Crash Data Improvement Program (CDIP) Final Report
APPENDIX A: CDIP INFORMATION REQUEST
CRASH DATA IMPROVEMENT PROGRAM PRE-SITE VISIT INFORMATION COLLECTION
PLEASE PROVIDE THE FOLLOWING RESOURCES:
- The most recent Traffic Records Assessment (can also be obtained thru NHTSA – with permission of the State).
- The most current version of their Traffic Records Strategic Plan used for Section 408 applications (can also be obtained thru NHTSA – with permission of the State).
- The CDIP questionnaires that are sent to the state for completion (attached).
- A current version of their crash production process flow chart (this is one of the items requested in the CDIP questionnaire).
- Any other materials the state feels may be relevant, e.g. research studies that highlight difficulties with using state crash data, or any previous documentation on measuring crash data quality.
CRASH DATA ANALYSIS
In addition to the above resources, the FHWA CDIP project team would like to opportunity to review the State's crash data prior to the site visit. Below is a list of crash data analyses to be conducted by the state. If the State does not have resources to conduct the analysis, please provide 3-5 years of historical crash data to the project team, and they can conduct the analyses.
Timeliness:
- A: Overall: Event date to availability on the statewide database (ready for use/analysis) - annual averages.
- B: If available, specific components of timeliness:
- Law Enforcement submission: Time from Event to receipt of crash reports - annual averages.
- Data Entry: Time from receipt to completion of primary data entry - annual averages.
- Location Coding: Time from completion of primary data entry to application of location codes - annual averages.
- Other: Any other processes (QC, data cleansing, financial responsibility, other?) required prior to data availability for use/analysis - annual averages.
Accuracy:
- A: Locations: Percent of reported crash locations that automatically code (without human intervention) to the appropriate state locating coding system (LRS, Rte # & milepoint, coordinate-based, whatever the appropriate system is). If applicable, the percentages of successful location coding can be provided separately for state- maintained (on-system) and local (off-system) roadways. Additionally, if the state tracks both initial automated location coding and a final percent coded successfully (with human intervention/correction), reporting both numbers is welcome.
- B: Form-level reject rate: Percent of crash reports rejected (i.e., failing to pass mandatory–fatal error–edit checks) upon initial submission. NOTE: this may not be relevant for your system or may only be relevant to those reports submitted on paper (if any).
- C: Specific Field Edits and Logic Checks:
- Agreement between time of day and lighting condition (e.g., percent of reports where Lighting_Condition and Time_of_Day are not in agreement.
- Agreement between number of vehicles/units & number of drivers (adjusting count for hit & run or recorded phantom vehicles) recorded
- Agreement between persons injured/killed in the person records and total injured/killed as reported at the crash level (this may be a calculated field in which case this analysis is moot)
- Agreement between highest severity level injury at the person level (individuals) and the crash-level injury severity (may also be a calculated field, in which case this analysis is moot)
- Results of any other data cleansing analysis/accuracy metric that are run routinely.
- VIN (if recorded, what percent decode to a valid vehicle year, make & model)
Completeness:
- A: Narrative/Diagram: Percent of reports received lacking either the narrative, the diagram, or both.
- B: Location: Percent of reports missing location information (e.g., GPS coordinates if required to report, or information such as on street name/identifier, nearest/intersecting street name/identifier, distance from street/landmark, or other required location information).
- C: Percent of other key fields left blank: Making allowances for hit & run or phantom vehicles (if recorded), look for unexplained blanks/missing data in select fields such as:
- Driver name/address
- VIN
- Harmful Event
- Environment, Vehicle & Driver contributing factors
- EMS data (if recorded)
- D: Under-reporting:
- Comparing across the most recent three complete years of data (e.g., 2007, 2008, and 2009), are there any law enforcement agencies that have seen a significant drop in crash report submissions that is (for example) double the year-to-year drop for the state as a whole?
- Are there any medium-to-large agencies (in terms of annual submissions historically) that have turned in either zero or an uncharacteristically low number of crash reports for 2010 YTD?
Consistency:
- A: Calculate the ratios to compare across years:
- Percent PDO crashes of total crashes
- Percent Injury crashes of total crashes
- Percent Fatal crashes of total crashes
The above should be calculated overall for the state and individually for each law enforcement agency–large agencies that are significantly at variance with statewide proportions may indicate a lack of uniform application of the reporting criteria.
Data Integration and Accessibility:
- If you have information on something like a CODES data linkage project (proportion of records linked and the strength of the matching achieved) for integration, or the percentage of linkage/matching you have when linking the crash database with roadway inventory or other traffic records databases. For accessibility, perhaps there is a record of number of requests, analyses run, web-enabled data analysis page hits, or any other indicator you might have.
