01 Epidemiology and crash data Flashcards
Characterize various levels of crash data
- Source of
information - Sampling area
- Level of detail
Describe the
process of in-depth crash data collection
Data collection
1. Crash notification
2. Accident scene inspection
3. Interview
4. Vehicale inspections
5. Medical records
Crash Analysis
* Injury Analysis
* Reconstruction
* Pre-Crash Analysis
Describe the role of exposure data in the analysis of crash data
Definition: Exposure data is a measure of the extent to
which road users are exposed to risk of crashes/injuries
”large” country -> ”more” traffic -> more
fatalities
Specify different measures of exposure
The number of accidents / injuries / fatalities in a region
are usually normalized by the following measures:
* Population (Good availability / limited representation)
* Number of registered vehicales (medium/ medium)
* Vehicale kilometers travelled (limited availability / good representation)
Confounding
Confounding: a variable
affects both the
analyzed factor and the
outcome
Stratification
- Stratification is a method to control for confounding
- Divide the subjects and results into
homogeneous groups (strata) with respect to suspected confounding variables
Validity
A variable is a valid measure of a
property if it is relevant and appropriate
as a representation of that property.
Reliability
A measurement process is reliable if repeated measurements on the same unit give the same (or approximately the same) results.
The reliability is dependent on the precision (absence of random errors) and the total number of cases (N).
Bias
Bias is a systematic over- or underestimation of the population parameter (i.e., consistent, repeated divergence in the same direction)
Precision
How clustered are the values
of the sample statistic
Risk
Let p be the injury risk, and estimate p as the share (or proportion) of the
population sustaining the injury:
* p(dach) = (total number of occupants sustaining the injury) / (total number of occupants in the situation concerned (N))
The point estimate p(dach) is our best guess of the injury risk based on the observed data.
Confidence interval
A confidence interval is a range of values, derived from sample data, that is likely to contain the true population parameter (such as a mean or proportion) with a specified level of confidence, typically expressed as a percentage (e.g., 95%).
How crash data can be applied in the vehicle safety
development process
- Identifying Priorities
- Safety System Development
- Evaluating Effectiveness
- Risk and Injury Analysis
- Benchmarking and Compliance
- System Testing and Simulation
- Feedback for Continuous Improvement
Source (5) of
information crash data (pro/con)
-
Police-reported
+ All types and severities
+ ”Representative” of general
crash population - Few vehicle-specific details
- No detailed injury data
-
insurance claims
+ Large dataset
+ Includes injury follow-up - Limited data access
- Potential bias: geographical
constraints, self-reporting -
In-depth crash investigation
+ Detailed data on vehicle,
occupants, infrastructure
+ Crash causation, injury
causation information - Small sample size
- Limited data access
-
Event Data Recorder (EDR) data
+ Reliable kinematic data - Needs to be combined with
other information -
Naturalistic driving data
+ Video recording
+ Kinematic data - Small sample size
- Little information on injury
Classification by area of data collection
- Multi-national databases
- National crash data
- Local crash data
*
Classification by data level (3)
Three levels of crash data
* Macroscopic level Police (lots but undetailed)
* Intermediate level Insurance
* Microscopic level Research (detailed but few)
Epidemiology
Epi - among
demos - people
logos - science
Simpson’s paradox
- Is airbag activation causing KSI (killed or seriously injured)?
- While airbag activation results in a worse KSI rate overall, it has better KSI rate in both low and high speed crashes
Confidence interval calculation
e ^ (ln(OR)+- wurzel(1/a+1/b+1/c+1/d)*1,96)
Calculation of injury reducing effect (effectiveness)
The injury reducing effect e of a safety system (e.g., restraint use) is defined as the injury risk reduction attributable to the system, given as a percentage of the injury risk without the system:
e = risk(without) - risk(with) / risk(without)
