Abstract
Background:
The control, management, and prevention of driving accidents and risky driving are regarded as concerns for numerous countries, according to the World Health Organization. In this regard, many technologies, such as count stations, are recommended. They count traffic offenses, such as speeding and unsafe distance, hourly and daily, and have different patterns according to the hour of the day and the location.Objectives:
This study aimed to investigate the risky driving behaviors according to traffic offenses in Iran and estimate their hourly and spatial patterns using generalized additive models (GAMs) and stochastic partial differential equation methods.Materials and Methods:
There were 2,316 count data stations for one month within MarchApril 2019. This study estimated the hourly average of each traffic offense, Pearson’s and Spearman’s correlations, and the energy statistics for testing the bivariate normal distribution. There are five distributions, such as univariate Poisson, quasilikelihood Poisson, Gaussian, locationscale Gaussian, and bivariate Gaussian in GAMs, to study the hourly patterns which were compared to the mean squared error (MSE) and correlation.Results:
The hourly average of total vehicles and number of speeding and unsafe distance offenses per count station had positive skew distributions with mean values equal to 347 ± 456, 22.5 ± 44.2, and 65.9 ± 150, respectively. The correlation between traffic offenses in most provinces was significant, not large, and different. The GAM with the bivariate Gaussian distribution had the best performance according to the MSE and correlation. It revealed three hourly patterns for count predictions; the first was that speeding is higher than unsafe distances; the second was that unsafe distances are higher than speeding; the third was that speeding and unsafe distances do not have a specific pattern in some hours. The percentage of speeding was higher in the central, northeast, and southeast regions than in other parts of Iran, and the percentage of unsafe distances was higher for the north, northwest, west, and some parts of the southwest than in other parts of Iran, respectively.Conclusions:
The hourly pattern of traffic offenses exists and has a complex structure. The spatial pattern of traffic offenses shows the riskiest points in Iran.Keywords
Stochastic Processes Generalized Additive Model Traffic Aggressive Driving Iran
1. Background
The traffic and speed cameras and count stations near the roads are some technologies developed to manage, control, and predict traffic status in different countries. These devices produce massive datasets hourly and daily, making them one of the primary resources for discovering patterns and relationships. For example, there are diverse indices, including the count of driving offenses and total vehicles based on their type, which could be considered (1, 2). According to the global status report on road safety by the World Health Organization, driving accidents and risky driving are among the remarkable causes of death worldwide (2).
This study investigated the hourly and spatial patterns of driving offenses using two advanced statistical methods, respectively. Firstly, the generalized additive model (GAM) extended the generalized linear model (GLM) idea by adding the smoothing functions, such as cubic regression splines for estimating, hypothesis testing, and confidence interval for coefficients. The GAM was enhanced to capture complex and nonlinear relationships between response variables and covariates (35). Secondly, the stochastic partial differential equation (SPDE) method is one of the techniques to study spatial variability with the integrated nested Laplace approximation (RINLA) (6).
2. Objectives
The traffic offenses dataset from count stations is a source for studying driving behaviors in Iran. The speeding and unsafedistance indices are the indicators of risky driving for the control and prevention for which governments worldwide have different laws (2). Firstly, this study estimated the hourly average of each traffic offense per count station nationalwide and provincewise. It shows how many traffic offenses occurred on average in each count station hourly. Secondly, the correlations and bivariate distribution between speeding and unsafe distance provincewise offenses were evaluated by Pearson’s and Spearman’s correlation coefficients and bivariate normal test based on the energy statistics, respectively. Thirdly, this study modeled the nonlinear relationship between the number of traffic offenses and the time of a day for each province with GAMs separately. Finally, this study introduced the percentage of traffic offenses among all transported vehicles as a key index to study the risky behavior of drivers on roads and investigated their spatial variability with SPDE. Moreover, temporal variability for four different ranges of hours was evaluated.
3. Materials and Methods
3.1. Population and Dataset
The population of this study was all count data stations near the interprovincial roads of Iran, which are available online from Iran Road Maintenance and Transportation Organization on the website of the Ministry of Roads and Urban Development (rmto.ir). All days of the first Iranian month, Farvardin, were considered in this analysis (31 days within 3/21/2019 to 4/20/2019). The number of individuals traveling between provinces increases in the first and second weeks of this month due to the Iranian New Year holidays. The condition of the roads gets back to regularly slow during the third and fourth weeks of this month. Therefore, the risky driving behavior of most Iranians could be estimated in this month.
The count data stations record hourly and daily different indices, such as total vehicles and the number of speeding and unsafe distance offenses. There were 31 separated provincial datasets. The active count stations record 60 minutes an hour, and only 2,316 count data stations remain. The count of unsafe distance offenses is the total number of vehicles with a distance shorter than 2 seconds between them. The count of speeding is the total number of vehicles on the road with a speed higher than the speed control limit (rmto.ir).
3.2. Statistical Analysis
3.2.1. Testing the Multivariate Normal Distribution
In order to study the bivariate distribution of these two traffic offenses, this study chose the energy test statistics that have the best performance among multivariate Gaussian distribution tests (7, 8). The observation Z ∈ R^{d} has a multivariate Gaussian distribution N_{d} (0, I_{d}) with the mean vector of 0 and the variancecovariance matrix I_{d} (9) as follows:
The y_{1},…, y_{n} are the standardized elements of the sample. The energy test statistics for the ddistribution standard normal are as follows:
And
For computation, the sample of observation is standardized based on the mean and correlation matrix. The obtained test statistic
3.2.2. Correlation Tests
The correlation between the two random variables Y_{1} and Y_{2} or ρ_{12} was assessed with the maximum likelihood estimation of the Pearson productmoment correlation coefficient if their distribution was bivariate normal. When the joint distribution of the two random variables was not multivariate normal, the correlation was evaluated utilizing Spearman’s Rank correlation coefficient. In this test, the ranks of Y_{1} and Y_{2} are R_{1} and R_{2} for calculation, respectively (10).
3.2.3. Generalized Additive Model
The GAM is a GLM with a set of smoothing functions of covariates. The general form is as follows:
where μ_{i} ≡ E (Y_{i}) and Y_{i} ~ EF (μ_{i}, ϕ). The response term is Y_{i} with the exponential family, a mean of μ_{i}, and the scale parameters of ϕ. The A_{i} is related to the parametric part of the model, and θ represents a vector related to the parameters. The f_{i} is the smoothing function of the covariate x_{k} (4). This study considered and compared the five distributions of univariate Poisson, univariate quasiPoisson, univariate Gaussian, univariate locationscale Gaussian, and bivariate Gaussian for Y_{i}. There are two models for each distribution, with the response variable in models 1 and 2 being the number of speeding and unsafe distance offenses, respectively. The only covariate in the model is the hour of the day, and the smoothing function f_{1}(x) is a cubic regression spline with k = 20. The mean squared error (MSE) and correlation between observations and fitted values are the criteria for choosing the best model (4, 5).
3.2.4. Spatial Data Analysis
The response variable was the percentage of total speeding and unsafe distances in the total traffic during one month for each count station, respectively. The count station location has some information that accounts for this model. The Y_{i} refers to the percentage of each traffic offense at count station s_{i} and has a Gaussian distribution as follows:
where Z(.) indicates a spatially structured random effect with zeromean Gaussian process and Matérn covariance. The fitting method was the SPDE approach with RINLA. The mesh was constructed with different margins to approximate Y_{i} to a discrete Gaussian Markov random field. The models predict the lower and upper limits with 95% credible intervals, and the result is a map plot (6). All the statistical analyses were performed using R 4.0.2 (rproject.org) with mgcv, energy, geoR, maps, rgdal, maptools, sf, viridis, and ggplot2 packages (1118).
4. Results
Table 1 shows the results of descriptive statistics and pvalues for the bivariate normal test based on the energy statistics and correlation coefficients. Diverse factors were associated with driving offenses, among which only the time of day as a spline function was considered in this study. Table 2 shows the results of fitting 1) univariate GAM with quasiPoisson distribution and 2) bivariate Gaussian GAM (other models, including univariate Poisson, univariate Gaussian, and univariate locationscale Gaussian, in Appendix 1). These two models had the best performance among their Poissonrelated and Gaussianrelated families according to the MSE and correlation indices.
Provinces  Number of Count Stations  Hourly Average per Station (SD)  PValue of Bivariate Normal Test ^{b}  Correlation  

Speeding  Unsafe Distance  Pearson  Spearman  
Ardebil  64  12.2 (12.3)  33.3 (29.2)  < 0.05  0.16 (< 0.05)  0.29 (< 0.05) 
Isfahan  120  26.5 (35.4)  30.3 (50.2)  > 0.05  0.26 (< 0.05)  0.50 (< 0.05) 
Alborz  40  99.4 (197.6)  320.8 (555.8)  < 0.05  0.32 (< 0.05)  0.49 (< 0.05) 
Ilam  48  5.5 (5.7)  14.5 (17.4)  > 0.05  0.21 (< 0.05)  0.46 (< 0.05) 
West Azarbayjan  90  17.9 (19.2)  40 (28.6)  < 0.05  0.25 (< 0.05)  0.46 (< 0.05) 
East Azarbayjan  82  15.4 (23.5)  18.2 (63.6)  < 0.05  0.01 (0.11)  0.08 (< 0.05) 
Bushehr  52  16.9 (14.8)  53.5 (68)  < 0.05  0.26 (< 0.05)  0.30 (< 0.05) 
Tehran  103  49.5 (79.3)  259.6 (334.7)  < 0.05  0.27 (< 0.05)  0.41 (< 0.05) 
Chaharmahal and Bakhtiari  62  23.7 (24.8)  26 (21.3)  > 0.05  0.46 (< 0.05)  0.49 (< 0.05) 
South Khorasan  72  20.5 (16.4)  13.9 (12.7)  > 0.05  0.51 (< 0.05)  0.61 (< 0.05) 
Razavi Khorasan  131  19.6 (31.5)  70 (133.9)  < 0.05  0.30 (< 0.05)  0.48 (< 0.05) 
North Khorasan  47  12.6 (17.4)  32.5 (27.5)  > 0.05  0.33 (< 0.05)  0.26 (< 0.05) 
Khuzestan  104  14.5 (17.8)  34.7 (31.2)  > 0.05  0.16 (< 0.05)  0.41 (< 0.05) 
Zanjan  74  7.3 (8.7)  40.5 (52.5)  > 0.05  0.38 (< 0.05)  0.48 (< 0.05) 
Semnan  54  43.9 (49.3)  33.1 (36.2)  > 0.05  0.58 (< 0.05)  0.69 (< 0.05) 
Sistan and Balouchestan  70  26.2 (24)  1.2 (1.3)  > 0.05  0.50 (< 0.05)  0.44 (< 0.05) 
Fars  126  33.7 (46.1)  74 (110.9)  > 0.05  0.21 (< 0.05)  0.35 (< 0.05) 
Qazvin  57  14.8 (25.1)  171.5 (268.2)  > 0.05  0.25 (< 0.05)  0.50 (< 0.05) 
Qom  76  49 (74)  82.4 (124.3)  > 0.05  0.16 (< 0.05)  0.20 (< 0.05) 
Kordestan  52  7.6 (9.9)  26.9 (28)  < 0.05  0.30 (< 0.05)  0.37 (< 0.05) 
Kerman  107  13.4 (13.4)  19.6 (17.2)  < 0.05  0.27 (< 0.05)  0.33 (< 0.05) 
Kermanshah  72  9.8 (12.1)  42.2 (47.9)  < 0.05  0.32 (< 0.05)  0.52 (< 0.05) 
Kohkilouye and Boyerahmad  28  7.1 (8.5)  24.6 (19.7)  < 0.05  0.34 (< 0.05)  0.50 (< 0.05) 
Golestan  52  13.1 (24.6)  78.5 (109.9)  < 0.05  0.00 (0.65)  0.29 (< 0.05) 
Gilan  84  9.8 (15)  193 (179.2)  < 0.05  0.09 (< 0.05)  0.30 (< 0.05) 
Lorestan  62  11.7 (12.5)  28.7 (31.8)  < 0.05  0.38 (< 0.05)  0.49 (< 0.05) 
Mazandaran  102  23.7 (34.7)  111.6 (164)  > 0.05  0.13 (< 0.05)  0.32 (< 0.05) 
Markazi  64  30.6 (29.7)  56.7 (64.2)  < 0.05  0.36 (< 0.05)  0.47 (< 0.05) 
Hormozgan  72  27.5 (31.7)  37.4 (37.9)  < 0.05  0.49 (< 0.05)  0.48 (< 0.05) 
Hamedan  71  15.3 (20.3)  44.4 (74.9)  > 0.05  0.47 (< 0.05)  0.69 (< 0.05) 
Yazd  78  19.6 (32.6)  19.8 (40.8)  < 0.05  0.58 (< 0.05)  0.53 (< 0.05) 
Provinces  Univariate QuasiPoisson  Bivariate Gaussian  

MSE  COR  MSE  COR  
Speeding  Unsafe Distance  Speeding  Unsafe Distance  Speeding  Unsafe Distance  Speeding  Unsafe Distance  
Ardebil  431.58  5585.6  0.26  0.37  313.039  3886.618  0.2655  0.4164 
Isfahan  4234.72  9435.8  0.25  0.24  3305.632  7955.29  0.2588  0.2578 
Alborz  100001.65  741558.34  0.11  0.22  86774.854  576048.427  0.1071  0.2286 
Ilam  115.86  2336.48  0.28  0.24  84.528  1887.848  0.2714  0.253 
West Azarbayjan  1083.11  2514.08  0.23  0.3  1544.517  12434.603  0.2564  0.2809 
East Azarbayjan  1943.81  15195.81  0.25  0.26  839.992  1988.403  0.2496  0.3423 
Bushehr  779.4  14589.36  0.22  0.29  540.01  10785.777  0.2256  0.3195 
Tehran  15866.28  354773.14  0.17  0.3  12766.231  242816.189  0.1775  0.3289 
Chaharmahal and Bakhtiari  1664.66  2716.23  0.3  0.41  1105.088  1737.031  0.3112  0.4543 
South Khorasan  1319.27  1133.26  0.33  0.32  834.305  873.05  0.3386  0.3355 
Razavi Khorasan  2241.12  50497.25  0.18  0.22  1682.416  40515.689  0.1797  0.2487 
North Khorasan  936.37  5419.72  0.16  0.38  767.099  3592.478  0.1647  0.4066 
Khuzestan  978.81  6959.98  0.2  0.33  737.537  4672.25  0.1983  0.3649 
Zanjan  229.03  19909.58  0.18  0.24  193.617  17284.791  0.1796  0.2557 
Semnan  10017.67  11546.95  0.25  0.24  7481.894  9765.432  0.2607  0.2609 
Sistan and Balouchestan  2478.86  12.47  0.34  0.3  1363.401  10.218  0.3528  0.3227 
Fars  5385.19  39761.19  0.21  0.27  4162.456  31543.184  0.2155  0.2845 
Qazvin  2263.28  360868.29  0.14  0.24  1600.491  244795.473  0.1557  0.218 
Qom  16924.48  61534.78  0.23  0.27  13419.011  49202.503  0.2369  0.2833 
Kordestan  235.64  3365.31  0.17  0.38  197.928  2299.921  0.1735  0.4272 
Kerman  659.95  2465.71  0.26  0.34  419.746  1635.215  0.2572  0.3666 
Kermanshah  744.38  11365.21  0.17  0.32  665.794  8546.778  0.1719  0.3573 
Kohkilouye and Boyerahmad  339.82  5697.06  0.25  0.4  238.177  3585.047  0.2614  0.4133 
Golestan  1439.64  35333.35  0.18  0.33  1237.653  24184.02  0.1821  0.3728 
Gilan  508.84  143528.05  0.14  0.37  439.298  88628.456  0.1437  0.4038 
Lorestan  483.89  6004.13  0.24  0.31  365.061  4710.623  0.2407  0.3312 
Mazandaran  4042.83  77117.08  0.14  0.31  3398.741  55020.8  0.1454  0.3335 
Markazi  3435.27  24601.29  0.28  0.3  2382.4  19328.955  0.2917  0.3174 
Hormozgan  3159.51  8827.06  0.22  0.29  2209.037  6432.453  0.2174  0.2983 
Hamedan  1741.22  22891.93  0.23  0.27  1362.917  18492.39  0.2293  0.2981 
Yazd  2615.35  7838.14  0.16  0.18  2153.476  7174.672  0.1643  0.1896 
Figure 1 shows the predicted responses for the bivariate Gaussian model in Sistan and Baluchestan, and Gilan provinces, Iran. The patterns of these two plots were in contradiction. All figures for both univariate quasiPoisson and bivariate Gaussian GAM models are presented in the appendix (Appendix 2). Figure 2 illustrates the predicted percentage of speeding and unsafe distance offenses with their spatial patterns in Iran. The two plots had a different pattern that indicated the percentage of speeding is higher in the central, northeast, and southeast regions than in other parts of Iran. The predicted percentage of speeding was higher than unsafe distance, indicating that speeding is high in Iran. The predicted percentage of unsafe distance with a range of 20  40% was higher for north, northwest, west, and some parts of southwest than other parts of Iran. These plots were produced with the triangulated mesh method. The other mesh method is available for comparison in Appendix 3.
The temporal and hourly patterns of the predicted percentage for traffic offenses are available in Appendix 4.1 and Appendix 4.2, respectively. There were four intervals of 6 hours, including 0  5, 6  11, 12  17, and 18  23. According to Appendix 4.1, the predicted percentage of speeding was the highest at 0  5. On the other hand, Appendix 4.2 shows that the predicted percentage of unsafe distance was the lowest at 0  5 and had the highest value at 12  17 (Appendix 4.1 and Appendix 4.2).
5. Discussion
The speed limit and speeding fines for intercity highways are different in European countries and Iran. As illustrated in Figure 3, Norway has the highest speeding fines (€711), and the Czech Republic has the lowest speeding fines (€19) among other countries, respectively. In some countries, the speeding limits are different based on the two lanes or other roads (Norway), winter and other seasons (Sweden and Finland), rainy and other weather conditions (Luxembourg and Italy), and free speed (Germany). The traffic fines for speeding in Iran within 2021  2022 is 2,100,000 Iranian Rial (IR) (equivalently about €44.30 if €1 = 47,377 IRR (i.e., official governmental NIMA, the Central Bank of the Islamic Republic of Iran exchange rate) and about €6.72 if €1 = 312,480 IRR (i.e., unofficial open market exchange rate) in Q1 2022). Nevertheless, this comparison of speeding fines is simple and naive according to the different purchasing power parity and gross domestic product between countries; therefore, it is suggested to compare them to new indices, such as Big Mac Index, in future studies (19). The Figure 3 dataset is made from speedingeurope.com and rahvar120.ir datasets.
A significant relationship has been reported between road traffic accidents (RTA) and time (e.g., the time of the day) in Yazd, Iran, during the New Year holidays and summer (20). It has been the other study for these holidays in the six most populous provinces of Iran, namely Fars, Khorasan Razavi, Tehran, Isfahan, Kerman, and Khuzestan, within 2011  2015, indicating that Fars and Khorasan Razavi, with attractive tourist sites, have different high RTA among others (21). The mortality rate due to traffic accidents is higher in Iran at midnight and summer (22, 23) and in spring and summer in Shiraz (24) than in other times. The present study showed that the speeding rate was higher from midnight to early morning throughout Iran. This finding might be due to the existence of black spots with low and not uniform lighting on the roads (25). It also suggests that risky driving, not darkness, is the main reason for accidents (26). However, there are some exceptions; for example, in the south of Iran (27) and Yasuj (28), the rate of accidents during the day is higher than at night. The provinces in the south of Iran are located in the warmest region in the country, with an average daily high temperature. The high rate of accidents during the day in these provinces might be due to heat stress on drivers. A similar finding was reported in a study in Saudi Arabia (29) and high ambient temperatures in Spain (30). The percentage of RTA in Fars, Isfahan, Ilam, and the southeast region of Iran has a nonlinear trend in 24 hours with different peaks of speeding and unsafe distance (26, 3134).
The rate of unsafe distance offense is almost high in Kermanshah, Iran, and it can be added as a new risk factor of RDA in this province (35). The highest rate of mortality for drivers, passengers, and pedestrians has been reported during 13:00  18:00 in the west of Iran (36), 18:00  20:00 in Mashhad (37), and 16:00  18:00 in the southeast of Iran (38). The findings of other studies showed that most of the collisions occurred in the early hours of the night. A part of these collisions is due to poor visibility. Inadequate visibility has a key role in crashes involving pedestrians, motorcyclists, cyclists, and drivers (39). Moreover, crashes at dark hours cause severe injuries (40, 41). According to the findings of a metaanalysis study, the odds ratio of mortality in darkhour crashes is 53% higher than in dayhour crashes (42). Consequently, traffic offenses might have a relationship with RTArelated mortality. In this regard, this study suggests adding traffic offenses statistics to the Iranian Integrated Road Traffic Injury Registry (43, 44) and RTA studies. The other risk factors are the spatial variations of traffic offenses and accidents, seatbelt and helmet status, gender, age group (45), and climatic conditions, such as fog in the north of Iran (46).
The traffic fines and risky driving in Iran are studied in different ways, including the relationship between the number of traffic offenses and fuel costs within 2011  2019 (47), the relationship between increasing traffic fines policies and the road traffic law enforcement (48), the prevalence and determination of speeding in Iran (49), the comparison study of traffic fines in Iran and other countries (50), risky driving fined by police in 2006 and 2007 in Tehran (51), the effect of cameras on speeding behavior of taxi drivers in two highways (52), and aggressive violations (e.g., “sound horn to indicate your annoyance”, “get angry, give chase”, and “aversion, indicating hostility”). Moreover, Iran and Great Britain, the Netherlands, and Finland are among the countries with higher speeding violations than other countries, such as Greek and Turkey (53). In addition, visual, perceptual, and cognitive capabilities and physiological condition of drivers (e.g., Barkley’s Attention Deficit Hyperactivity Disorder Screening Test, Risk Perception Questionnaire, Risk Taking Questionnaire, Sensation Seeking Scale Survey, and Driver Behavior Questionnaire), among other factors in SHRP2 naturalistic driving study, are assessed in the USA (54).
Advanced and sophisticated statistical methods are in demand for trafficrelated datasets. The GAMs are among the statistical models that can be used for complex relationships, such as risky driving in Iran (4) (e.g., driving offenses near public places, such as airports (1)). The bivariate structure of the response can estimate the correlation and compare the traffic offenses between provinces at distinct times of the day (5). It also calculates the peak hours and 95% confidence interval with their pattern for each province. The GetisOrd General G* statistic in geographically weighted regression models revealed that the hotspot for fatal pedestrian accidents is in Mazandaran, Iran, and it is more common in Yazd, East Azerbaijan, and Ardebil (55). Future studies can investigate clustering methods, statistical learning methods (56), functional data analysis (57), and GAM for location, scale, and shape techniques to estimate the exact distribution with many parameters and their estimation for the underlying distribution (58).
5.1. Conclusions
The present study concluded that the risky driving behaviors due to traffic offenses can be estimated straightforwardly at different times and locations and add new information about the time of the days and roads that have not registered or occurred any traffic accidents. In this regard, they are predictive models. The geographical status of the roads, such as mountains or deserts, is shown to be related to the type of traffic offenses. For example, speeding violations on desert roads are higher than mountain roads, and unsafe distance violations on mountain roads are higher than desert roads. The daynight, rush hours, and holidays are the main timerelated factors for occurring traffic offenses. The future direction of this study is to investigate the relationships between the percentage of traffic offenses and traffic accident occurrence, climate status (e.g., raining, foggy, and sunny), and holidays and restrictions (e.g., coronavirus disease 2019 restriction) on all roads in different times of the day.
In highway safety research, crash modification factors and safety performance are introduced based on the traffic volume and road characteristics, and different statistical methods are proposed to estimate them (59, 60). Therefore, defining new and easytocompute indices is needed for future studies to measure and model the percentage of risky driving. This study had some limitations. Firstly, the police statistics have crime classification errors (e.g., some errors in detecting speeding and unsafe distance) and systematic errors (e.g., the failure of count stations in some hours) (61). Secondly, the statistics on traffic accidents are not publicly available.
References

1.
Fayaz M, Abadi AR, Khodakarim S, Hoseini MR, Razzaghi AR. The DataDriven Pattern for Healthy Behaviors of Car Drivers Based on Daily Records of Traffic Count Data from 2018 to 2019 near Airports: A Functional Data Analysis. JP J Biostat. 2020;17(2):53957. doi: 10.17654/bs017020539.

2.
World Health Organization. Global status report on road safety 2018: summary. Geneva: World Health Organization; 2018, [cited 2022]. Available from: https://www.who.int/publications/i/item/WHONMHNVI18.20.

3.
Hastie T, Friedman J, Tibshirani R. The Elements of Statistical Learning: Data Mining, Inference, and Prediction. New York, NY: Springer Series in Statistics; 2001.

4.
Wood SN. Generalized Additive Models: An Introduction with R. New York: CRC Press; 2017. doi: 10.1201/9781315370279.

5.
Wood SN, Pya N, Säfken B. Smoothing Parameter and Model Selection for General Smooth Models. J Am Stat Assoc. 2016;111(516):154863. doi: 10.1080/01621459.2016.1180986.

6.
Moraga P. Geospatial Health Data: Modeling and Visualization with RINLA and Shiny. New York: Chapman and Hall/CRC; 2019. doi: 10.1201/9780429341823.

7.
Chen W, Genton MG. Are You All Normal? It Depends. Ithaca, NY: arXiv; 2020, [cited 2022]. Available from: https://www.researchgate.net/publication/343877013.

8.
Joenssen DW, Vogel J. A power study of goodnessoffit tests for multivariate normality implemented in R. J Stat Comput Simul. 2012;84(5):105578. doi: 10.1080/00949655.2012.739620.

9.
Székely GJ, Rizzo ML. A new test for multivariate normality. J Multivar Anal. 2005;93(1):5880. doi: 10.1016/j.jmva.2003.12.002.

10.
Kutner MH. Applied linear statistical models. New York: McGrawHill Irwin; 2005.

11.
Rizzo ML, Szekely GJ, Rizzo MM. Package ‘energy’. 2021. Available from: https://cran.rproject.org/web/packages/energy/energy.pdf.

12.
Wood SN. mgcv: GAMs and generalized ridge regression for R. R News. 2001;1/2:205.

13.
Garnier S, Ross N, Rudis R, Camargo AP, Sciaini M, Cédric S. viridis  ColorblindFriendly Color Maps for R. Newark, NJ: Sjmgarnier; 2021, [cited 2022]. Available from: https://sjmgarnier.github.io/viridis/.

14.
Becker RA, Wilks AR, Ray Brownrigg R, Minka TP, Deckmyn A. maps: Draw Geographical Maps. Wien, Austria: The Comprehensive R Archive Network; 2018, [cited 2022]. Available from: https://pdf4pro.com/view/mapsdrawgeographicalmaps62d0bf.html.

15.
Bivand R, LewinKoh N, Pebesma E, Archer E, Baddeley A, Bearman N, et al. maptools: Tools for Handling Spatial Objects. Wien, Austria: The Comprehensive R Archive Network; 2022, [cited 2022]. Available from: https://cran.rproject.org/web/packages/maptools/index.html.

16.
Ribeiro Jr PJ, Diggle P, Christensen O, Schlather M, Bivand R, Ripley B, et al. geoR: Analysis of Geostatistical Data. Wien, Austria: The Comprehensive R Archive Network; 2022, [cited 2022]. Available from: https://cran.rproject.org/web/packages/geoR/index.html.

17.
Bivand R, Keitt T, Rowlingson B, Pebesma E, Sumner M, Hijmans R, et al. rgdal: Bindings for the 'Geospatial' Data Abstraction Library. Wien, Austria: The Comprehensive R Archive Network; 2021, [cited 2022]. Available from: https://cran.rproject.org/web/packages/rgdal/index.html.

18.
Wickham H. ggplot2: Elegant Graphics for Data Analysis. New York: Springer; 2009. doi: 10.1007/9780387981413.

19.
Clements KW, Si J. Simplifying The Big Mac Index. J Int Financial Manag Account. 2017;28(1):8699. doi: 10.1111/jifm.12058.

20.
Lotfi M, Montazeri M, Lashkardoost H, Shamsi F, Askari M, Hamedi E, et al. Road traffic accidents in Yazd province, Iran: A longitudinal study (2012–2016). Arch Trauma Res. 2018;7(2):6872. doi: 10.4103/atr.atr_9_18.

21.
Besharati MM, Azizi Bondarabadi M, Memariyan M, Tavakoli Kashani A. Patterns of road traffic fatalities in the six most populous provinces of Iran, 2011–2015. Arch Trauma Res. 2019;8(3):17781. doi: 10.4103/atr.atr_91_18.

22.
SadeghiBazargani H, Ayubi E, AzamiAghdash S, Abedi L, Zemestani A, Amanati L, et al. Epidemiological Patterns of Road Traffic Crashes During the Last Two Decades in Iran: A Review of the Literature from 1996 to 2014. Arch Trauma Res. 2016;5(3). e32985. doi: 10.5812/atr.32985. [PubMed: 27800461]. [PubMed Central: PMC5078874].

23.
Shahbazi F, Soori H, Khodakarim S, Ghadirzadeh MR, Hashemi Nazari SS. Analysis of mortality rate of road traffic accidents and its trend in 11 years in Iran. Arch Trauma Res. 2019;8(1):1722. doi: 10.4103/atr.atr_72_18.

24.
Ghaem H, Hajipour M, Tababataee HR, Yadollahi M, Izanloo F. Time Series Analysis of Mortalities Resulting from Car Accidents in the Injured Individuals Hospitalized in Shiraz Shahid Rajaee Hospital During 2010  2016. Trauma Mon. 2018;23(1). e13573. doi: 10.5812/traumamon.13573.

25.
Mohan A, Landge VS. Identification of Accident Black Spots on National Highway. Int J Civ Eng Technol. 2017;8(4):58896.

26.
Mohammadi G. The pattern of fatalities by age, seat belt usage and time of day on road accidents. Int J Inj Contr Saf Promot. 2009;16(1):2733. doi: 10.1080/17457300802406963. [PubMed: 19058047].

27.
Rakhshani T, Rakhshani F, Asadi ZS, Hadiabasi M, Khorramdel K, Zarenezhad M. Study of the pattern of mortality caused by Traffic Accidents (TAs) in The South of Iran. J Pak Med Assoc. 2016;66(6):6449. [PubMed: 27339561].

28.
Rakhshani T, Kashfi M, Amirian I, Ebrahimi M, Hashemi Nazari S. Epidemiology of Fatal Road Traffic Accidents in Iran, Yasouj, 20142015. J Health Sci Surveill Syst. 2018;6(1):2935.

29.
Nofal FH, Saeed AA. Seasonal variation and weather effects on road traffic accidents in Riyadh city. Public Health. 1997;111(1):515. doi: 10.1038/sj.ph.1900297. [PubMed: 9033225].

30.
Basagana X, EscaleraAntezana JP, Dadvand P, Llatje O, BarreraGomez J, Cunillera J, et al. High Ambient Temperatures and Risk of Motor Vehicle Crashes in Catalonia, Spain (20002011): A TimeSeries Analysis. Environ Health Perspect. 2015;123(12):130916. doi: 10.1289/ehp.1409223. [PubMed: 26046727]. [PubMed Central: PMC4671248].

31.
Hasanzadeh J, Moradinazar M, Najafi F, AhmadiJouybary T. Trends of Mortality of Road Traffic Accidents in Fars Province, Southern Iran, 2004  2010. Iran J Public Health. 2014;43(9):125965. [PubMed: 26175980]. [PubMed Central: PMC4500428].

32.
Mansouri M, Javad Kargar M. Analysis and Monitoring of the Traffic Suburban Road Accidents Using Data Mining Techniques; A Case Study of Isfahan Province in Iran. Open Transp J. 2014;8(1):3949. doi: 10.2174/1874447801408010039.

33.
Mohammadfam I, Karami Naserkhani R, Soltanian AR. The analysis of deaths caused by driving accidents in Ilam province, western Iran and the related factors by using the method of time series. Int J Occup Hyg. 2016;8(4):2007.

34.
Khorshidi A, Ainy E, Hashemi Nazari SS, Soori H. Temporal Patterns of Road Traffic Injuries in Iran. Arch Trauma Res. 2016;5(2). e27894. doi: 10.5812/atr.27894. [PubMed: 27703958]. [PubMed Central: PMC5037289].

35.
Izadi N, Khoram Dad M, Jamshidi P, Zanganeh AR, Shafiei J, Firouzi A. Epidemiological Pattern and Mortality Rate Trend of Road Traffic Injuries in Kermanshah Province (20092014). J Community Health Res. 2016;5(3):15868.

36.
Hamzeh B, Najafi F, Karamimatin B, Ahmadijouybari T, Salari A, Moradinazar M. Epidemiology of traffic crash mortality in west of Iran in a 9 year period. Chin J Traumatol. 2016;19(2):704. doi: 10.1016/j.cjtee.2015.12.007. [PubMed: 27140212]. [PubMed Central: PMC4897842].

37.
Sarbaz M, Kimiafar K, Khadem Rezaiyan M, Banaye Yazdipour AR. Epidemiology of transport accidents based on international statistical classification of diseases (ICD10) in Mashhad, Iran. Int Electron J Med. 2018;7(1):239. doi: 10.31661/iejm801.

38.
Rad M, Martiniuk AL, AnsariMoghaddam A, Mohammadi M, Rashedi F, Ghasemi A. The Pattern of Road Traffic Crashes in South East Iran. Glob J Health Sci. 2016;8(9):14958. doi: 10.5539/gjhs.v8n9p149. [PubMed: 27157159]. [PubMed Central: PMC5064071].

39.
World Health Organization. World report on road traffic injury prevention. Geneva: World Health Organization; 2004, [cited 2022]. Available from: https://www.who.int/publications/i/item/worldreportonroadtrafficinjuryprevention.

40.
Ackaah W, Apuseyine BA, Afukaar FK. Road traffic crashes at nighttime: characteristics and risk factors. Int J Inj Contr Saf Promot. 2020;27(3):3929. doi: 10.1080/17457300.2020.1785508. [PubMed: 32588731].

41.
Ramadani N, Zhjeqi V, Berisha M, Hoxha R, Begolli I, Salihu D, et al. Public Health Profile of Road Traffic Accidents in Kosovo 20102015. Open Access Maced J Med Sci. 2017;5(7):103641. doi: 10.3889/oamjms.2017.214. [PubMed: 29362641]. [PubMed Central: PMC5771275].

42.
Yousefifard M, Toloui A, Ahmadzadeh K, Gubari MIM, Madani Neishaboori A, Amraei F, et al. Risk Factors for Road Traffic InjuryRelated Mortality in Iran; a Systematic Review and MetaAnalysis. Arch Acad Emerg Med. 2021;9(1). e61. doi: 10.22037/aaem.v9i1.1329. [PubMed: 34580659]. [PubMed Central: PMC8464012].

43.
SadeghiBazargani H, Sadeghpour A, Lowery Wilson M, Ala A, Rahmani F. Developing a National Integrated Road Traffic Injury Registry System: A Conceptual Model for a Multidisciplinary Setting. J Multidiscip Healthc. 2020;13:98396. doi: 10.2147/JMDH.S262555. [PubMed: 33061404]. [PubMed Central: PMC7520136].

44.
Marin S, Pourasghar F, Moghisi AR, Samadirad B, Haddadi M, KhorasaniZavareh D, et al. Development and psychometric evaluation of data collection tools for Iranian integrated road traffic injury registry: Registrarstation data collection tool. Arch Trauma Res. 2019;8(3):1706. doi: 10.4103/atr.atr_40_18.

45.
Fathollahi S, Saeedi Moghaddam S, Rezaei N, Jafari A, Peykari N, Haghshenas R, et al. Prevalence of behavioural risk factors for roadtraffic injuries among the Iranian population: findings from STEPs 2016. Int J Epidemiol. 2019;48(4):118796. doi: 10.1093/ije/dyz021. [PubMed: 30843066].

46.
KhodadadiHassankiadeh N, Rad EH, Koohestani HS, KouchakinejadEramsadati L. The Pattern of Road Accidents in Fog and the Related Factors in North of Iran in 20142018 . Durham, North Carolina: Research Square; 2020, [cited 2022]. Available from: https://www.researchsquare.com/article/rs73501/v1.

47.
Delavary M, Ghayeninezhad Z, Lavallière M. Evaluating the Impact of Increased Fuel Cost and Iran’s Currency Devaluation on Road Traffic Volume and Offenses in Iran, 2011–2019. Safety. 2020;6(4):49. doi: 10.3390/safety6040049.

48.
Delavary Foroutaghe M, Mohammadzadeh Moghaddam A, Fakoor V. Impact of law enforcement and increased traffic fines policy on road traffic fatality, injuries and offenses in Iran: Interrupted time series analysis. PLoS One. 2020;15(4). e0231182. doi: 10.1371/journal.pone.0231182. [PubMed: 32302374]. [PubMed Central: PMC7164613].

49.
Rahimi H, Hashemi Nazari SS, Soori H, Motevalian SA, Momeni E, Azar A. Traffic Police Effectiveness and Efficiency Evaluations, an Overview of Methodological Considerations. Arch Trauma Res. 2017;6(1). e36927. doi: 10.5812/atr.36927.

50.
Safarzadeh M, Bagheri R. [Comparative studies of traffic fines by traffic police in Iran and other countries]. Rahvar. 2012;9(17):5974. Persian.

51.
Shams M, RahimiMovaghar V. Risky driving behaviors in Tehran, Iran. Traffic Inj Prev. 2009;10(1):914. doi: 10.1080/15389580802492280. [PubMed: 19214883].

52.
Tavolinejad H, Malekpour MR, Rezaei N, Jafari A, Ahmadi N, Nematollahi A, et al. Evaluation of the effect of fixed speed cameras on speeding behavior among Iranian taxi drivers through telematics monitoring. Traffic Inj Prev. 2021;22(7):55963. doi: 10.1080/15389588.2021.1957100. [PubMed: 34424783].

53.
de Winter JCF, Dodou D. National correlates of selfreported traffic violations across 41 countries. Pers Individ Differ. 2016;98:14552. doi: 10.1016/j.paid.2016.03.091.

54.
Antin J. Design of the InVehicle Driving Behavior and Crash Risk Study: In Support of the SHRP 2 Naturalistic Driving Study. Washington, DC: Transportation Research Board; 2011.

55.
Kavousi A, Moradi A, Soori H, Rahmani K. Environmental factors influencing the distribution of pedestrian traffic accidents in Iran. Arch Trauma Res. 2020;9(1):815. doi: 10.4103/atr.atr_76_19.

56.
James G, Witten D, Hastie T, Tibshirani R. An introduction to statistical learning. New York, NY: Springer; 2013.

57.
Ramsay JO; Silverman. Functional Data Analysis. New York, NY: Springer; 2005. doi: 10.1007/b98888.

58.
Stasinopoulos D, Rigby RA. Generalized Additive Models for Location Scale and Shape (GAMLSS) inR. J Stat Softw. 2007;23(7):146. doi: 10.18637/jss.v023.i07.

59.
Banks D, Persaud B, Lyon C, Eccles K, Himes S. Enhancing Statistical Methodologies for Highway Safety Research – Impetus from FHWA. McLean, VA; 2014. Contract No.: FHWAHRT14081.

60.
Donnell E, Hanks E, Porter RJ, Cook L, Srinivasan R, Li F, et al. The Development of Crash Modification Factors: Highway Safety Statistical Paper Synthesis. McLean, VA; 2020. Contract No.: FHWAHRT20069.

61.
Lohr S. Measuring Crime: Behind the Statistics. New York: Chapman and Hall/CRC; 2019. doi: 10.1201/9780429201189.