COMPARATIVE STUDY OF REGIONAL CRASH DATA IN TURKEY.
Even though there has been significant public policy attention and improvements in traffic safety policies and practices in Turkey, 61 people died per billion vehicle-km in traffic crashes in 2016 (TGDH, 2017; TurkStat, 2018a). In spite of significant improvements in national highway network, there has been an increase in fatal and/or injury (FI) crashes over the last decade (TurkStat, 2018a). The distribution of crashes across the nation is also of importance to transportation system owners. National and local safety programs aim to reduce crashes and the severity of their outcomes within their jurisdictions. Development of geographically appropriate safety strategies requires estimating pertinent crash and exposure data at the relevant spatial scale. While data required to identify safety risks are collected at the local level, published databases are typically available only at larger scales. Thus, there is a need to deduce data at the local level (i.e., lower levels of spatial aggregation) from partially complete or surrogate datasets that are available at a higher level of aggregation.
FI crashes are reported by the traffic police and gendarmerie units according to their areas of responsibility in Turkey. Disaggregate statistics of these crashes are published annually by Turkish Statistical Institute (TurkStat). This aggregate database provides temporal and provincial distribution of the crashes as well as type of vehicles involved, classification of the crash locations as well as gender and age distribution of the crash victims. Due to the lack of disaggregate crash level data at the national level, province and regional variations of traffic safety have not been examined in detail. Recently, Atalay and Tortum (2015) compared the number of fatalities per traffic crashes and per kilometer of road network across the 81 provinces of Turkey. The results showed that number of fatalities per crash are higher in less developed provinces, whereas number of fatalities per length of road network are higher in developed provinces. In other study, Erdogan (2009) studied the provincial level differences in number of FI crashes and number of fatalities. Population and number of registered vehicles were used to quantify safety and results indicated that provinces with higher FI crashes and fatalities were located in the provinces that contain the roads connecting the Istanbul, Ankara, and Antalya provinces. However, there is no study focusing on traffic safety at the regional level in Turkey.
This study provides a comparative analysis of the FI crashes across the seven geographic regions in Turkey from 2006 to 2016 (additional information is provided in Appendix A). The comparisons are performed in relative terms and absolute terms. Since vehicle-km data are not available either province or regional level, number of FI crashes per million population and per million registered vehicles are used to quantify safety. The principal sources of data used in this study is TurkStat.
Number of FI crashes per million population and per million registered vehicles were determined for each geographic region annually for the study period. A Geographic Information Systems based thematic maps were used to support these efforts.
Traditional statistical tests based on the normality assumption of the data. Since FI crash rates do not follow normal distribution either across the regions or over the years, nonparametric methods need to be used to study FI crash rates. An appropriate test to use for this purpose is the Kruskal-Wallis nonparametric test. In this study, hypotheses of the Kruskall-Wallis H test was that:
HO: FI crash rates are the same for each region from 2006 to 2016
H1: FI crash rates are not the same for each region from 2006 to 2016.
Based on the Kruskall-Wallis test, the null hypothesis, Ho, is to be rejected at the (100-[alpha]) percent level of confidence if the test statistic, H, falls in the critical region H > [[chi].sub.a.sup.2] with v = (k-1) degrees of freedom. To control the familywise type I error in Kruskall-Wallis H test; the probability of rejecting at least one pair hypothesis given all pairwise hypotheses are true, adjusted p-values are calculated and used to make the decision for each pair. The following equations was used to calculate adjusted p-values for each of pairwise hypothesis. If the adjusted p-value is bigger than 1, it is set to 1.
[p.sub.adj] = pK (K - 1)/2 (1)
where; K = number of pairwise hypothesis, and p = significance level of pairwise hypothesis.
FI crash rates were calculated annually for each geographic region based on per million population and per million registered vehicles. The results are presented thematically in Tables B1 to B2 (see Appendix). It is noted that the numbers of the regions are given randomly. In these tables, a graded color pattern is used to indicate FI crash rates. The color gradation ranges from red to yellow or green. Dark red is used to indicate the higher FI crash rates and worse safety records, and dark green is used to indicate lower FI crash rates and best safety records. Lighter red, yellow and lighter green colors are used to achieve gradation.
Table B1 presents FI crash rates of each region per million population for each year during the study period. Table B2 presents FI crash rates of each region per million registered vehicles for each year during the study period. In addition, the average FI crash rates for each measure for the entire 11 year period as a whole are given in these tables. It is seen that FI crash rates for regions significantly increased for each measure from 2006 to 2016. Furthermore, Table B1 and B2 clearly indicate the stability of the relative FI crash rates of regions across the years. They show that regions that tended to have lower FI crash rates, had lower crash rates across the years; and, regions that tended to have higher FI crash rates, had higher crash rates across the years.
Kruskall-Wallis pairwise comparisons implied that FI crash rates per million population are not the same across the regions from 2006 to 2016 (i.e. H = 31.50 > [[chi].sub.0.05,9.sup.2]= 12.59). Fig. 1 and 2 present box plot and 95% confidence interval of FI crash rates of regions per million population. It is seen that FI crash rates in Central Anatolia Region (Region 5), Mediterranean Region (Region 4) and Aegean Region (Region 2) seems relatively higher than the others. FI crash rates in Southeastern Anatolia Region (Region 3) and Eastern Anatolia Region (Region 6) seems relatively lower than the others.
Fig. 3 presents graphical Kruskal-Wallis multiple pairwise comparisons. The number below each region represents the average rank of regional FI crash rates over the 11 years period. Fig. 4 provides Kruskal-Wallis tests results for significant pairwise comparisons. However, most of them are not significant based on adjusted p-value (see Fig. 3). In Fig. 3, yellow lines represent the significant pairwise comparisons based on adjusted p-values. FI crash rates per million population for Central Anatolia Region (Region 5) and Aegean Region (Region 2) are significantly higher than Southeastern Anatolia Region (Region 3) and Eastern Anatolia Region (Region 6); for Mediterranean Region (Region 4) is significantly higher than Southeastern Anatolia Region (Region 3).
Kruskall-Wallis pairwise comparisons implied that FI crash rates per million population are not the same across the regions from 2006 to 2016 (i.e. H = 44.98 > [chi.sub.0.05,9.sup.2] = 16.92). Fig. 5 and 6 present box plot and 95% confidence interval of FI crash rates of regions per million population. It is seen that FI crash rates in Eastern Anatolia Region (Region 6) seems relatively higher than the others. FI crash rates in Marmara Region (Region 1) seems relatively lower than the others. Fig. 7 presents graphical Kruskal-Wallis multiple pairwise comparisons. Furthermore, Fig. 8 provides Kruskal-Wallis tests results for significant pairwise comparisons. However, most of them are not significant based on adjusted p-value. FI crash rates per million registered vehicles for Eastern Anatolia Region (Region 6) are significantly higher than Marmara Region (Region 1), Eagan Region (Region 2) and Mediterranean Region (Region 4). In addition, FI crash rates per million registered vehicles for Marmara Region (Region 1) are significantly lower than Southeastern Anatolia Region (Region 3), Central Anatolia Region (Region 5) and Black Sea Region (Region 7).
Fig. 9 and 10 provide thematic maps based on the average ranks of the provinces for each of the safety measures used in this study. In these maps, the red colored provinces have the highest rates while the green colored provinces have the lowest rates. An examination for Fig. 1 to 2 reveal some interesting patterns in the spatial distribution of the relative safety ranks of the regions. Overall, it can be seen that Marmara Region (Region 1) tend to have best safety records. Relative safety records of Aegean Region (Region 2), Eastern Anatolia Region (Region 6) and Southeastern Anatolia Region (Region 3) are significantly different for million population and million registered vehicles measures. For instance, Eastern Anatolia Region (Region 6) has the best safety records for FI crash rates per million population, however, it has the worst safety records for FI crash rates per million registered vehicles.
This paper summarized efforts of and findings from a study to examine regional level FI crash trends and perform comparative analyses of safety records 2006 to 2016. The comparisons were performed in relative terms (ordinal scale or based on rates) and absolute terms (cardinal or rank ordered scale). Two safety measures were used to evaluate safety: million population and million registered vehicles. Data were obtained from publications maintained by TurkStat.
An examination of the results indicated that the relative ranks of the regions were stable over the study period for each safety measure. Non-parametric statistical tests and thematic maps used to support comparative analyses. Specifically, the Kruskal-Wallis nonparametric test was used in this study. The results showed that the FI crash rates are not the same across the regions. Furthermore, the analyses also revealed that depending on the safety measure used, the relative rankings of regions varied (i.e., a region ranked at the top (high crash rate) for one safety measure does not need to be ranked again at the top for other safety measure). This figure is resulted from significantly different vehicle ownership rate across the regions in Turkey. For the cardinal analysis the computed rates were used. These results were consistent with those from the ordinal analysis, but it was showed that FI crash rates significantly increased over the time.
For broad macro level analyses a more representative vehicle-km measure is required to study relative safety records of regions. However, it is available only for national level in Turkey. Furthermore, if specific analyses are required, then safety measures should be defined based on the desired evaluations. For example, if the goal were to address rural safety, the measures should be computed using rural fatal and/or injury crashes, rural vehicle-km, and the extent of rural kilometers of road network. This paper explored methods to analyze regional differences in road traffic safety. The results document the validity and promise of the methods. These methods could be expanded for policy and operational analyses.
Erdogan, S. (2009). "Explorative spatial analysis of traffic accident statistics and road mortality among the provinces of Turkey". Journal of safety research, Vol. 40, No. 5, 341-351.
OECD/ITF. (2016). Organisation for Economic Cooperation and Development/International Transport Forum. Road Safety Annual Report 2016, Paris. OECD Publishing, 2016.
TGDH, Turkish General Directorate of Highways. (2017). Karayollari Uzerindeki Seyir ve Tasimalar. Turkey. http://www.kgm.gov.tr/SiteCollectionDocuments/KGMdocuments/Istatistikler/SeyirveTasimalar/SeyirVeTasimalar.pdf [Accessed 10 Dec 2017].
Tortum, A. and Atalay, A. (2015). "Spatial analysis of road mortality rates in Turkey." In Proceedings of the Institution of Civil Engineers-Transport, Vol. 168, No. 6, pp. 532-542.
TurkStat, Turkish Statistical Institute. (2018a). Road Traffic Accident Statistics, 2015. http://www.turkstat.gov.tr/PreHaberBultenleri.do?id=21611 [Accessed 10 Dec 2017].
TurkStat, Turkish Statistical Institute. (2018b). Road Traffic Accident Statistics, 2016. http://www.turkstat.gov.tr/PreHaberBultenleri.do?id=24606 [Accessed 10 Dec 2017].
Appendix A: Geographic Regions in Turkey
It is noted that the numbers of the regions are given randomly.
* Marmara Region (Region 1): Balikesir, Bilecik, Bursa, Canakkale, Edirne, Istanbul, Kirklareli, Kocaeli, Sakarya, Tekirdag, Yalova.
* Aegean Region (Region 2): Afyon, Aydin, Denizli, Izmir, Kutahya, Manisa, Mugla, Usak.
* Southeastern Anatolia Region (Region 3): Adiyaman, Batman, Diyarbakir, Gaziantep, Kilis, Mardin, Siirt, Urfa, Sirnak.
* Mediterranean Region (Region 4): Adana, Antalya, Burdur, Hatay, Isparta, Kahramanmaras, Mersin, Osmaniye.
* Central Anatolia Region (Region 5): Aksaray, Ankara, Cankiri, Eskisehir, Karaman, Kayseri, Kirikkale, Kirsehir, Konya, Nevsehir, Nigde, Sivas, Yozgat.
* Eastern Anatolia Region (Region 6): Agri, Ardahan, Bingol, Bitlis, Elazig, Erzincan, Erzurum, Hakkari, Igdir, Kars, Malatya, Mus, Tunceli, Van.
* Black Sea Region (Region 7): Amasya, Artvin, Bartin, Bayburt, Bolu, Corum, Duzce, Giresun, Gumushane, Karabuk, Kastamonu, Ordu, Rize, Samsun, Sinop, Tokat, Trabzon, Zonguldak
Murat Ozen (*1)
Mersin University, Engineering Faculty, Department of Civil Engineering, Mersin, Turkey ORCID ID 0000-0002-1745-7483
(*) Corresponding Author
Received: 28/01/2018 Accepted: 10/04/2018
Sample1 Test Std. Std. Test Sig. Adj.Sig. -Sample2 Statistic Error Statistic R3-R5 -37.000 9.539 -3.879 .000 .002 R3-R2 34.273 9.539 3.593 .000 .007 R6-R5 33.000 9.539 3.459 .001 .011 R3-R4 -32.091 9.539 -3.364 .001 .016 116-R2 30.273 9.539 3.173 .002 .032 R1-R5 -28.909 9.539 -3.030 .002 .051 R6-R4 28.091 9.539 2.945 .003 .068 R1-R2 -26.182 9.539 -2.745 .006 .127 R1-R4 -24.000 9.539 -2.516 .012 .249 R3-117 -19.455 9.539 -2.039 .041 .870 117-R2 14.818 9.539 1.553 .120 1.000 R6-R1 4.091 9.539 .429 .120 1.000 R7-R4 12.636 9.539 1.325 .669 1.000 R4-R2 2.182 9.539 .229 .185 1.000 R7-R5 17.545 9.539 1.839 .066 1.000 R3-R1 8.091 9.539 .848 .396 1.000 R4-R5 -4.909 9.539 -.515 .607 1.000 R6-R7 -15.455 9.539 -1.620 .105 1.000 R2-R5 -2.727 9.539 -.286 .775 1.000 113-116 -4.000 9.539 -.419 .675 1.000 R1-R7 -11.364 9.539 -1.191 .234 1.000 Fig. 4. Kruskal-Wallis multiple pairwise comparisons of FI crash rates per million population Sample1 Test Std. Std. Test Sig. Adj.Sig. -Sample2 Statistic Error Statistic R1-R3 -41.455 9.539 -4.346 .000 .000 R2-R6 -41.727 9.539 -4.374 .000 .000 R1-R6 -57.727 9.539 -6.051 .000 .000 R4-R6 -34.182 9.539 -3.583 .000 .007 R1-R7 -33.182 9.539 -3.478 .001 .011 R1-R5 -31.727 9.539 -3.326 .001 .019 R5-R6 -26.000 9.539 -2.726 .006 .135 R2-R3 -25.455 9.539 -2.668 .008 .160 R7-R6 24.545 9.539 2.573 .010 .212 R1-R4 -23.545 9.539 -2.468 .014 .285 R7-R3 8.273 9.539 .867 .386 1.000 R5-R3 9.727 9.539 1.020 .308 1.000 R4-R3 17.909 9.539 1.877 .060 1.000 R1-R2 -16.000 9.539 -1.677 .093 1.000 R4-R5 -8.182 9.539 -.858 .391 1.000 R5-R7 -1.455 9.539 -.152 .879 1.000 R2-R4 -7.545 9.539 -.791 .429 1.000 R3-R6 -16.273 9.539 -1.706 .088 1.000 R2-R5 -15.727 9.539 -1.649 .099 1.000 R4-R7 -9.636 9.539 -1.010 .312 1.000 R2-R7 -17.182 9.539 -1.801 .072 1.000 Fig. 8. Kruskal-Wallis multiple pairwise comparisons of FI crash rates per million population Appendix B: FI Crash Rates Table B1. FI crash rates for regions per million population Region 2006 2007 2008 2009 2010 2011 2012 2013 2014 Marmara (R1) 930 1061 1051 1087 1091 1233 1376 1557 1621 Aegean (R2) 1479 1564 1489 1544 1597 1826 2197 2766 2961 Southeastern 707 717 706 794 834 936 1153 1538 1564 Anatolia (R3) Mediterranean 1388 1481 1432 1522 1608 1848 2164 2749 2796 (R4) Central 1565 1678 1558 1720 1803 2000 2371 2675 2621 Anatolia (R5) Eastern 699 749 726 808 942 1017 1159 1558 1688 Anatolia (R6) Black Sea (R7) 1065 1181 1093 1207 1304 1433 1608 2192 2304 Region 2015 2016 Average 2006-2016 Marmara (R1) 1707 1718 1312 Aegean (R2) 3183 3150 2160 Southeastern 1718 1524 1108 Anatolia (R3) Mediterranean 3061 3020 2097 (R4) Central 2745 2786 2138 Anatolia (R5) Eastern 1815 1797 1178 Anatolia (R6) Black Sea (R7) 2529 2702 1692 Table B2. FI crash rates for regions per million registered vehicle Region 2006 2007 2008 2009 2010 2011 2012 2013 Marmara (R1) 4927 5378 5141 5265 5191 5669 6098 6666 Aegean (R2) 6276 6278 5725 5846 5852 6305 7263 8859 Southeastern 8163 7839 7329 7771 7616 8077 9414 12098 Anatolia (R3) Mediterranean 6626 6673 6130 6346 6469 6997 7802 9571 (R4) Central 7636 7826 6986 7502 7551 7943 8989 9754 Anatolia (R5) Eastern 10189 10294 9449 9871 10621 10811 11614 14783 Anatolia (R6) Black Sea (R7) 7384 7710 6788 7189 7313 7483 7935 10243 Region 2014 2015 2016 Average 2006-2016 Marmara (R1) 6715 6731 6456 5840 Aegean (R2) 9175 9428 8979 7271 Southeastern 12058 12916 11335 9511 Anatolia (R3) Mediterranean 9403 9867 9459 7758 (R4) Central 9214 9202 8963 8324 Anatolia (R5) Eastern 15399 15784 14859 12152 Anatolia (R6) Black Sea (R7) 10255 10651 10920 8534
|Printer friendly Cite/link Email Feedback|
|Publication:||Turkish Journal of Engineering (TUJE)|
|Date:||Sep 1, 2018|
|Previous Article:||INVESTIGATION OF DISSOLUTION KINETICS OF Zn AND Mn FROM SPENT ZINC-CARBON BATTERIES IN SULPHURIC ACID SOLUTION.|
|Next Article:||PERFORMANCE COMPARISON OF ANFIS, ANN, SVR, CART AND MLR TECHNIQUES FOR GEOMETRY OPTIMIZATION OF CARBON NANOTUBES USING CASTEP.|