This is a mock report exploring data from STATS19’s R package1 made by someone who has never worked with that kind of dataset before. STATS19 provides three types of datasets: accidents, vehicles and casualties.

Accidents 2018

Initial exploration

Let’s see how many observations do we have as well as the variables’ number and types.

Data summary
Name accidents2018
Number of rows 122635
Number of columns 31
_______________________
Column type frequency:
Date 1
factor 22
numeric 8
________________________
Group variables None

Variable type: Date

skim_variable n_missing complete_rate min max median n_unique
datetime 13 1 2018-01-01 2018-12-31 2018-07-05 365

Variable type: factor

skim_variable n_missing complete_rate ordered n_unique top_counts
accident_index 0 1.00 FALSE 122635 201: 1, 201: 1, 201: 1, 201: 1
police_force 0 1.00 FALSE 51 Met: 25390, Wes: 5490, Ken: 4403, Wes: 4132
accident_severity 0 1.00 FALSE 3 Sli: 97799, Ser: 23165, Fat: 1671
date 0 1.00 FALSE 365 201: 504, 201: 498, 201: 491, 201: 488
day_of_week 0 1.00 FALSE 7 Fri: 20021, Thu: 18656, Wed: 18397, Tue: 17950
time 13 1.00 FALSE 1438 17:: 1154, 18:: 1093, 17:: 1086, 16:: 1069
local_authority_district 0 1.00 FALSE 380 Bir: 2614, Lee: 1548, Wes: 1509, Lam: 1287
local_authority_highway 0 1.00 FALSE 207 Ken: 3811, Sur: 3113, Lan: 2676, Ham: 2615
first_road_class 0 1.00 FALSE 6 A: 53499, Unc: 43355, B: 14210, C: 7005
road_type 0 1.00 FALSE 6 Sin: 88323, Dua: 19473, Rou: 7573, One: 3366
junction_detail 0 1.00 FALSE 10 Not: 52076, T o: 35958, Cro: 11422, Rou: 9974
junction_control 0 1.00 FALSE 5 Dat: 54842, Giv: 53259, Aut: 13323, Sto: 750
second_road_class 52211 0.57 FALSE 6 Unc: 48631, A: 12213, B: 4662, C: 4168
pedestrian_crossing_human_control 0 1.00 FALSE 4 Non: 117924, Dat: 3173, Con: 1116, Con: 422
pedestrian_crossing_physical_facilities 0 1.00 FALSE 7 No : 94877, Ped: 9753, Pel: 7169, Zeb: 4583
light_conditions 0 1.00 FALSE 5 Day: 88435, Dar: 24746, Dar: 6120, Dar: 2477
weather_conditions 0 1.00 FALSE 10 Fin: 99221, Rai: 12789, Unk: 3666, Oth: 2603
road_surface_conditions 0 1.00 FALSE 6 Dry: 90546, Wet: 28215, Fro: 1417, Dat: 1223
special_conditions_at_site 0 1.00 FALSE 9 Non: 118495, Dat: 1524, Roa: 1372, Aut: 284
carriageway_hazards 0 1.00 FALSE 7 Non: 119170, Dat: 1325, Oth: 1072, Any: 376
urban_or_rural_area 1 1.00 FALSE 3 Urb: 82583, Rur: 39996, Una: 55
lsoa_of_accident_location 6445 0.95 FALSE 27965 E01: 165, E01: 123, E01: 84, E01: 82

Variable type: numeric

skim_variable n_missing complete_rate mean sd p0 p25 p50 p75 p100 hist
longitude 55 1 -1.26 1.40 -7.27 -2.19 -1.15 -0.14 1.76 ▁▁▅▇▃
latitude 55 1 52.43 1.38 49.91 51.47 51.89 53.39 60.76 ▇▆▁▁▁
number_of_vehicles 0 1 1.85 0.72 1.00 1.00 2.00 2.00 24.00 ▇▁▁▁▁
number_of_casualties 0 1 1.31 0.76 1.00 1.00 1.00 1.00 59.00 ▇▁▁▁▁
first_road_number 0 1 836.74 1670.33 0.00 0.00 41.00 580.00 9621.00 ▇▁▁▁▁
speed_limit 0 1 37.11 14.07 20.00 30.00 30.00 40.00 70.00 ▇▁▁▂▁
second_road_number 0 1 291.80 1129.17 -1.00 0.00 0.00 0.00 9620.00 ▇▁▁▁▁
did_police_officer_attend_scene_of_accident 0 1 1.29 0.47 -1.00 1.00 1.00 2.00 3.00 ▁▁▇▃▁

The table above shows that overall we do not have significant missing data in any of the 16 variables as well as some basic statistics of the (few) numerical variables. Now let’s see the mode for every variable.

The tables above pose interesting (basic) research questions to be explored. As an example, seeing that the day of the week were most accidents take place is Friday, I would like to know if most accidents happen during weekdays or weekend. We could use that data as a proxy to infer if professional drivers are more or less involved in accidents than amateurs, especially if we combine that with the hour of the day.

Surprisingly, most accidents take place on dry conditions with sunny days and good visibility, so, apparently, weather does not have such as big impact as I might have guessed on the first sight, although verifying it would require further analysis.

Accidents’ evolution over time

There have been a total of 122,635 accidents in 2018, out of which a 1% were fatal, 19% were serious, and 80% were slight. However, let’s see how these figures have been evolved through time and if there has been an increase or decrease on the number of accidents.

Wile the number of accidents in UK is high, we can see an overall tendency in number of accidents to decrease over time, but can we observe other patterns?

year Fatal Fatal variation Serious Serious variation Slight Slight variation Total Total variation
2009 2057 NA 21997 NA 139500 NA 163554 NA
2010 1731 -18.83% 20440 -7.62% 132243 -5.4876% 154414 -5.919%
2011 1797 3.67% 20986 2.60% 128691 -2.7601% 151474 -1.941%
2012 1637 -9.77% 20901 -0.41% 123033 -4.5988% 145571 -4.055%
2013 1608 -1.80% 19624 -6.51% 117428 -4.7731% 138660 -4.984%
2014 1658 3.02% 20676 5.09% 123988 5.2908% 146322 5.236%
2015 1616 -2.60% 20038 -3.18% 118402 -4.7178% 140056 -4.474%
2016 1695 4.66% 21725 7.77% 113201 -4.5945% 136621 -2.514%
2017 1676 -1.13% 22534 3.59% 105772 -7.0236% 129982 -5.108%
2018 1671 -0.30% 23165 2.72% 97799 -8.1524% 122635 -5.991%

As can be seen in the table above, total number of accidents has been decreasing over time and 2018 is the year with less total accidents since 2009. This might seem good news (with plenty of room for improvement, provided that the accidents figures are still high), but we can also observe that there has been a slight increment on serious accidents, being 2018 the year whith most serious accidents in 2009, at the cost of slight accidents. This means that while there is a tendency of fatal accidents to decrease since 2009, it is also true that the number of fatal accidents has been more or less stable during the last 3 years.

Accidents distribution by time

Day of the Week 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
Sunday 490 369 321 270 203 176 205 237 296 498 691 802 1022 1033 975 1013 926 923 864 728 574 463 397 322
Monday 218 136 125 96 104 174 389 1000 1446 973 778 845 984 966 1082 1460 1512 1630 1288 845 562 454 397 274
Tuesday 165 104 67 68 68 146 472 1015 1590 1029 774 848 899 941 996 1384 1538 1740 1389 942 656 456 403 258
Wednesday 163 100 95 64 85 171 375 1103 1646 953 766 850 985 888 1063 1542 1562 1705 1400 942 646 514 456 322
Thursday 176 111 88 77 83 177 422 1052 1632 934 851 828 944 1007 1047 1474 1558 1732 1475 957 674 533 476 348
Friday 245 148 114 80 83 179 408 941 1471 894 768 937 1096 1169 1261 1606 1714 1732 1468 1117 784 654 590 558
Saturday 380 299 243 180 172 175 210 290 457 594 861 1000 1186 1134 1130 1021 1095 1153 1007 948 801 602 547 584

As can easily be seen in the table above, most accidents take place during peak hours in weekdays and there is a tendency to increase the closer it gets to Friday evening, which is probably the busiest time and when people is more tired.

Accidents’ spatial distribution

Let’s see how accidents are spatially distributed to see if we can identify hot areas. The following map displays accidents by type, displaying slight accidents, as they the most significant ones.

Having the coordinates of every accident, we could also analyse them at a closer scale. As suggested in the Active Travel Podcast Pilot: Media reporting of Active Travel, it could be interesting to view a picture of the places where accidents took place in order to identify possible correlation with their physical features and the number of accidents and casualties. As a protoype, the following code gets the picture from mapillary of the top-5 location whith more casualties, which could be the foundations of a larger research based on machine learning.

## [1] "Displaying mapillary image close to lon=-0.818005 and lat=52.43432"

## [1] "Displaying mapillary image close to lon=-4.328339 and lat=55.873593"

## [1] "Displaying mapillary image close to lon=-0.561374 and lat=51.914048"

## [1] "Displaying mapillary image close to lon=0.003746 and lat=52.614004"

## [1] "Displaying mapillary image close to lon=-1.197392 and lat=51.250871"

Casualties

Initial exploration

Let’s see how many observations do we have as well as the variables’ number and types.

Data summary
Name casualties2018
Number of rows 160597
Number of columns 16
_______________________
Column type frequency:
factor 13
numeric 3
________________________
Group variables None

Variable type: factor

skim_variable n_missing complete_rate ordered n_unique top_counts
accident_index 0 1 FALSE 122635 201: 59, 201: 29, 201: 23, 201: 20
casualty_class 0 1 FALSE 3 Dri: 103371, Pas: 34794, Ped: 22432
sex_of_casualty 0 1 FALSE 3 Mal: 95252, Fem: 65305, Dat: 40
age_band_of_casualty 0 1 FALSE 12 26 : 33242, 36 : 24225, 46 : 22454, 21 : 18187
casualty_severity 0 1 FALSE 3 Sli: 133302, Ser: 25511, Fat: 1784
pedestrian_location 0 1 FALSE 12 Not: 138163, In : 9153, Cro: 3603, On : 2308
pedestrian_movement 0 1 FALSE 10 Not: 138163, Cro: 7274, Unk: 6205, Cro: 4648
car_passenger 0 1 FALSE 4 Not: 131009, Fro: 18048, Rea: 11057, Dat: 483
bus_or_coach_passenger 0 1 FALSE 6 Not: 157064, Sea: 2218, Sta: 956, Ali: 160
pedestrian_road_maintenance_worker 0 1 FALSE 4 No : 153620, Not: 6852, Yes: 87, Dat: 38
casualty_type 7 1 FALSE 21 Car: 90913, Ped: 22432, Cyc: 17550, Mot: 7221
casualty_home_area_type 0 1 FALSE 4 Urb: 115934, Dat: 16594, Rur: 15534, Sma: 12535
casualty_imd_decile 0 1 FALSE 11 Dat: 27345, Mor: 16684, Mos: 16007, Mor: 15893

Variable type: numeric

skim_variable n_missing complete_rate mean sd p0 p25 p50 p75 p100 hist
vehicle_reference 0 1 1.48 2.56 1 1 1 2 999 ▇▁▁▁▁
casualty_reference 0 1 1.40 2.70 1 1 1 1 991 ▇▁▁▁▁
age_of_casualty 0 1 37.06 19.66 -1 22 34 50 102 ▃▇▅▂▁

The table above shows that overall we do not have significant missing data in any of the 16 variables as well as some basic statistics of the (few) numerical variables. Now let’s see the mode for every variable.

From the tables above, we can profile the average casualty in 2018 as a male between 26-35 years old, driver of a car that has an accident in urban areas and gets slightly injured after the accident. Let’s further explore the casualties’ demographics.

Casualties’ demographics

At this level of detail, we cannot see notable differences between genders. Both male and female seem to follow the same age distribution, although admittedly, females absolute numbers are notably smaller in all the ages.

Let’s see if both genders follow same distribution according to accident severity.

As can be seen in the plots above, the number of young females involved in fatal and severe accidents are much lesser than those to their male equals.

Future actions and research

This is the end (for now) of this mock report aimed to know about the STATS19 dataset as well as some new coding. There is still lots of data to be explored that, in turn, will lead to research questions, especially if we combine the different datasets together (thankfully they have an accident_index that will make it possible).

We have seen many unanswered questions in this document, and others that have not been directly mentioned, such as the role of women involved in accidents are usually drivers or not.

Another thing I would love to do is to join vehicles and accidents to see if accidents’ severity follows a similar distribution according to the type of vehicles involved. My hypothesis here is that fatal accidents involving cars will be much higher than those involving bicicles, which I expect them to be quite marginal.

Also, I would love to study the impact of the physical conditions of the highways and environment. Although accidents dataset has some information about it, I don’t think it is enough, so, as an OpenStreetMap contributor and advocate, I would love to combine both datasets.

  1. Lovelace, R., Morgan, M., Hama, L., Padgham, M., Ranzolin, D., & Sparks, A. (2019). stats 19: A package for working with open road crash data. The Journal of Open Source Software, 4(33), 1181. https://doi.org/10.21105/joss.01181↩