|

H. W. Heinrich and the Biggest Scientific Error in Road Safety Research History

Hours Posted on Hours Updated on

Author(s): Knipling, af Whlberg

Slidedeck Presentation:

a Pres CARSP Knipling et al 5-23-22

Abstract:

Background:

H. W. Heinrich was an influential mid-20th century industrial safety engineer. He espoused a theory of “identical causal mechanisms,” whereby serious accidents, minor ones, and even no-injury operator errors all had identical or highly similar causes. The theory led to Behavior-Based Safety and, using today’s technology, Onboard Safety Monitoring.

Heinrich never studied traffic crashes, but his theory was the explicit basis for today’s Naturalistic Driving (ND) Mixed-Safety Critical Event (SCE) method, a dominant research paradigm over recent decades. The method captures and aggregates mostly non-crash avoidance maneuvers (e.g., hard-brakings, swerves) to create databases ostensibly causally representative of harmful crashes. A further assumption, never proven, is that the mix of SCEs is representative of the mix of crashes.

Although ND captures mostly non-crashes, the SHRP2 database was large enough to capture 1,400+ impacts. Only ~30% of SHRP2 crashes were police-reportable; the rest were sensor-detected. Heinrich would have said that a mixture of mostly inconsequential crashes informs us about the causes of those causing harm. Yet, for extrapolations to higher-severity and to diverse crash scenarios to be valid, crashes must be homogeneous both “vertically” and “horizontally” in regard to their causes.

Aims:

This paper debunks the theory of identical causal mechanisms and the Naturalistic Driving Mixed-SCE method. The theory is scientifically groundless and easily disproven. In its place, we posit the Heterogeneity Principle. Crash characteristics and causes are pervasively heterogeneous both vertically by severity and horizontally by scenario type.

Methods:

Primary methods have included literature reviews and direct analysis of U.S. crash databases, including crash causation databases.

Results:

  • Crashes are pervasively heterogeneous in regard to when, where, how, and why they happen. Crash epidemiology is complex and rich for study.
  • Extrapolations from one category of crashes (or from non-crashes) to other categories are likely to be spurious. This is true vertically by severity and horizontally by type (e.g., rear-end, road departure).
  • Human harm resides predominantly at the highest severity levels. Datasets dominated by low-severity and zero-severity events have weak relevance to harmful human consequences.
  • The causal robustness of different crash categories is most striking when one compares single-vehicle and multi-vehicle crash involvements. They are each robust in their causal profiles, but are dissimilar to each other.
  • Excess minor single-vehicle crashes in SHRP2 distort interpretations vis--vis significant crashes.
  • Another schism, obvious but often ignored, is that between at-fault and not-at-fault crash involvements.

Discussion:

The ND Mixed-SCE method was a wrong turn scientifically and has had adverse societal effects. For example, the U.S. DOT implemented a mandatory truck driver break rule in 2011 which was based on spurious Mixed-SCE data.

Conclusions:

Mixed-SCE datasets are created by researchers, not sampled from target populations. Empirical science is the systematic observation of target phenomena. In road safety, our target should be harmful crashes. Methods must be directly relevant to harmful crashes or, if the claimed relevance is indirect, they must be put through validity tests. The Mixed-SCE method fails these tests. Debunking the method defends the rigor of our science and highlights the diversity and rich epidemiology of motor vehicle crashes.