Finalization of 2022 Crash Data
Synopsis
We finalized 2022 crash data on 2023-05-12 at about 15:45. As with 2021 data, this was an updatable freeze, which was will allow us to add Reference Point coding data once (or as) they become available without making changes to any other data.
We have created and tested analysis extracts of these finalized data. These extracts have been deployed to Community Maps and WisTransportal (though not labeled as such). Creation of the SAS libraries for finalized 2022 data ought to be a routine matter of changing a few parameters in well established code. If the workgroup approves release of 2022 finalized data, these datasets will be available as such by the first business day of next week.
Overall Numbers
Parameter | 2022 | 2021 | 2020 | 2019 | 2018 | 2017 |
---|---|---|---|---|---|---|
Total Crashes | 128830 | 128296 |
| 145288 | 144212 | 139870 |
Fatal Crashes | 547 | 546 | 540 | 511 |
| 539 |
Fatalities | 595 | 595 | 593 | 551 | 576 | 594 |
Serious Injury Crashes | 2740 | 2880 | 2651 | 2621 | 2714 | 2931 |
Serious Injuries | 3213 | 3473 | 3186 | 3133 | 3212 | 3492 |
Motorcycle Crashes | 1921 | 2082 | 2104 | 1806 | 1967 | 2206 |
Bicycle Crashes | 668 | 691 | 619 | 759 | 852 | 906 |
Pedestrian Crashes | 1324 | 1273 | 1137 | 1460 | 1505 | 1533 |
Construction Zone Crashes | 2034 | 2186 | 1805 | 2489 | 3166 | 2795 |
FMCSA Reportable Crashes | 2720 | 2654 | 2327 | 3126 | 3097 | 3229 |
These Numbers
differ from those presented in Final year-end crash statistics
A Note on Fatal Crashes
The current state of the 2022 crash data reflects our current assessment of the FARS status of these crashes. There are 3 fatalities whose FARS reportability hinges on cause of death information that has not yet been released. One of these is a presumed FARS case. It was announced in a Daily Fatality Report, is tabulated at the Department’s crash statistics pages, and appears as a fatality in the crash data. The other two are presumed Non Motor Vehicle Traffic Fatalities (NMVTFs), and we flagged them as not fatalities prior to finalization. If the cause of death information contradicts any of these presumptions, we will have late changes to the FARS totals that will not be reflected in these finalized data.
Edit Checks
Critical Checks
We were able to most of the critical edit checks for 2022 finalization. Three remain in the finalized 2022 crash data, but their impacts on data analysis should be small.
Fatality without Date of Death
There are two fatalities without dates of death. These individuals died after the crash and outside of Wisconsin. We have thus far been unable to obtain death certificate information for these individuals and do not have a definite date of death for them.
Flag Check: Restricted Citations
An error in the code that maintains the crash restricted flag causes some crashes to be incorrectly flagged as restricted (for containing juvenile criminal record information).
The most significant impact of this bug is that it makes crash reports that ought to be easily accessible to the public somewhat more difficult to obtain.
Changes to Critical Checks
As a part of the CRU’s FARS case management modernization project, we have discontinued 12 fatality related critical batch edit checks, replacing them with a new system of edit checks run against the FARS case data and the crash data. There are currently over 30 of these new checks.
The Narrative: Bicycle Mentioned, but not a Unit check had been a new critical for 2022, but we decided to lower its priority to important in the face of insufficient resources to follow-up on edit checks.
Important Checks
After Tejal’s departure in mid-2022, the CRU has been unable to keep up it’s work of prompt follow-up to these edit checks. So 2022 data are not as clean with respect to important checks as we would like. we are currently at work on creating a systematic edit check follow-up system based on Jira and Confluence. Edit check follow-up will be classified and managed in Jira and solution patterns (which include how-to fix documentation) will be used to more quickly and easily make these follow-ups. We have already seen some success with this strategy in our transmission failure follow-up work.
Crash Data Flags
The Witness Flag is Tainted
In checking the 2022 finalized data extracts against a (several hours older) productions extract, we discovered an issue with the Witness flag. Investigation of this anomaly showed that the Witness flag degrades as changes are made to the crash data in Resolve.
This bug has been around since the beginning and affects finalized data going all the way back to 2017.
Year | Crashes | Incorrect Witness Flag | Percent Incorrect |
---|---|---|---|
2017 | 139870 | 4229 | 3.0 |
2018 | 144212 | 4475 | 3.1 |
2019 | 145288 | 3965 | 2.7 |
2020 | 114697 | 3226 | 2.8 |
2021 | 128296 | 1438 | 1.1 |
2022 | 128830 | 849 | 0.7 |
Interestingly this problem appears to show up most often as RP updates are made, which accounts for the lower frequency of this error in the 2021 and 2022 finalized data.
Flag Review, Verification and Validation
This discovery that the Witness flag is being degraded by seemingly unrelated changes to crash data and the fact that it when unnoticed for so long indicates that we should consider a systematic examination of the crash database flags. It seems unlikely that some of the more commonly used flags would have this issue and that it be unseen for long.
Review
Create a document for each flag. Describe its meaning and show the rules used to set it.
Verification
Establish edit checks to verify that these flags, as they appear in analysis extracts, satisfy their identified rules.
Validation
Establish edit checks that compare the flags to additional or different data. Effectively this compares the defined flags to alternative definitions that will allow us to determine how robustly they identify crashes as intended.