Relationships between vehicle type and crash severity
To explore the relationship between vehicles and crash severity, we must first join on the ‘accidents’ table:
a = get_stats19(year = 2018, type = "accidents")
#> Files identified: dftRoadSafetyData_Accidents_2018.csv
#> http://data.dft.gov.uk.s3.amazonaws.com/road-accidents-safety-data/dftRoadSafetyData_Accidents_2018.csv
#> Data already exists in data_dir, not downloading
#> Data saved at ~/stats19-data/dftRoadSafetyData_Accidents_2018.csv
#> Reading in:
#> ~/stats19-data/dftRoadSafetyData_Accidents_2018.csv
#> date and time columns present, creating formatted datetime column
va = dplyr::inner_join(v, a)
#> Joining, by = "accident_index"Now we have additional variables available to us:
dim(v)
#> [1] 226409 24
dim(va)
#> [1] 226409 56
names(va)
#> [1] "accident_index"
#> [2] "vehicle_reference"
#> [3] "vehicle_type"
#> [4] "towing_and_articulation"
#> [5] "vehicle_manoeuvre"
#> [6] "vehicle_location_restricted_lane"
#> [7] "junction_location"
#> [8] "skidding_and_overturning"
#> [9] "hit_object_in_carriageway"
#> [10] "vehicle_leaving_carriageway"
#> [11] "hit_object_off_carriageway"
#> [12] "first_point_of_impact"
#> [13] "was_vehicle_left_hand_drive"
#> [14] "journey_purpose_of_driver"
#> [15] "sex_of_driver"
#> [16] "age_of_driver"
#> [17] "age_band_of_driver"
#> [18] "engine_capacity_cc"
#> [19] "propulsion_code"
#> [20] "age_of_vehicle"
#> [21] "driver_imd_decile"
#> [22] "driver_home_area_type"
#> [23] "vehicle_imd_decile"
#> [24] "vehicle_type2"
#> [25] "location_easting_osgr"
#> [26] "location_northing_osgr"
#> [27] "longitude"
#> [28] "latitude"
#> [29] "police_force"
#> [30] "accident_severity"
#> [31] "number_of_vehicles"
#> [32] "number_of_casualties"
#> [33] "date"
#> [34] "day_of_week"
#> [35] "time"
#> [36] "local_authority_district"
#> [37] "local_authority_highway"
#> [38] "first_road_class"
#> [39] "first_road_number"
#> [40] "road_type"
#> [41] "speed_limit"
#> [42] "junction_detail"
#> [43] "junction_control"
#> [44] "second_road_class"
#> [45] "second_road_number"
#> [46] "pedestrian_crossing_human_control"
#> [47] "pedestrian_crossing_physical_facilities"
#> [48] "light_conditions"
#> [49] "weather_conditions"
#> [50] "road_surface_conditions"
#> [51] "special_conditions_at_site"
#> [52] "carriageway_hazards"
#> [53] "urban_or_rural_area"
#> [54] "did_police_officer_attend_scene_of_accident"
#> [55] "lsoa_of_accident_location"
#> [56] "datetime"Let’s see how crash severity relates to the variables of interest mentioned above:
xtabs(~vehicle_type2 + accident_severity, data = va) %>% prop.table()
#> accident_severity
#> vehicle_type2 Fatal Serious Slight
#> Bus 0.0002605903 0.0035820131 0.0178791479
#> Car 0.0082063876 0.1146465026 0.6043487671
#> Cycle 0.0004814296 0.0174993044 0.0620735041
#> Goods 0.0020626389 0.0144075545 0.0631202823
#> Motorbike 0.0017181296 0.0259574487 0.0513407153
#> Other 0.0004019275 0.0025308181 0.0094828386
xtabs(~vehicle_type2 + accident_severity, data = va) %>% prop.table() %>% plot()
As expected, crashes involving large vehicles such as buses and good vehicles tend to be more serious (involve proportionally more deaths) than crashes involving smaller vehicles.
To focus only on cars, we can filter the va table as follows:
The best proxy we have for car type in the open STATS19 data (there are non-open versions of the data with additional columns) is engine capacity, measured in cubic centimetres (cc). The distribution of engine cc’s in the cars dataset created above is shown below.
summary(vac$engine_capacity_cc)
#> Min. 1st Qu. Median Mean 3rd Qu. Max.
#> -1 1198 1560 1440 1968 19997The output shows that there are some impossible values in the data, likely due to recording error. Very few cars have an engine capacity above 5 litres (5000 cc) and we can be confident that none have an engine capacity below 300 cc. We’ll identify these records and remove them as follows:
We have set the anomolous vehicle size data to NA meaning it will not be used in the subsequent analysis.