roadusers in Ghana (Ackaah and Adonteng, 2011; Damsere-Derry et al.,

0001-4575/$ http://dx.doi.ond injury associated with road trafc accidents areomena that require urgent attention (Tay et al., 2009).o Peden et al. (2004), approximately 1.2 million peopleare killed each year as a result of road trafc accidents.on to increases in death of passengers involved in roadents, pedestrian deaths that result from being knockedhicles have also increased in recent years. Pedestrians been found to be one of the single largest causes ofbility and death in the developed world (Sullman et al.,

accidents involving pedestrians are caused by the neg-oth the pedestrian and the driver. According to Hamed

drivers and pedestrians in developing countries ignorelations and thus take greater risks and exercise lesssenbloom et al. (2004) described pedestrians as a con-ctor to trafc accidents, and they constitute a large

those who are seriously and fatally injured. Pedestrians be the road user with the highest risk of death in trafcnd their actions predispose them to injury (Damsere-

2010; Afukaar et al., 2011). The number of people who the roads in Ghana is unacceptably high, and the death

ding author at: CSIR-Building and Road Research Institute, Box UP 40,. Tel.: +233 247581500; fax: +233 3220 60080.ress: en.aidoo@yahoo.com (E.N. Aidoo).

2010). Ackaah and Adonteng (2011) showed that pedestrian colli-sion alone resulted in approximately 51.7% of all fatal accidents inGhana during the period 20052007.

The death of many pedestrians who were involved in road trafcaccidents resulted from the driver leaving the pedestrian who wasstruck at the scene of the accident. Pedestrians who did not dieimmediately after the collision later died as a result of poor traumacare (Mock et al., 1997; Peden et al., 2004). For example, most ofthese accident victims may not have died if they had been rushedto the hospital for medical treatment immediately following theaccident. Hit-and-run drivers also increase the victims exposure tobeing struck again by a subsequent vehicle (MacLeaod et al., 2012).Hit-and-run is described as the drivers decision to leave the sceneafter hitting a pedestrian without reporting to the police.

Reducing the number of drivers who ee after hitting apedestrian will help to reduce deaths associated with pedestrianaccidents. According to Tay et al. (2009), one way to reduce hit-and-run accidents is to increase the level of trafc enforcement.These authors further argue that educating drivers may changetheir attitudes and beliefs about hit-and-run accidents.

There are numerous diverse factors that contribute to hit-and-run accidents in general and in Ghana, in particular. This researchconsiders factors such as trafc control, road characteristics andsurface conditions, accident severity, light condition and day ofthe week. Previous research has conrmed that these factors are

see front matter 2012 Elsevier Ltd. All rights reserved.rg/10.1016/j.aap.2012.12.021ect of road and environmental charactets in Ghana

ako Aidoo , Richard Amoh-Gyimah, Williams Ackand Road Research Institute, Kumasi, Ghana

e i n f o

ugust 2012vised form012ecember 2012

sion

a b s t r a c t

The number of pedestrians who have din addition to vehicle passenger deathas a result of the driver leaving the peThis paper seeks to determine the efand-run accidents in Ghana. Using pAccident Database at the Building anIndustrial Research (CSIR), Ghana, a biestimated model indicate that fatal accroad sections without medians and juwill leave the scene after hitting a pinto road design and construction andhit-and-run accidents in Ghana.m/locate /aap

tics on pedestrian hit-and-run

s a result of being hit by vehicles has increased in recent years,ny pedestrians who were involved in road trafc accident diedrian who was struck unattended at the scene of the accident.f road and environmental characteristics on pedestrian hit-rian accident data extracted from the National Road Trafcd Research Institute (BRRI) of the Council for Scientic and

logit model was employed in the analysis. The results from thets, unclear weather, nighttime conditions, and straight and atns signicantly increase the likelihood that the vehicle driverrian. Thus, integrating median separation and speed humpslling street lights will help to curb the problem of pedestrian

2012 Elsevier Ltd. All rights reserved.

24 E.N. Aidoo et al. / Accident Analysis and Prevention 53 (2013) 23 27

signicant inuences on the drivers decision to leave or stay on thescene after a pedestrian accident. For instance, light condition, dayof the week and trafc condition have been found to be contribut-ing factors in a drivers decision to leave the scene of the accidentafter strikin1997; Tay eMacLeaod etrian and endrivers decof the accidlikely to eans are invo

This papsafety in Gheffect of roaand-run acroad and enaccidents inand ensuringrams thataccidents.

The remdescribes ththe data, emcluding rem

2. Method

To explaconditions scene of pewas emploThis type oof the dephit-and-runthe probabthe indepenThe indepecan be cateregression the general1972; Agresthe binary dalso known

log[

P(Y 1 P(Y

= + 1x

where P(Y ing the scenP(Y = 1)] rescene (Agreof the drivebetween 0 aIn Eq. (1), are the regrof covariatecan be estimmated modtest to deteBecause themation, logbased on thcient, whichthe odds rat

Table 1Descriptive statistics of pedestrian accidents (20042010).

Variable No. ofaccidents

No. of hit-and-run

Hit-and-run (%)

45 6 7 8 90

anti 3730 195 5.2g Ahafo 1149 73 6.4tral 2070 113 5.5ern 2929 145 5.0ter Accra 8561 980 11.4thern 255 22 8.6er East 224 17 7.6er West 92 5 5.4a 1140 64 5.6tern 1428 54 3.8

nvironmentan 13,940 1197 8.6-urban 7625 471 6.2ing 13 0

weekkdays 14,900 1124 7.5kend 6678 544 8.1

nt severityl 5444 525 9.6-fatal 16,134 1143 7.1

er conditionr 20,088 1492 7.4ers (rain, fog/, etc.)

1490 176 11.8

ondition 14,612 934 6.4t no light 4186 451 10.8t light on 2767 281 10.2

escriptionight and at 19,631 1568 8.0e and inclined 1927 94 4.9

eparationian 4238 343 8.1edian 17,312 1319 7.6

urface condition21,386 1652 7.7

143 5 3.5

urface repaird 19,259 1520 7.9gh 2269 135 5.9

on typeat junction 17,173 1434 8.4tion 4360 228 5.2

conditione 12,055 944 7.8age 9497 719 7.6

rcentages for missing observations are not shown in the table.

ccident with respect to a change in the covariates being con- in the model. The data used in this research were extractedhe National Road Trafc Accident Database at the Buildingad Research Institute (BRRI) of the Council for Scienticdustrial Research (CSIR), Ghana. The database is compiledad trafc accident les from the Motor Trafc and TransportTTU) of the Ghana Police Service. The database at the BRRI

ect to two shortfalls: non-reporting and under-recording.g a pedestrian (Solnick and Hemenway, 1994; Johnson,t al., 2008, 2009). Solnick and Hemenway (1995) andt al. (2012) concluded in their research that pedes-vironmental characteristics signicantly inuence theision to participate in a hit-and-run by leaving the sceneent. These researchers further found that drivers are lesse the scene of the collision if younger or older pedestri-lved.er aims to contribute to the existing literature on roadana by providing empirical results and identifying thed and environmental characteristics on pedestrian hit-cidents by drivers in Ghana. Identifying the effects ofvironmental characteristics on pedestrian hit-and-run

Ghana is important for designing road safety policiesg the successful implementation of road safety pro-

will reduce pedestrian mortality due to hit-and-run

ainder of the paper is organized as follows: Section 2e methods employed in the study; Section 3 presentspirical analysis and results; and Section 4 provide con-arks.

ology

in the effect of the road, environmental and weatheron the decision of the driver to leave or stay at thedestrian accidents, a binary logistic regression modelyed (Solnick and Hemenway, 1995; Tay et al., 2009).f model was used because of the dichotomous natureendent variable (i.e., hit-and-run as opposed to not). This type of model makes it possible to estimateility of the driver leaving the accident scene based ondent variables incorporated into the regression model.ndent variables incorporated into the regression modelgorical or quantitative in nature. The binary logisticmodel is among the family of regression models inized linear model framework (Nelder and Wedderburn,ti, 2007). The model assumes a binomial distribution forependent variable. A binary logistic regression model,

as a logit model, is dened in the following manner:

= 1) = 1)

]= log

(

1 )

= log it()

1 + 2x2 + + nxn, (1)

= 1) = describes the probability of the driver leav-e after a pedestrian accident (hit-and-run), while [1 presents the probability of the driver not leaving thesti, 2007). The logit model predicts the probability ()r leaving the scene of the accident. This probability fallsnd 1 (0 1) for all possible independent variables.

is the intercept term in the model, is (i = 1, 2, . . . , n)ession coefcients for each covariate and xs are the sets/independent variables. The parameters in the modelated using a maximum likelihood approach. The esti-el can be evaluated by performing a likelihood ratiormine the signicance of the covariates in the model.

dependent variable is modeled using a log transfor- it(), the interpretation of the estimated coefcient ise exponential transformation of the estimated coef-

is commonly known as the odds ratio. For this study,io indicates the likelihood of the driver leaving the scene

Year200200200200200200201

RegionAshBronCenEastGreaNorUppUppVoltWes

Road eUrbNonMiss

Day ofWeeWee

AccideFataNon

WeathCleaOthmist

Light cDayNighNigh

Road dStraCurv

Road sMedNo m

Road sDry Wet

Road sGooRou

LocatiNot Junc

TrafcNonSign

Note: Pe

of the asideredfrom tand Roand Infrom roUnit (Mis subj2704 236 8.73024 217 7.23208 257 8.03297 222 6.72982 243 8.13323 259 7.83040 234 7.7

E.N. Aidoo et al. / Accident Analysis and Prevention 53 (2013) 23 27 25

In-depth information on the database and its shortfalls can be foundin Salifu and Ackaah (2011). The data used in this study representthe period 20042010.

3. Empiric

3.1. Descrip

Within twere record7.7%, of the(7.7%) is highas a 1.8% r8.1%, 17.9% gary (Tay etrespectively

Table 1 with respecther describas well as Annually, apedestrian dents varieproportion lowest proptionally, thurban areaapproximathit-and-rundents was hand fog), as out street liproportion good surfachit-and-runsignage on

3.2. Results

To deterhit-and-runtted to thInformationwas used fnal modeindependensignicanceand Hemenprobability nal modellem by perf(deviance =to be non-story and dooverall signusing a likeference betshown in this satisfacto

Table 2 tted moderelative to t

Comparenumber ofhigher in a

Greater Accra and Northern regions compared to all other regions inGhana, at approximately 279% and 100% higher, respectively, thanthat of the Western region. The differences in hit-and-run accidentsamong the regions in Ghana may be due to differences in ethnicity

cio-cand-thout (p-n cearease drban nt scen thto leastaif the-runvolvhe inreferlear utedThe -runimatss a

thed 38ht wuateperc

Highthesety toiver the ent wacLeass mses ah red sigter himated oncreasnt covolumts this con

et ale thatie

ay.ardints onry suugh f theompe drimay g it dtionduceal results and discussion

tive statistics

he study period, a total of 21,578 pedestrian collisionsed, with hit-and-run accidents amounting to 1668, or

total. Ghanas pedestrian hit-and-run accident rateh when compared to countries such as Singapore, whichate (Tay et al., 2008), but it is low when compared withand 18.1% reported for California (Tay et al., 2009), Cal-

al., 2010) and the United States (MacLeaod et al., 2012),.provides summary statistics of hit-and-run incidentst to the independent variables considered. Table 1 fur-es the frequency distribution of pedestrian collisions

the number and percentage of hit-and-run accidents.n average of approximately 238 drivers who struck aed the scene. The table shows that hit-and-run acci-d among the ten regions in Ghana with the highest(11.4%) occurred in the Greater Accra region and theortion (3.8%) occurred in the Western region. Addi-

e proportion of hit-and-run accidents was higher ins than in non-urban areas (8.6% vs. 6.2%). In Ghana,ely 9.6% of fatal pedestrian accidents were caused by

drivers. The incidence of pedestrian hit-and-run acci-igher during unclear weather conditions (such as rainwell as during nighttime hours, especially in areas with-ghts. Compared to roads with other characteristics, theof hit-and-run cases was lower on dry, straight, at ande roads without median separation. The proportion of

accidents was not affected by the presence/absence ofroads.

of logistic regression modeling

mine the possible factors contributing to pedestrian accidents, the binary logistic regression model wase set of available independent variables. The Akaike

Criterion (AIC) with backwards deletion techniqueor the selection of the nal model for the data. Thel selected for the study omitted two non-signicantt variables (i.e., Year and Day of week). Thus, the non-

of Day of week contradicted the studies of Solnickway (1995) and Tay et al. (2009), who found a higherof hit-and-run on weekends than on week days. The

was also evaluated for an over/under dispersion prob-orming a likelihood ratio test on the deviance residuals

11, 009, df = 21, 390). The result of the test was foundignicant at 1%, indicating that the model is satisfac-es not have a problem of over/under dispersion. Theicance (goodness-of-t) of the tted model was testedlihood ratio test. The test examines the signicant dif-ween the null model and the tted model. The results,e last column of Table 2, indicate that the tted modelry.summarizes the maximum likelihood estimates of thel. The coefcients for all of the variables were estimatedhe selected reference category.d to the Western region (the reference group), the

pedestrian hit-and-run incidents was signicantlyll other regions. The incidence was quite high in the

and soof hit-areas, nicanin urbaurban that thnon-uraccide

Whdriver trian suscene ohit-andtions inwhen tthus p

Unccontribroads. hit-andapproxDarkneleaving68% anthe nigInadeqof the 2008).under inabilithe drgiving consistand Mdarknewitnes

Witinclineee afapproxoccurrtion inaccidetrafc prevenresult to Tayvolumbe idenhighw

Regincidenfrom dwas rohood olower cthat thspeed makininstallawill reultural environment. There was also a higher likelihoodrun incidents in urban areas compared to non-urbangh this difference was found to be only marginally sig-value = 0.0546). The incidence of hit-and-run accidentsnters was approximately 13% higher than that of non-. This difference is because there is a higher possibilityiver may be identied by community members in theareas and thus prevented the driver from leaving theene.e effect of accident severity on the likelihood of theve the scene was examined, it was found that the pedes-ning a fatal injury generally led to the driver leaving the

accident. The association between fatal accidents and incidents might be due to the cost and legal implica-ed. The driver is more likely to face severe punishmentjury sustained by the pedestrian is fatal, and s/he may

to leave the scene.weather conditions, such as rain, fog and mist, greatly

to the incidents of hit-and-run accidents on Ghanaianestimated parameters suggested that the likelihood of

as a result of these unclear weather conditions wasely 26% higher than for the clear weather condition.lso appeared to increase the likelihood of the driver

scene of the accident. The effect was approximately% higher during the night without light and duringith a light on, respectively, compared to the daytime.

lighting may encourage hit-and-run behavior becauseeived lower probability of being identied (Tay et al.,er incidents of hit-and-run during unclear weather and

environmental conditions might be due to the drivers see the pedestrian. Additionally, for such conditions,is less likely to be identied by any witness, therebydriver the opportunity to ee the scene. This result isith the ndings of a similar study by Tay et al. (2009)od et al. (2012). According to MacLeaod et al. (2012),akes it easier to protect the drivers identity even whenre available.spect to road description, roads that were curved andnicantly reduced the likelihood that the driver woulditting a pedestrian. The estimated value found wasely 25% lower than the hit-and-run incidents that

straight and at roads. Roads without median separa-ed the likelihood of the driver leaving the scene of thempared to the roads with median separation. Thus, thee of separated roads, which is frequently higher, likely

e driver from eeing from the scene after a accident. Thissistent with the ndings of Tay et al. (2009). According. (2009), divided highways typically carry more trafcn undivided highways and drivers are more likely tod when they are involved in an accident on a divided

g the road surface condition, the number of hit-and-run roads with wet surfaces was not signicantly differentrface roads. In addition, when the road surface repair(i.e., roads with potholes or untarred roads), the likeli-

driver leaving the scene of the accident was signicantlyared to roads with a good surface. This nding suggestsvers ability to leave the scene of the accident at a highbe inuenced by the rough nature of the road surface,ifcult for the driver to leave the scene. In this regard,

of speed humps in settlement areas with tarred roads the speed of vehicles and likely prevent drivers from

26 E.N. Aidoo et al. / Accident Analysis and Prevention 53 (2013) 23 27

Table 2Parameter estimates for logistic regression model.

Variable Coefcient Standard error p-Value Odds ratio 95% CI (OR)

Intercept 3.279 0.165

E.N. Aidoo et al. / Accident Analysis and Prevention 53 (2013) 23 27 27

Although the study was limited with respect to the effect of roadand environmental characteristics on pedestrian hit-and-run acci-dents, we believe that both drivers and pedestrian characteristicsare likely to inuence the drivers decision to leave the scene aftera pedestrian accident. For instance, the value of the pseudo R2 alsosuggests the inclusion of other variables to the tted model. Hence,it is therefore recommended that future research relating to pedes-trian hit-and-run accidents should consider these characteristics toimprove policies regarding road safety.

Acknowledgements

We are grateful to Mr. Francis K. Afukaar (Chief Research Sci-entist) from the Building and Road Research Institute, Prof. LouisMunyakazi from Kumasi Polytechnic, Dr. Moudud Md Alam fromDalarna University and Prof. Johan Lyhagen from Uppsala Univer-sity for their useful comments and suggestions. Any remainingerrors are our own responsibility. We are also thankful to the staffof the Building and Road Research Institute.

References

Ackaah, W., Adonteng, D.O., 2011. Analysis of fatal road trafc crashes in Ghana.International Journal of Injury Control and Safety Promotion 18, 2127.

Afukaar, F.K., Agyemang, W., Ackaah, W., Larbi, J., 2011. Road Trafc Crashes inGhana: Statistics 2010. Consultancy Service Report for National Road SafetyCommission of Ghana.

Agresti, A., 2007. An Introduction to Categorical Data Analysis, 2nd ed. John Willeyand Sons Inc., Hoboken, NJ.

Damsere-Derry, J., Ebel, B.E., Mock, C.N., Afukaar, F., Donkor, P., 2010. Pedes-trians injury patterns in Ghana. Accident Analysis and Prevention 42, 10801088.

Hamed, M.M., 2000. Analysis of pedestrians behavior at pedestrian crossings. SafetyScience 38, 6382.

Johnson, C., 1997. Pedestrian fatalities on interstate highways: characteristics andcountermeasures. Transport Research Record 1578, 2329.

MacLeaod, K.E., Griswold, J.B., Lindsay, S.A., Ragland, A.D., 2012. Factors associatedwith hit-and-run pedestrian fatalities and driver identication. Accident Anal-ysis and Prevention 45, 366372.

Mock, C., Nii-Amon-Kotei, D., Maier, R.V., 1997. Low utilization of formal medicalservices by injured persons in a developing nation: health service data under-estimate the importance of trauma. Journal of Trauma 42, 504513.

Nelder, J.A., Wedderburn, R.W.M., 1972. Generalized linear models. Journal of RoyalStatistical Society A 140, 4877.

Peden, M., Scureld, R., Sleet, D., Mohan, D., Hyder, A.A., Mathers, C., 2004. WorldReport on Road Trafc Injury Prevention. World Health Organization, Geneva.

Rosenbloom, T., Nemrodov, D., Barkan, H., 2004. Crossing at a red light: behaviourof individuals and groups. Transportation Research Part F 7, 395404.

Salifu, M., Ackaah, W., 2011. Under-reporting of road trafc crash datain Ghana. International Journal of Injury Control and Safety Promotion,http://dx.doi.org/10.1080/17457300.2011.628752.

Solnick, S.J., Hemenway, D., 1994. Hit the bottle and run: the role of alcoholin hit-and-run pedestrian fatalities. Journal of Studies on Alcohol 55 (6),679684.

Solnick, S.J., Hemenway, D., 1995. The hit-and-run in fatal pedestrian accidents: vic-tims, circumstances and drivers. Accident Analysis and Prevention 27, 643649.

Sullman, M.J.M., Gras, M.E., Font-Mayolas, S., Masferrer, L., Cunill, M., Planes, M.,2011. The pedestrian behaviour of Spanish adolescents. Journal of Adolescence34, 531539.

Tay, R., Barua, U., Kattan, L., 2009. Factors contributing to hit-and-run in fatal crashes.Accident Analysis and Prevention 41, 227233.

Tay, R., Kattan, L., Sun, H., 2010. Logistic model of hit and run crashesin Calgary. Canadian Journal of Transportation, North America. Avail-able from: http://journals.hil.unb.ca/index.php/CJT/article/view/12412/17172(accessed 03.05.12).

Tay, R., Rifaat, S.M., Chin, H.C., 2008. A logistic model of the effects of roadway,environmental, vehicle, crash and driver characteristics on hit-and-run crashes.Accident Analysis and Prevention 40, 13301336.

Further reading

Ackaah, W., Salifu, M., 2011. Crash prediction model for two-lane rural highways inthe Ashanti Region of Ghana. Journal of International Association of Trafc andSafety Sciences 35, 3440.

The effect of road and environmental characteristics on pedestrian hit-and-run accidents in Ghana1 Introduction2 Methodology3 Empirical results and discussion3.1 Descriptive statistics3.2 Results of logistic regression modeling

4 ConclusionAcknowledgementsReferences

Further reading