National Academies Press: OpenBook

Hazardous Materials Transportation Incident Data for Root Cause Analysis (2009)

Chapter: Chapter 4 - Database Analysis

« Previous: Chapter 3 - Summary of Interviews with Carriers, Shippers, and Database Managers
Page 32
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 32
Page 33
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 33
Page 34
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 34
Page 35
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 35
Page 36
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 36
Page 37
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 37
Page 38
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 38
Page 39
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 39
Page 40
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 40
Page 41
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 41
Page 42
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 42
Page 43
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 43
Page 44
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 44
Page 45
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 45
Page 46
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 46
Page 47
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 47
Page 48
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 48
Page 49
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 49
Page 50
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 50
Page 51
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 51
Page 52
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 52
Page 53
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 53
Page 54
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 54
Page 55
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 55
Page 56
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 56
Page 57
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 57
Page 58
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 58
Page 59
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 59
Page 60
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 60
Page 61
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 61
Page 62
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 62
Page 63
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 63
Page 64
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 64
Page 65
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 65
Page 66
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 66
Page 67
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 67
Page 68
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 68
Page 69
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 69
Page 70
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 70
Page 71
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 71
Page 72
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 72
Page 73
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 73
Page 74
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 74
Page 75
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 75
Page 76
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 76
Page 77
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 77
Page 78
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 78
Page 79
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 79
Page 80
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 80
Page 81
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 81
Page 82
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 82
Page 83
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 83
Page 84
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 84
Page 85
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 85
Page 86
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 86
Page 87
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 87
Page 88
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 88
Page 89
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 89
Page 90
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 90
Page 91
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 91
Page 92
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 92
Page 93
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 93
Page 94
Suggested Citation:"Chapter 4 - Database Analysis." National Academies of Sciences, Engineering, and Medicine. 2009. Hazardous Materials Transportation Incident Data for Root Cause Analysis. Washington, DC: The National Academies Press. doi: 10.17226/14336.
×
Page 94

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

This section analyzes the major crash databases and identifies fields that can provide answers to “why” questions in determining root cause. Each of the following major databases is included in this discussion: • Motor Carrier Management Information System (MCMIS) • Hazardous Materials Incident Reporting System (HMIRS) • Fatality Analysis Reporting System (FARS) • Trucks Involved in Fatal Accidents (TIFA) • Large Truck Crash Causation Study (LTCCS) • Railroad Accident/Incident Reporting System (RAIRS) • Marine Information for Safety and Law Enforcement (MISLE) In addition, the NTSB and Hazardous Materials Serious Crash Analysis: Phase 2 (Battelle 2005) approaches to accident analysis are described. 4.1 Motor Carrier Management Information System (MCMIS) 4.1.1 MCMIS Database Description MCMIS includes four major files named Registration, Crash, Inspection, and Company Safety Profile. For this project, the Registration and Crash files are the most relevant. Although the focus of this analysis will be on the Crash file, when trying to identify the contributing and root causes of accidents, information in the Registration file can provide useful supplemental information. Specif- ically, the Registration file has carrier information on the number and configuration of vehicles, the number of drivers, annual miles driven, and accident rates. Such information is needed to determine the extent to which a class of accidents with similar contributing or root causes might be occurring annually. Vehicle miles traveled (VMT) helps estimate the risk of occurrence by pro- viding a measure of exposure. Identifying the risk enables officials to target resources for reducing the number of accidents. The design of the Crash file was developed in 1992 to record information on serious accidents involving a truck, bus, or light vehicle transporting hazmat. The types of information collected on each heavy truck or bus involved in these serious accidents has changed little since it was developed. The process of reporting serious accidents begins with a law enforcement official fill- ing out a police accident report (PAR). The state agency responsible for filing the MCMIS crash report screens the PARs to identify serious heavy truck and bus accidents. Once the state report- ing agency finds an accident that meets the requirements for reporting the incident to FMCSA, the information for the vehicle from the PAR is coded into the MCMIS Crash file format and 32 C H A P T E R 4 Database Analysis

electronically transferred to FMCSA or manually entered by the state agency for MCMIS Crash file inclusion. The MCMIS dataset records are at the vehicle level, as opposed to the accident level. There- fore, if multiple trucks or buses are involved in the same crash, there will be one MCMIS crash record for each truck and bus. The vehicle focus means that there will always be more records than accidents. Information coded into the MCMIS Crash file and obtainable from the PAR includes the date, time, and location of the accident; a description of the vehicle; the name and address of the carrier; personal information, including driver licensure; parameters that describe the roadway at the accident location; and the accident impacts. Because the database contains driver information—name, address, sex, and birth date—the full MCMIS Crash file is publicly available but only after data that can identify individuals have been stripped out. However, since driver information was used in this project for various purposes, the project team signed a con- fidentiality agreement and obtained a version of the file that included the driver fields. 4.1.2 Location and Ownership of Data The MCMIS Registration and Crash files are maintained by FMCSA and available only to authorized governmental users on a routine basis. 4.1.3 Database Format Initially, the MCMIS Crash file contained only one table, but recently it has been divided into the four tables (CRASH_MASTER, CRASH_CARRIER, CRASH_DRIVER, and CRASH_EVENT) shown in Figure 4-1. Database Analysis 33 Figure 4-1. Tabular relationship in MCMIS Crash file.

The CRASH_MASTER Table contains information on the date and location of the crash, road- way type, road surface condition, weather conditions, and vehicle configuration. It also contains information indicating if a hazmat release occurred, the number of fatalities, and serious injuries (those requiring transport to a medical facility for treatment). Since the state is required to pro- vide a report for trucks and buses involved in either intrastate or interstate commerce, this table has two yes and no fields to flag FEDERAL_REPORTABLE and STATE_REPORTABLE crashes. The CRASH_MASTER Table has five fields that are used when hazardous materials are being carried by the vehicle involved in the crash. The CRASH_CARRIER Table lists the carrier name and address information. The CRASH_ DRIVER Table provides the name, address, sex, and birth date of the driver. The table also con- tains licensure information. The CRASH_EVENT Table contains information on the accident sequence, beginning with the first dangerous event (e.g., crossing the median). These events do not consider pre-crash conditions such as speeding, slippery road conditions, poor visibility, etc., which are captured in other fields. There is a one-to-one relationship between the records in the CRASH_CARRIER, CRASH_DRIVER, and CRASH_MASTER tables. Although any number of events can be entered under a single vehicle incident listed in the CRASH_MASTER Table, for the majority of records, only one event is listed. In 2005, the maximum number of events asso- ciated with a single vehicle crash was four. (In the original structure of the data, up to four events could be coded.) 4.1.4 Threshold for Exclusion or Inclusion The MCMIS Crash file was developed to capture vehicle data on all serious crashes of trucks and buses involved in commerce. A crash is considered serious if there is a fatality, an injury requiring prompt medical attention at a facility away from the accident location, or if one of the vehicles involved in the crash had to be towed from the scene due to disabling damage. Since the database reports vehicle involvements, if two qualifying vehicles are involved in an accident, there will be two records, one for each qualifying vehicle. 4.1.5 Years of Coverage DOT started to provide funding to states to report serious heavy truck and bus accidents in 1992. In the early years, the underreporting of accidents was high. Although it has improved in recent years, the non-reporting rate is currently estimated to be about 20%. Some states have a much higher reporting rate than others. 4.1.6 Types of Fields Covered The CRASH_MASTER Table contains information about the location of the crash, the vehicle configuration, the type of roadway, and the weather at the time of the crash. It also contains the DOT number of the carrier and crash consequences (number of fatalities, injuries, and if it was a tow-away accident). Most importantly to this project, this table contains the following fields: • VEHICLE_HAZMAT_PLACARD, • VEHICLE_HAZMAT_NUMBER, • VEHICLE_HAZMAT_MATERIAL, • VEHICLE_HAZMAT_CLASS_ID, and • HAZMAT_RELEASED. The first and last fields are yes/no flags. The intent of the middle three is to provide the UN number, the name of the hazardous material, and the one-digit hazard class number, respectively. The intent of these three fields is to provide information consistent with the 34 Hazardous Materials Transportation Incident Data for Root Cause Analysis

172.101 Hazardous Materials Table in 49 CFR Part 172. There is no requirement that the name of the material be taken from the Part 172 Table and, as a result, the trade name of the hazardous material is often entered. The CRASH_CARRIER Table contains information on the carrier and the CRASH_DRIVER Table contains information on the driver of the vehicle described in the CRASH_MASTER Table. The CRASH_EVENT Table can have any number of events for each entry in the CRASH_MASTER Table. This is where the event sequence for the accident is specified. Prior to the restructuring that occurred in about 2003, four separate fields named EVENT_1 EVENT_2, EVENT_3, and EVENT were provided for the reporter to describe the accident sequence following the initiat- ing event. The event codes are shown in Table 4-1. For example, if the accident sequence was a collision with a motor vehicle in transit, followed by the vehicle running off the road, followed by the vehicle rolling over, there would be three EVENT_ID records entered and the entries for SEQ_NO 1, 2, and 3 would be 13, 1, and 3, respectively. Note that these are all events and not conditions that would be useful for identifying contribut- ing causes of the accident. Conditions would have to be inferred from the CRASH_MASTER Table in fields such as WEATHER, ROAD_CONDITION, and LIGHT_CONDITION. Although there is an accident description narrative in all PARs, this narrative must be captured in code numbers in the CRASH_MASTER and EVENT tables. 4.1.7 Database Purpose and Function FMCSA is responsible for ensuring the safety of commercial interstate motor vehicle truck and bus safety. In order to ensure safety, the MCMIS Crash file has been developed to provide data on the number of serious truck and bus crashes that are occurring each year. From such information, FMCSA can monitor trends, evaluate the effectiveness of current regulations, and Database Analysis 35 Table 4-1. Event codes and descriptions. Event Codes Event Description 1 Non collision ran off road 2 Non collision jackknife 3 Non collision overturn (rollover) 4 Non collision downhill runaway 5 Non collision cargo loss or shift 6 Non collision explosion or fire 7 Non collision separation of units 8 Non collision cross median/centerline 9 Non collision equipment failure (brake failure, blown tire, etc.) 10 Non collision other 11 Non collision unknown 12 Collision involving pedestrian 13 Collision involving motor vehicle in transit 14 Collision involving a parked motor vehicle 15 Collision involving a train 16 Collision involving pedalcycle 17 Collision involving animal 18 Collision involving fixed object 19 Collision with work zone maintenance equipment 20 Collision with other movable object 21 Collision with unknown movable object 98 Other

provide information that can be used to develop more effective regulations. Note that the data- base was never designed to identify contributing and root causes of accidents. 4.1.8 Data Collection Data collection for MCMIS is a complex process that begins with the police officer filling out and filing a PAR. These reports are compiled and sent to the appropriate state agency, where a determination is made of whether an accident involving a truck, light vehicle, or bus also met the crash severity definition and should be reported to FMCSA. For it to be classified as serious—and therefore reportable—the crash must either have resulted in a fatality, required that someone be transported to a remote facility for emergency medical treatment, or required that one of the vehi- cles involved in the crash had to be towed from the scene. Once it has been determined that the accident should be reported to FMCSA, the information is transcribed from the PAR either into an electronic file that is transmitted to FMCSA or manually entered through a Web interface. FMCSA then performs certain checks and enters the data into the MCMIS Crash database. Complexity arises not only from the process of going from the PAR to the MCMIS Crash form, but also from the number of agencies and individuals involved in the reporting. A query of the indi- viduals or organizations filling out the PARs in 2005 totaled over 60,000 entries for approximately 145,000 crash records entered (only about 2% of these are hazmat crashes). The exact number of individuals filling out the form could not be determined because many of the entries were orga- nization names, not the names of the individuals filling out the form. If any one of the thousands of police officers filling out a PAR fails to record the value for a parameter or any of the state agency staff fail to transcribe a parameter value or transcribe it incorrectly, the data submitted in the MCMIS Crash file is incomplete or inaccurate. 4.1.9 Data Compilation Over the years, efforts have been made to develop some standardization in the format used by each state and territory for its PARs. Although there have been some successes, there are still vast differences among the forms. Over time, there also has been an effort to keep the types of informa- tion reported in the Crash file consistent. Although the MCMIS file has been segmented into sev- eral tables from the one table used initially, the information requested has remained the same. This has enabled the states and territories to develop a standard protocol for translating the data in their PARs into the MCMIS format. Essentially, there remain 56 different translation protocols since each state or territory has a slightly different PAR. Although the MCMIS report file prepared by the state or territory is electronically transmitted to FMCSA, there appears to be little automation on the front end of the process. In the majority of the states and territories, the PARs are prepared by hand by the police officers and the translation of the PAR information into the MCMIS electronic file structure is also a manual operation. If the police officer filling out the PAR uses an abbreviated notation for the name of the carrier, unless the person transcribing the data into the MCMIS Crash file format realizes it is a shortened carrier name, the spelling of the carrier name listed in the PAR becomes the carrier name listed in the MCMIS Crash file. 4.1.10 Accuracy and Completeness of Data Studies have shown that the complicated process of filling out PARs, identifying those truck and bus accidents that meet the definition of serious accidents, and then entering the data from the PARs into the MCMIS format is neither complete nor accurate. Over the last five years, the University of Michigan Transportation Research Institute (UMTRI) has been under contract to FMCSA to assess the accuracy and completeness of the MCMIS crash reporting system on a 36 Hazardous Materials Transportation Incident Data for Root Cause Analysis

state-by-state basis. To date, 27 states have been evaluated. It is known that states are taking the evaluations into consideration in improving their systems and have instituted changes to correct the problems identified by UMTRI. The UMTRI program is just one facet of a comprehensive pro- gram at FMCSA to assist the states in improving their accident reporting. In general, significant progress has been made, but the completeness and accuracy of the data in MCMIS remains a seri- ous issue. Moreover, there are many fields in the PAR that are either not filled out or not translated into the MCMIS format, and when fields that must be used to identify serious accidents are left blank or inaccurately filled out, then either serious accidents that should be reported are not reported or accidents that are not serious are filed because the inaccurate information provided makes them appear serious. Additional analyses reflecting on the accuracy and completeness of the data in the MCMIS Crash file are found in Appendix C (available on the TRB website at www.TRB.org by searching for HMCRP Report 1). 4.1.11 Identification of Hazmat Incidents in MCMIS Table 4-2 lists some of the key parameters recorded in MCMIS and the percentage of the entries for which no information is presented in calendar year (CY) 2005. Overall, there is about a 20% underreporting rate. As shown in Table 4-2, the percentage of blanks in the MCMIS Crash file tables varies from zero to about 30%. For fields like FATALITIES and INJURIES, there are no blank entries because zero is entered if there were no injuries or fatalities. Similarly, there are no blank entries for Y/N fields such as TOW AWAY. There is one parameter, DRIVER CONDITION CODE, which is left blank in all crash records after CY 2001. In 2001, the field was coded as “driver appeared normal” for about 94% of the crashes. It is the other 6% of the crashes where the contributing cause, or even the root cause, might have been “driver condition.” Since there is a location on the PAR to enter this information and since a police officer is trained to observe a person’s behavior, some weight can be assigned to the officer’s opinion regarding the driver’s suitability to be operating a motor vehicle. Since this is the Database Analysis 37 Parameter Name Percentage Blank Carrier Name Provided for Each Incident 0% Fatalities 0% Injuries 0% Tow Away 0% County Code 1% Driver Name Provided for Each Vehicle 2% Vehicle Configuration 4% Weather 7% Light Condition 8% Road Surface Condition 8% Cargo Body 9% No Event Sequence Provided for Each Incident 16% Vehicle Identification Number 17% Traffic Way 18% Access Control 23% Accident Location Adequate 23% Vehicle Hazmat (Y/N) 24% GVWR 26% DOT Number 31% Table 4-2. Percentage of entries blank by parameter name.

only driver condition parameter included in the MCMIS Crash file, leaving this parameter blank is considered a significant loss. For the remaining parameters that have less than complete coverage, either the PAR did not pro- vide the value or the state staff person translating the information into the MCMIS crash format did not enter a value for the parameter. In Hazardous Materials Serious Crash Analysis: Phase 2 (Battelle 2005), PARs were obtained for all vehicles that were believed to be transporting hazardous materials in 2001, over 1,800 PARs. For almost all the parameter fields that were left blank in the MCMIS crashes in 2001, it was possible to fill in missing parameter values from the PARs. There were some notable exceptions. One state did not provide the commercial vehicle truck supplement to the PAR and, without that, it was impossible to fill in the vehicle configuration and gross vehi- cle weight rating (GVWR). That supplement also contained the hazmat data. If the staff person has access to all the PAR data, all the blank percentages would probably be less than 10%. The reasons for the information being lost are unknown. The last four entries in Table 4-2 provide some statistics regarding how well the informa- tion is coded into MCMIS. The MCMIS Catalog, published on the FMCSA website, provides no standard regarding how to fill out the location field. The information provided is sufficient to locate the accident on a map only 23% of the time. These entries typically occurred at inter- sections or the location was specified as a route name or number along with the number of the nearest milepost. There are a variety of reasons for this low percentage. In many cases, only the route name is given and if the state and county are given, which is normally the case, then the best that can be done is to locate the accident on a route somewhere in a county through which the route passes. Evidently, at some point in the translation between the PAR and the MCMIS Crash file, a 30-character limit was imposed on the location field. There were numer- ous cases where the entry stopped mid-word and, as a result, truncated the milepost informa- tion needed to locate the accident on the route. A comparison of the entries in the CRASH_EVENT and CRASH_MASTER Tables reveals that no crash event is provided for 16% of the crashes. Of the remaining 84% of the crashes with event sequences, 57% list one event, 13% list two events, 5% list three events, and 9% list four events. Based on these percentages, for 16% of the crashes, it is impossible to even identify the type of crash. For slightly more than one-half of the crashes, one event sequence is provided. Based on the statistics shown in Table 4-3, the use of one event to define the acci- dent often can be justified. The dominant single-event sequence accident in which a truck is involved is coded as EVENT_ID=13, “collision involving motor vehicle in transit.” This is the event code for 82% of the vehicle incidents with a single event listed in the CRASH_EVENT Table. The next most likely entry is “other,” occurring 4% of the time. Table 4-3 lists, in order of decreasing percentages, the name of the single event followed by the percentage of crash records and the number of crash records listing this single event to describe the accident sequence. In going down the list shown in Table 4-3, there are several significant features. If the number of single events that fall into the equipment failure category were totaled, although individually they each show a zero percentage, the combined number would be above 1%. There also are a fair number that could be better described using more than one sequence. For example, it is highly unlikely that striking an animal would cause sufficient damage to the vehicle to result in an injury to one of the truck occupants or result in the truck being towed from the scene, mak- ing it a serious accident that would be reported to MCMIS. Did striking the animal result in a jackknife or the truck running off the road or overturning? A two-element event sequence prob- ably should have been used for that class of accidents. Overall, the biggest concern remains underreporting, closely followed by failure to enter parameter values for a significant fraction of the records. 38 Hazardous Materials Transportation Incident Data for Root Cause Analysis

4.1.11.1 Overall Reporting Rating for the States Table 4-4 provides a summary of state reporting rates for different factors for those states that UMTRI has evaluated. The rates are recorded as the percent of reportable cases that were actu- ally reported, by crash severity (in the MCMIS scale). Table 4-4 shows that reporting rates have tended to improve in recent years and that the lowest rates are associated with earlier years. For completeness, UMTRI has compiled the percentage of missing data for the primary vari- ables. They also compare variables as reported in MCMIS to how they appear in the state file. Unfortunately, comparing the values doesn’t reveal if the MCMIS information is accurate, just whether it is the same as what was reported to the state. To get a further handle on accuracy, it would be necessary to compare the information with an independent source. This is accomplished in the next section for fatal crashes using the TIFA database. In the TIFA survey, individuals are called and asked about the crash. These contacts include the driver, owner, safety director, and reporting police officer. 4.1.11.2 Comparing MCMIS with TIFA to Evaluate Crash Data Accuracy The TIFA file maintained by UMTRI provides a unique opportunity to evaluate the accuracy of the data reported to the MCMIS Crash file. TIFA data are collected independently from the MCMIS Crash file data, using a different methodology. MCMIS Crash file data are extracted by the states from their crash data and uploaded to the MCMIS Crash file using the SafetyNet System. Some states extract the data from a supplementary crash reporting form, while others incorporate the required data on their primary crash report. Many states use a computer algo- rithm to identify reportable crashes, while others manually identify reportable crashes and extract the data. Whatever the method, the data originate with the officer responsible for filing the crash report. Database Analysis 39 Table 4-3. Crashes described by a single event. Event Description Percentage of Single Events Number of Crash Records Collision involving motor vehicle in transit 82% 73,591 Other 4% 3,358 Collision involving a parked motor vehicle 3% 2,462 Collision involving fixed object 3% 2,424 Non collision overturn (rollover) 2% 2,147 Collision with other movable object* 1% 1,074 Non collision other 1% 901 Non collision ran off road 1% 792 Collision involving pedestrian 1% 767 Collision involving animal 1% 606 Non collision explosion or fire 0% 352 Non collision jackknife 0% 308 Non collision cargo loss or shift 0% 299 Non collision equipment failure (brake failure, blown tire, etc.) 0% 254 Collision involving pedalcycle 0% 237 Collision involving a train 0% 134 Non collision separation of units 0% 96 Collision with unknown movable object 0% 70 Non collision downhill runaway 0% 44 Collision with work zone maintenance equipment 0% 35 Non collision unknown 0% 33 Non collision cross median/centerline 0% 31 *Previously “collision involving other object.”

In contrast, the TIFA protocol uses state crash reports, but primarily to identify persons with knowledge of the crash for a telephone interview. Interviewers typically contact the driver, owner, or safety director of the carrier for a detailed interview about the truck, driver, and carrier that operated the truck. If the driver or carrier can not be contacted or refuses to cooperate, the inter- viewer will contact the reporting officer, tow operator, or other witness. But the great majority of interview information comes from sources that actually operated the truck. In addition, all sur- vey information is reviewed by experienced editors who decode the vehicle identification num- ber (VIN), look up the manufacturer’s original specifications for the vehicle to compare with interview information, and also consult a library of information on typical cargoes, trailers, and carrier operations. Information that is ambiguous or unusual is clarified by return calls to the orig- inal respondent. Rarely, if no other information is available, some limited descriptive informa- tion may be coded from the police report. But the overwhelming majority of information in the TIFA file is collected directly from the vehicle operators. Because TIFA data are assembled by a completely separate process and entity, the TIFA file can serve as a relatively independent view of trucks involved in traffic crashes that are reported to the MCMIS Crash file. We say relatively because there are occasional instances where certain information may be coded from a police report. But there are good reasons for regarding TIFA data as reasonably accurate. Most cases have more than one source for the data (unless all infor- mation can be collected from a single source). Any ambiguous or contradictory information is clarified by further calls. TIFA editors have over 20 years of experience in working with large trucks, and they are knowledgeable about the variety of vehicles and operations. Finally, all cases are checked using computer algorithms for consistency, and to identify unusual cases for further 40 Hazardous Materials Transportation Incident Data for Root Cause Analysis Table 4-4. Reporting rates of states to MCMIS Crash file compiled from UMTRI reports. State Data Year Overall Reporting Rate Fatal Injury Tow Away Alabama 2005 76.0 91.4 76.4 75.0 Arizona 2005 78.2 93.8 83.4 75.6 Connecticut 2005 Likely <30% California 2003 72.0 84.2 73.9 70.9 Florida 2003 24.0 55.6 26.5 20.0 Georgia 2006 68.1 78.8 68.4 67.4 Idaho 2006 72.9 92.3 90.5 60.7 Illinois 2003 43.0 71.0 42.3 42.6 Indiana 2005 80.5 90.3 81.9 79.6 Iowa 2005 71.6 94.1 86.4 61.4 Louisiana 2005 56.6 79.6 57.0 54.7 Maryland 2005 31.1 84.6 56.0 15.6 Michigan 2003 73.7 92.4 73.1 73.4 Missouri 2000 60.9 76.8 63.7 58.8 Missouri 2005 83.3 94.6 84.9 81.8 Nebraska 2005 86.8 100.0 82.0 82.7 New Jersey 2003 82.5 67.4 81.5 83.2 New Mexico 2003 9.0 27.5 11.0 6.8 North Carolina 2003 48.2 63.3 49.4 47.1 Ohio 2000 38.8 50.7 58.2 28.6 Ohio 2005 42.5 85.4 52.7 32.3 Pennsylvania 2006 77.0 91.7 74.5 77.6 South Dakota 2005 66.4 78.9 64.9 66.4 Tennessee 2004 51.3 93.5 54.8 47.4 Washington 2003 37.6 to 53.7 67.2 About 40

review. The TIFA file is not free of errors, but there are multiple layers of checks to keep the error rate low. Accordingly, the TIFA file can be used to check the accuracy of fatal truck crashes in MCMIS. The TIFA file includes only fatal crash involvements, significantly limiting the number of MCMIS records that can be checked. Nevertheless, since fatal crash involvements often receive the most detailed investigations by police, the records in MCMIS for fatal involvements arguably should be the most accurate. In that sense, the results of the comparison with TIFA represents the best-case scenario. To perform the comparison, records in four years of TIFA crash data were matched to the cor- responding records in the MCMIS Crash file. Crash involvements for 2002 through 2005 were matched, producing 11,914 records for comparison. Table 4-5 shows the comparison of the identification of hazardous material in the MCMIS Crash file and the matching record in the TIFA file for the period from 2002 to 2004. In the TIFA file, the value recorded was whether the cargo body had amounts of hazardous materials suffi- cient to require a placard. In the table, cases where that information was unknown for all cargo bodies are counted as no hazardous materials. In MCMIS, the unknowns are shown as “(blank),” to indicate the value was left unknown. Valid entries for the variable in MCMIS are “Y” or “N.” The one case with an “M” is considered to be a typo in which “N” was meant. Table 4-5 shows some disconcerting patterns. Of the cases where the TIFA interview indicated the truck had placarded amounts of hazmat, the MCMIS hazmat placard was coded “Y” for 212, “N” for 172, and left blank for 50. There were also 95 cases where the hazmat placard was coded “Y” in MCMIS, but the TIFA interview showed that the truck did not have hazmat cargo. Thus, of the 434 cases in the TIFA file recorded as carrying hazmat, only 212 or 48.8% were coded as carrying hazardous material in the MCMIS file. Similarly, of the 307 cases coded as carrying haz- ardous material in the MCMIS file, only 212 or 69.1% actually had hazardous material. One possible explanation for the discrepancy could be that the MCMIS variable captures trucks showing a hazmat placard, regardless of whether the truck actually had hazmat. In fact, there were 60 cases identified in the MCMIS Crash file with hazmat placard coded “Y” for trucks that were empty at the time of the crash. Another 34 had some cargo, but the TIFA interview showed that it did not include hazmat. Comparisons also were made for several other variables. For truck cargo body and configuration, a comparison was made of the detailed code levels. The TIFA file includes more detailed types of cargo bodies and truck configurations than allowed in the MCMIS Crash file but, for the purpose here, both were aggregated to the levels permitted in MCMIS. Table 4-6 shows the distribution of Database Analysis 41 Table 4-5. Identification of hazmat cargoes in TIFA and MCMIS, 2002–2004. TIFA Code MCMIS Code N Hazmat Cargo Hazmat Placard Total N 172 Y 212 Yes (blank) 50 M 1 N 8,484 Y 95 No (blank) 2,900 Total 11,914

cargo body types from the TIFA file, the number of such body types in TIFA, and the percentage identified correctly in MCMIS. Table 4-7 makes a similar comparison for truck configuration. The primary truck types, straight trucks with no trailer and tractor-semitrailers, are identified accurately 87.5% and 75.5% of the time, respectively. Less recognizable types like straight trucks pulling a trailer or bobtail tractors are less often accurately identified in MCMIS. Finally, Table 4-8 shows the percentage of selected variables that are coded the same in TIFA and the MCMIS Crash file. Variables shown in Table 4-8 are drawn from the FARS file and not from the TIFA interview. GVWR class in MCMIS aggregates the classes to 1 to 2, 3 to 6, and 7 to 8. The variable is left unknown in 62% of the cases, so the last row of the table shows the accuracy of the variable in MCMIS excluding unknowns. 4.1.12 Quality Control Process The MCMIS reporting methodology presents a difficult quality control process. First, there are a large number of jurisdictions filling out PARs that vary from state to state. Although many reporting agencies do not break down the reporting to the officer’s level by providing a badge number, the 151,000 reports filed in 2005 were filled out by more than 61,000 agencies or indi- vidual officers. This means that, on average, a police officer from a specific agency might fill out less than three truck PARs in a given year. Assuming there are about 3,000 placarded shipments involved in crashes each year, the probability that a police officer will have to fill out a PAR for a placarded truck is on average, less than once every 20 years (60,000/3,000). This poses a signif- icant training problem if the officer will be filling out the hazmat supplement only a few times in his or her career. Requiring or sponsoring a formal training program in 50 states for an event that occurs a few times in an officer’s career is probably not cost effective. Providing a guide to 42 Hazardous Materials Transportation Incident Data for Root Cause Analysis Cargo Body Type Frequency % Correct No cargo body (e.g., bobtail) 251 49.0 Van 5,432 71.9 Flatbed 1,713 70.1 Tank 1,011 77.7 Auto carrier 77 68.8 Dump 1,947 65.5 Refuse 288 64.9 Table 4-6. Proportion of cargo body type in MCMIS coded correctly, based on comparison with TIFA data. Truck Configuration Frequency % Correct Straight truck 2,839 87.5 Straight truck plus trailer 373 42.9 Other straight truck 8 75.0 Bobtail tractor 175 61.7 Tractor-semitrailer 7,956 75.5 Tractor doubles 439 76.1 Table 4-7. Proportion of truck configuration in MCMIS coded correctly, based on comparison with TIFA data.

filling out the state PAR and the MCMIS crash form would be more appropriate and is the avenue recommended for consideration. Although FMCSA has an elaborate training program, it does not focus on hazmat crashes. Because of unique characteristics of hazmat crashes, train- ing should include special attention for this category of cargo. Tests to evaluate the effectiveness of the current (quality assurance) Q/A process are challeng- ing because of the number of agencies providing data. Each agency likely has its own Q/A process, so queries at the national level contain considerable uncertainty regarding their quality. A few years ago, when one compared carrier names and DOT numbers, the agreement was poor and the number of ways the carrier name was presented ran for pages, particularly for a carrier with a name that can be easily misspelled or mistyped, especially considering that most of the entries are being entered from a handwritten PAR. Fortunately, many of these previously observed problems are no longer present. The overall Q/A process was checked selecting a single-truck carrier, “Schneider Trucking.” This company was picked because in 2001, the number of ways this company’s name could be spelled ran on for pages. Within the umbrella of “Schneider Trucking,” the company appears to be operating under three DOT numbers, “Schneider Specialty Carriers,” “Schneider Bulk Car- riers,” and “Schneider National Carriers Inc.” Although there are slight variations in the names and addresses for the divisions, for the DOT number reported for Schneider Specialty Carriers, all the variations are in the name and address and none of the names or addresses are the same as those reported for the other two Schneider divisions. While the Q/A could be improved so only one name and address was recorded for each of the Schneider divisions, the variations are not considered a major impediment toward carrier-specific analysis since the DOT number appears to always be reported correctly. Since it is known that the source of the variation in the name and address starts when the carrier’s name and address are handwritten on the PAR, the only way the variation could be eliminated would be to not faithfully record the information in the PAR but instead refer back to a pick list taken from the MCMIS Carrier Registration file where the DOT number was assigned. Since this is not a serious analysis impediment, improve- ment in other areas, such as reducing the number of blank entries and ensuring the consistency among current entries, would be more cost effective. 4.1.13 Interconnectivity with Other Databases The MCMIS Crash file could be connected with HMIRS, TIFA, and—for grade-crossing accidents—RAIRS. The date and state where the incident occurred, along with the carrier’s DOT number, provides a way to link accidents reported in the other databases. An attempt was made Database Analysis 43 Variable % Correct Interstate carrier 83.8 Number of vehicles in crash 85.2 Number of fatalities 99.2 Light condition 89.7 Road surface condition 91.0 Weather 84.0 Trafficway flow 41.0 GVWR class 34.8 GVWR class where known 91.5 Table 4-8. Proportion of selected variables in MCMIS coded correctly, based on comparison with TIFA data.

to couple MCMIS with the LTCCS, but the coder of the accident data in the causation study have obscured any information that can be used to link the data in the LTCCS to the other databases. The data are only given by month and year, no state or day of the month is provided. The VIN number of the vehicle has been truncated, so it is difficult to match the numbers shown in the LTCCS with the numbers in MCMIS. No other common fields could be found. One useful improvement that FMCSA could make to the MCMIS Crash file that would increase the possibility of linking the file to other databases would be to restore the rule for how the REPORT_NUMBER field is constructed. Prior to 2001, the instructions to the state were to use the police report number in the REPORT_NUMBER field. That rule is no longer required, although some states clearly embed the police crash report number in the REPORT_NUMBER field. The actual police report number of a crash would permit a hard link to a specific crash, not just the probabilistic link obtained by using date, time, and geographic location. 4.1.14 Analyses Using Database The previous sections have described some of the characteristics of the MCMIS Crash data- base. Since the focus of this analysis is hazmat accidents, here the focus will be on techniques to identify accidents involving hazmat vehicles and then to show the characteristics of those crashes. The ways to join MCMIS crash records with the records from other accident databases will be discussed in this section. The results from joining two datasets when the same accident is reported in both (for example, MCMIS and HMIRS) will be summarized separately after the individual databases have been described. Hazardous Materials Serious Crash Analysis: Phase 2 (Battelle 2005), described in Chapter 2 of this report, developed a matrix listing the parameters that were believed to provide a comprehensive understanding of the accident environment. The project reviewed PARs and made telephone calls to the carrier to obtain data on those parameters. The parameters, divided into the five classifications of vehicle, driver, packaging, infrastructure, and situational are shown in Table 4-9. The unshaded boxes are not recorded in MCMIS and, for those that are recorded, the color codes show the percentage of the hazmat accidents that correspond to entries in MCMIS. When 44 Hazardous Materials Transportation Incident Data for Root Cause Analysis Table 4-9. Accident parameter coverage in MCMIS based on percentage not null. Vehicle Driver Packaging Infrastructure Situational Configuration Age Package Type Road Surface Pre-CrashCondition Cargo Body Experience Quantity Shipped Road Condition Dangerous Event GVW Condition Quantity Lost Road Type Vehicle Speed Vehicle Defect Valid License Age (Cargo Tank) Traffic Way Impact Location Vehicle Response Citation Issued Rollover Protection Access Control Primary Reason Driver Response Inspection History Speed Limit Accident Type Training Design Specification No. of Lanes Weather Condition Location Light Condition Time of Day Health Consequences Key: > 95% 50% to 95% < 50%

attempting to identify root and contributing causes of accidents, any cells in Table 4-9 that are not shaded could not be shown to be contributing or root causes of accidents. The table clearly shows that the MCMIS dataset must be supplemented considerably before it can be a good source of information for identifying root or contributing causes of hazmat truck accidents. 4.1.15 Summary and Potential Measures for Improving Root Cause Analysis The MCMIS Crash file is formed using a complex set of operations that vary from one PAR originator to another, and Crash file preparation that varies from one state to another. Although the managers of the MCMIS Crash file have made great strides in improving the quality of the data, additional improvements are required for this database to be a useful tool in an information system that is capable of identifying contributing and root causes of accidents. The single biggest improvement in MCMIS crash reporting would be MCMIS parameter fields that are completely populated. Some fields should be required to be filled out, particularly those related to the vehicle, carrier, driver, route characteristics, and point-of-contact information. Additional improvements include the following: • Require that the DRIVER_CONDITION_CODE field be filled out. In Hazardous Materials Seri- ous Crash Analysis: Phase 2 (Battelle 2005), the code “Appeared Normal” was the common entry for about 94% of the vehicle crash records. The codes for “Asleep” and “Fatigued” totaled about 3%, and the total for “Drugs or Alcohol Impairment” is about 1% and less than 1% for “Being Ill.” The DRIVER_CONDITION_CODE field is clearly valuable, especially for programs designed to improve driver performance. This is the only field that captures driver performance in MCMIS and provides a valuable indicator of whether, in the opinion of the police officer fill- ing out the PAR, the vehicle driver was truly incapacitated. • Fill all five hazmat fields completely and accurately for trucks carrying hazardous materials. Presently, in records where one or more of the fields indicates a vehicle as carrying haz- ardous materials, all five fields are completely filled out less than 15% of the time. When several of the fields are filled out, the entries are often inconsistent, making it difficult to make an accurate determination of when a truck was transporting a hazardous material, and the UN number, class, and/or name of the hazardous material actually being transported. Although it is normally possible to identify the name of the hazardous material from the data reported in the VEHICLE_HAZMAT_MATERIAL field, it should be noted that in either the recording of the information or in the electronic transmission of the data, the field is often truncated. • Enter the DOT number for all serious crashes involving hazardous materials. Currently, a DOT number is entered for only 80% of the vehicles carrying hazardous materials. A carrier transporting hazardous materials, even if not involved in interstate commerce, must register with FMCSA and be assigned a DOT number. For hazmat shipments, it should always be pos- sible to assign a DOT number. • Fill out the VEHICLE_CONFIGURATION_ID and ROAD_CONFIGURATION_ID fields. • Specify the LOCATION field in a manner that enables the accident location to be found on a map. Presently, this is the case roughly 30% of the time. Specifying the route number or street name followed by the longitude and latitude is a straight-forward way to present location information. The difficulty in identifying the accident location on a map is aggravated by trun- cation of the field occurring somewhere in the recording or record transmission process, thereby eliminating key information. Give state personnel entering the data into the MCMIS crash record system access to databases containing information such as the MCMIS Registration file and the 49 CFR Part 172 Hazardous Database Analysis 45

Material Table. Having access to this information could enable the state personnel to verify the hazmat entries and even fill in any information missing from the PAR. Linking the data entry process with these and other files so the data entry personnel could choose from “pick lists” that are narrowed down as additional characters are entered could make it easier to accurately popu- late fields in the MCMIS Crash file. The MCMIS Crash file data dictionary could be enhanced so it contains not only the def- inition of a parameter and the format for the field in the database but also the format of the data to be entered. Specifying the format in the database does not necessarily define the data entry format as evidenced by past records. A section answering some commonly asked ques- tions would be valuable as well. One question might be: If the PAR lists the carrier location as one of the carrier’s freight depots, should that address be entered in the MCMIS Crash file or should the address of the carrier’s home office, taken from the MCMIS Registration file, be entered? Another question might be related to the choice of entering a street address or a postal box number. Questions asked about the many situations that occur when filling out the LOCATION field also would be worthwhile given the different formats currently being listed in the MCMIS Crash file. For example, an Interstate route could be designated as I-70, IR70, I070, I70, or some other format. If the potential measure is adopted to use longitude and latitude when specifying a location, then the format and accuracy must also be specified. If the coordinates were expressed in decimal degrees, then specifying the longitude and lati- tude to two decimal places would place the accident on a highway, but if specified to three decimal points the location would be shown as either being on the left- or right-hand side of the right-of-way. Build data quality consistency checks into the data entry process. For example, if a number is entered into the DOT_NUMBER field that is not in the MCMIS Registration file or is inconsis- tent with the carrier’s name and address in the Registration file, then the number should be flagged and held until the correct number can be determined. If the UN number is not listed in the 49 CFR 172 Hazardous Material Table, it should not be possible to enter it into the VEHICLE_ HAZMAT_NUMBER field. 4.2 Hazardous Materials Incident Reporting System (HMIRS) HMIRS is maintained by PHMSA. In accordance with 49 CFR 171.16, all carriers of haz- ardous materials by road, rail, water, or air must fill out DOT Form F 5800.1 and submit it to PHMSA within 30 days of a reportable hazmat incident. The reportable incident could occur during loading, while in transit, during unloading, or while in temporary storage when en route between the origin and the final destination for the hazardous material. An incident is reportable if (1) the National Response Center (NRC) was notified, (2) there is an uninten- tional release of a hazardous material or the discharge of any quantity of hazardous material, (3) a cargo tank with a capacity of 1,000 gallons or greater containing any hazardous substance suffers structural damage to the lading retention system or damage that requires repair to a system intended to protect the lading system (even if there is no release of hazardous material), or (4) an undeclared hazardous material is discovered. In accordance with 49 CFR 171.15(b), NRC must be notified immediately if there is (1) an injury or fatality directly result- ing from exposure to the hazardous material, an evacuation lasting more than one hour, a major artery closed for more than one hour, or alteration of an operational flight pattern or aircraft routine; (2) a fire, breakage, or spillage of a radioactive material; (3) a fire, breakage, or spillage of an infectious substance; (4) the release of a marine pollutant; or (5) a situation exists that poses a continuous danger to life at the scene. 46 Hazardous Materials Transportation Incident Data for Root Cause Analysis

4.2.1 Database Description Prior to 2005, the HMIRS database consisted of three tables. The CON Table provided informa- tion on the incident. the MAT Table contained the name and address of the shipper, origin and des- tination address, the hazardous material being shipped, the amount released, and any damage to the packaging. The RMK Table was used for remarks. In most cases, there is one MAT entry for every CON entry. The RMK Table limits the text field to 80 characters, a legacy from the 80-character entry on an IBM card. Thus, there are often several RMK entries for each CON entry. Beginning in 2005, the HMIRS database was significantly restructured. The new database structure is shown in Figure 4-2. Figure 4-2 shows that all of the tables are related to the IREPORT Table. This table assigns a unique report number to each carrier-reported hazmat incident. IREPORT contains information on the carrier, the incident location, and impacts in terms of fatalities, injuries, and economic dam- age. It also provides contact information so that PHMSA data entry personnel can request additional data when certain fields are left blank. The IEVENT Table performs a function similar to the old RMK Table. The big difference is that after 2005, each line defines part of the event sequence. Although most IRECORDS have only one IEVENT record, about 10% have four IEVENT records. In 2005 and 2006, there were no IREPORT entries that had more than four IEVENT records. The IACTION Table was new in 2005 and gives the carrier the opportunity to identify changes that have been made to its operations as a result of the incident. An examination of the action statements demonstrates that some carriers have prepared thorough accident investigations and probably know the contributing causes and root causes of the incident. This table provides a way of identifying improvements that have been made without providing evidence of negligence that could be used in any litigation arising as a result of the incident, which is a major concern to the accident reporter. Database Analysis 47 Figure 4-2. Relationship among HMIRS tables from 2005 onward.

The SHIPPER Table provides a listing of the shippers with hazardous materials involved in the incident. If the shipment was not “exclusive use,” there could be several SHIPPER records for each IREPORT record. Similarly, a shipper could put more than one MATERIAL in the shipment, meaning that if an incident occurs, there could be several MATERIAL records for each SHIPPER record. The restructured database provides the opportunity to identify individual packages of the same material by having multiple PACKAGE records entered for a single MAT_ID record in the MATERIAL Table. Lastly, it is possible to describe where, how, and why each of the individual packages failed in the PKGFAIL Table. If there are multiple layers to the package, those can be described in the PKGLAYER Table. Since both the PKGLAYER and PKGFAIL Tables are related to the MATERIAL Table and not to each other, it is not possible to describe how the packaging layers failed in the incident. The post-2005 structure of the HMIRS database provides the opportunity to report the per- formance of individual packages in shipments carrying multiple packages of more than one kind of hazardous material from more than one shipper. Although this level of detail is not often needed or used when reporting an incident, the structure of the database permits it. In all other databases, when there are multiple classes of hazardous materials in the shipment, their struc- ture permits only one hazmat entry. The reporter must choose which hazardous material to des- ignate, and this can become a source of disagreement when the records in two different data- bases report that the vehicle contained different classes/divisions of hazardous materials. In actuality, both types of hazardous materials were present. The structure of the HMIRS database is ideally suited for examining package behavior in both the normal shipping environment and following an accident. Table 4-10 shows the total num- ber of incidents reported in 2005 and 2006 by mode and phase, and Table 4-11 shows the total number of accidents reported by mode and phase for the same two-year period. A comparison between the columns in Table 4-10 and Table 4-11 shows that all of the inci- dents reported for water and air are related to normal transport and are not related to accidents. Even with truck and rail, approximately 10% of the en route incidents are related to accidents. The focus of HMIRS is clearly not the transport accident environment. For the normal transport 48 Hazardous Materials Transportation Incident Data for Root Cause Analysis T_PHASE Air Truck Rail Water En Route 1,049 4,871 1,303 83 Loading 553 4,542 30 6 Unloading 595 19,487 70 24 En Route Storage 1,868 1,709 46 24 Table 4-10. Total number of incidents reported by phase and mode, 2005 and 2006. T_PHASE Truck Rail En Route 558 91 1 Loading 1 Unloading 13 En Route Storage 1 Table 4-11. Total number of accidents reported by phase and mode, 2005 and 2006.

environment, the root cause of spills is probably related to the handling of the packages, infor- mation that HMIRS captures very well. The focus of this project, however, is transportation acci- dents. Specifically, these include the 558 truck and 91 rail transportation accidents that occurred en route over the two-year period shown. Table 4-12 presents a summary of the HMIRS database accidents for 2005 and 2006. Table 4-12 shows eight incidents of undeclared shipments. Five of these shipments were empty cargo tank shipments and the regulations require them to be placarded even when empty. The other three contained hazardous materials and were not placarded. The new requirement to report damage to cargo tanks having a capacity of 1,000 gallons or greater when they suffer damage to the lad- ing system or its protective system even though there is no release, manifests itself in the record- ing of 33 additional truck accidents over a two-year period. 4.2.2 Purpose and Function HMIRS was developed in the early 1980s and has been maintained ever since. Its purpose has always been to provide regulatory agencies with the information they need to monitor the safety of hazmat transport, document the effectiveness of current regulations, and—if shown to be warranted—provide the data required to support new regulatory initiatives. As stated previously, the focus is on package performance. Most of the incidents reported occur during normal transport and are not related to accidents. Since this project focuses on acci- dents, the majority of the records reported in HMIRS are not referenced in the following analysis. 4.2.3 Data Collection When an HMIRS reportable incident occurs, the carrier is required to fill out DOT Form F 5800.1 and submit it to PHMSA. As follows, there are four filing methods, but most of the reports are received by the first three of these: 1. XML submissions—five carriers do this on a nightly basis and most follow up with a paper copy, 2. Online 5800.1 incident reporting application, 3. PDF—some follow up with an e-mail that includes a PDF attachment, and 4. Faxes from some filers and package carriers from other filers who deliver, on a monthly basis, the paper forms for all incidents within the past month. A carrier has 30 days in which to file a report. PHMSA does Web searches of newspapers and also receives a list of incidents in which NRC was notified. By comparing the list of incidents that have been reported by these sources, carriers that have not filed within the 30-day period are identified. Those carriers are notified by phone or mail regarding their delinquent status. Database Analysis 49 Report Type Truck Rail Incidents 521 87 Undeclared Shipments 4 4 Cargo Tanks, No Release 33 0 Total In-Transit Accidents 2005-2006 558 91 Table 4-12. Further breakdown of en route accidents, 2005 and 2006.

4.2.4 Data Compilation Reports filed by XML submission or online 5800.1 go directly into the database, which includes high-level quality control processing. For PDFs and faxes or paper forms delivered by package carriers, PHMSA scans and performs optical character recognition (OCR) for accuracy and then enters the record into the database. PHMSA also employs character-to-character checks to ensure that the OCR process translated the paper form properly. As part of the data entry process, PHMSA examines the form for personally identifiable information (PII), and such information is scrubbed from the report. For example, if the carrier name is the name of a per- son, the name is blanked out in the report. The HMIRS database is updated monthly. 4.2.5 Accuracy and Completeness of Data Aside from the inconsistencies that are discussed in Section 4.2.6, data accuracy is mostly con- trolled by the accuracy of the carrier reporting the information. The carrier has access to first- hand knowledge of the incident through personal interviews with the vehicle operator involved in the accident and repair information that documents the extent of the damage and the costs to complete the repairs. PHMSA staff, in interviews, complained that it is evident that many of the reports are reviewed by a lawyer prior to submission and they believe this review greatly weak- ens the value of the information reported. No individual or business wants to put information on the record that could be used against them in a civil suit. This is particularly true for the acci- dent reports that make up a small fraction of the HMIRS database. When questionable or incomplete information is provided, PHMSA staff contact the carrier and request additional information or clarification of the information they have received. In the case of injuries and fatalities, since HMIRS distinguishes among injuries and fatalities that are the direct result of exposure to the hazardous material and those as a result of the traffic acci- dent, all fatalities and injuries are validated by their own process to determine if they were caused by a hazardous material. This includes obtaining coroner reports and death certificates. Data compilers also look for any mention of injuries if none were marked on the form. In fiscal year 2009, PHMSA will introduce an online Incident Reporting System that will require filers to fix incorrect data before the submission will be accepted. However, since the car- riers will also be able to file the reports via the other methods available, the effectiveness of these checks will be limited to electronically submitted reports. 4.2.5.1 Comparing HMIRS with TIFA to Evaluate Crash Data Accuracy As was the case for MCMIS, the TIFA file provides a unique opportunity to evaluate the accu- racy of the crash data reported by carriers to PHMSA and entered in the HMIRS database. Although the TIFA data is collected differently, the data associated with fatal truck crashes are checked rigorously for data quality. Consequently, TIFA data is highly reliable. In this discussion, and in subsequent comparative analyses of the accuracy of HMIRS data, it is important to recognize that not every attribute, including some cause codes, in the HMIRS database is considered. Rather, those that can be compared because similar information theo- retically exists in multiple databases are used to estimate the level of accuracy of the HMIRS data that are being collected. The TIFA file provides an opportunity to evaluate the completeness of reporting of a subset of hazmat incidents. The TIFA file includes all trucks that were involved in a traffic crash that included at least one fatality. Data collected includes whether the cargo on the truck was plac- arded as hazardous material and whether the cargo spilled. The TIFA file is based on NHTSA’s FARS file, which is a comprehensive database of all fatal traffic crashes. UMTRI receives police 50 Hazardous Materials Transportation Incident Data for Root Cause Analysis

reports on all crashes involving a truck. Interviewers contact the truck owner, its operator, the carrier’s safety director, the original reporting officer, or any other person knowledgeable about the truck at the time of the crash. The interviewers collect a set of detailed information about the configuration of the vehicle, which is cross-checked with multiple sources, such as manufac- turer’s specifications. As such, the TIFA file should provide a complete record of hazmat releases in fatal truck crashes, incidents that should be in the HMIRS data. Given the seriousness of fatal truck crashes, one would expect these events to be among the most likely to be reported to HMIRS. Accordingly, estimates of reporting completeness in this subset may be regarded as close to the upper boundary of the plausible range for the true report- ing rate. TIFA and HMIRS data for 2005, the most recent year when both were available, were used in the comparison. The TIFA file includes all traffic accidents involving a medium or heavy truck and a fatality that occurred as a consequence of the accident, whether from hazmat release or not. Cases required to be reported to HMIRS involve either a release or damage to the lading system. The TIFA file records whether there was a hazmat spill, so HMIRS cases that meet that criterion can be iden- tified in TIFA. TIFA, however, does not capture information that indicates damage to the lading system, although it does capture rollover, which is probably strongly associated with damage to the lading system. Thus, the TIFA file should include all HMIRS-reportable cases that involve a fatal- ity and truck transportation. However, only the cases reportable because there was a hazmat release can be specifically identified in TIFA. Cases that are reportable because of damage to the lading sys- tem cannot be specifically identified, although many are very likely among those in which the truck rolled over. Of the 53 HMIRS highway cases in 2005, 50 were matched to a record in the TIFA file. The match had to be done manually; that is, by reviewing individual records, since there are no case identifiers in common between the two files. In addition, other potential match variables— location and time—all had various problems. The time and even the date recorded in TIFA var- ied from the self-reported data in HMIRS. Time varied most frequently, often by +/−30 minutes. The only location information useful for the match that was common between the two files was the county name. In the TIFA file, county is captured as a numeric code (the Federal Informa- tion Processing Standards [FIPS] code, as found at http://www.census.gov/geo/www/fips/fips. html, which is part of a standard and widely used geographic identification system). In HMIRS, county is an alphabetic string, and therefore subject to the vagaries of spelling. There were vari- ations in spelling the county name and, in one case, an incident was reported as occurring in Albany County, NY, in the HMIRS data, but actually occurred in Albany County, WY. Hand- matching is unavoidable when such materials are all that is available. Three HMIRS cases did not appear in the TIFA file. This may occur if the vehicles transport- ing the hazardous material were not medium or heavy trucks. Each case was searched for in the 2005 FARS file. One case from Bronx, NY, was in FARS, but not TIFA. The vehicle was coded as a truck, so the case should have been included in TIFA. The reason it was not included is that it was added to FARS after the complete FARS file was released to the public. A corrected version of FARS was released with the case after the TIFA file was itself completed. It is likely that the two other records in HMIRS but not TIFA were inaccurately reported. One record was from Douglas County, KS, on 12/22/2005 at 6:30. This record seemed to be matched by three records in the 2005 FARS, which occurred on the same date at 6:38. But all three vehi- cles were light vehicles, and none coded in FARS as carrying hazmat cargo. However, it is known that FARS underreports hazardous material in cargo. Finally, a record from Lynchburg City, VA, on 1/31/2005, was not found in FARS at all. No fatal crash occurred in Virginia on that day. There may have been an error in the date of the incident in HMIRS, or some other error. Database Analysis 51

One criterion for reporting to the HMIRS file is a spill of hazardous materials. All fatal truck crash involvements in which hazmat cargo is released should be reported to the HMIRS. The TIFA data show that there were 67 trucks in a fatal crash in which hazardous material spilled (see Table 4-13). Forty of the records were found in HMIRS, for a reporting rate of 59.7%. Cargo spillage clearly is associated with reporting, as the data in the table show that only 9 of the 115 trucks with hazmat cargo that did not spill were reported. Overall, 26.9% of trucks carrying hazardous materials that were involved in a fatal crash were reported to HMIRS. Non-spill incidents, which nonetheless included damage to protection for the lading reten- tion system, are also required to be reported. This could account for the nine non-spill cases that were found in HMIRS and, likewise, the 106 non-spill cases that were not. The TIFA data do not include any information on whether there was damage to the system protecting the cargo reten- tion system of the truck, so there is no way to determine directly if such damage accounts for the observed pattern of reporting. However, the TIFA data include a variable that identifies whether the vehicle rolled over. Rollover should be strongly associated with damage to the truck, includ- ing the system protecting the cargo retention system, so trucks with hazardous material that rolled over as part of the crash would be expected to be reported to HMIRS. Again, there is no way to determine directly if they must be reported, but certainly a high proportion would be expected to qualify. In Table 4-14, the results of matching the HMIRS to the TIFA file are shown disaggregated by cargo spillage and whether the truck rolled over. In the top half of the table, match results are shown for trucks that did not have cargo spill in the crash. Of the cases that rolled over, and thus likely damaged the protection for the cargo retention system and qualified for reporting to HMIRS, only 4 of the 15 (26.7%) were actually reported to HMIRS. Only 5% of the 100 hazmat trucks in a fatal crash, with no spill and no rollover, were reported to HMIRS. Note, however, that trucks with a hazmat spill and rollover were reported at only a 61.2% rate. Crashes in which there was a spill, but no rollover, were reported at a 55.6% rate, somewhat lower but not practi- cally different. The TIFA file includes some limited information about the motor carrier, including whether the carrier was private, operating trucks incidental to another business, or for-hire, and whether the carrier operated in interstate or intrastate commerce. 52 Hazardous Materials Transportation Incident Data for Root Cause Analysis Matched Cargo Spillage No Yes Total None 106 9 115 Spill of hazmat 27 40 67 Unknown 3 1 4 Total 136 50 186 Row Percentages None 92.2 7.8 100.0 Spill of hazmat 40.3 59.7 100.0 Unknown 75.0 25.0 100.0 Total 73.1 26.9 100.0 Table 4-13. TIFA hazmat crash involvements matched with HMIRS TIFA 2005/HMIRS 2005.

Considering all records matched, there was little difference in reporting rates between private and for-hire carriers or between interstate or intrastate carriers. Fatal crash involvements of a truck carrying hazmat cargo operated by an intrastate carrier were reported 19.5% of the time, compared to 29.8% of involvements where the carrier was operating interstate. The rate for interstate carriers is somewhat higher, but there are too few cases for the difference to be consid- ered statistically significant. Rates for private and for-hire carriers are virtually identical, 26.5% and 28.3%, respectively. Although it is not possible to determine definitively whether each case qualified for report- ing either because of a spill or damage to the protection to the lading retention system, the results here at least bound the probable HMIRS reporting rate. On the high side is the 59.7% reporting rate observed for cases in which there was a hazmat release. All of these cases defini- tively qualified for reporting, yet less than three-fifths were actually reported. The lower bound for HMIRS reporting might be the overall reporting rate of 26.9% of fatal truck crash involve- ments with hazmat cargo. However, it is quite unlikely that the overall reporting rate is anywhere near as high as 59%. In the case of the MCMIS crash file, the overall reporting rate is about 75% of the reporting rate for fatal crashes. If the same ratio is applied to the HMIRS file, this would mean the overall reporting rate is about 45%. But the MCMIS crash file is reported by state agencies, and the FMCSA has an intensive program to increase reporting, including paying for changes to systems to increase reporting. The obligation to report to HMIRS falls on the thousands of private haz- mat carriers, and there is no systematic program to make sure that all appropriate cases are entered into the database. Thus, the overall reporting rate may be even lower than that of HMIRS. However, given current data, it is not possible to provide a more realistic estimate. Database Analysis 53 Table 4-14. TIFA/HMIRS match results, by cargo spillage and rollover, TIFA 2005/HMIRS 2005. No Cargo Spill No spill Matched Rollover No Yes Total No roll 95 5 100 Rollover 11 4 15 Total 106 9 115 Row percentages No roll 95.0 5.0 100.0 Rollover 73.3 26.7 100.0 Total 92.2 7.8 100.0 Hazmat Spill Matched Rollover No Yes Total No roll 8 10 18 Rollover 19 30 49 Total 27 40 67 Row percentages No roll 44.4 55.6 100.0 Rollover 38.8 61.2 100.0 Total 40.3 59.7 100.0

4.2.6 Quality Control Process Reports that are not submitted electronically are checked twice. The first check ensures that the translation from the submitted form to the electronic record has been completed accurately. After the records have been placed in an electronic form, the records are checked for business rule inconsistencies, invalid dates, and invalid commodities (by cross-checking with the com- modities in the database). Additional checks include cases of city/county inconsistencies or when the report shows that 5.5 gallons were spilled from a 5-gallon container. In these instances, the filer is contacted and asked whether there were multiple packages that failed, following which, the information is corrected. An analysis of the data demonstrates that some obvious Q/A checks are not being performed. For example, several carriers that file thousands of reports each year file under more than a dozen names and several DOT numbers. The multiple DOT numbers are probably valid and are the result of acquisitions and mergers. However, since all hazmat carriers must annually register with PHMSA, the name of the carrier could be required to be selected from the Registration file. In cases where no entry of the DOT number is provided, the submission should be rejected and the carrier required to resubmit the report with the DOT number completed. Unfortunately, sev- eral large carriers that are submitting thousands of reports never provide a DOT number on a single report. 4.2.7 Interconnectivity with Other Databases The interconnectivity of HMIRS with other databases varies by mode. For trucks, the DOT number, date of incident, and county of incident provides a fairly good way to link records in HMIRS and MCMIS databases. Incident date, county, and the occurrence of a fatal accident might be a good technique to link HMIRS and TIFA. Although some effort was made, there seems to be no good way to link HMIRS and MISLE for ship or barge hazmat accidents. For cer- tain accidents, FRA and/or NTSB might select an accident for more study. In such cases, the addi- tional information that is collected may result in the identification of some contributing causes or even a root cause for the accident. 4.2.8 Analyses Using Database The HMIRS database underwent a major structural change that took effect at the start of 2005. The first part of this evaluation will look at the effect of the HMIRS structural change. This will be followed by an evaluation of some of the parameters that could identify contributing causes of hazmat accidents. Since HMIRS really focuses on package behavior, it is likely that any capa- bility for identifying contributing causes will be limited to package behavior. The final section will look at the gains that might be realized when HMIRS is coupled with other databases such as MCMIS, TIFA, RAIRS, and MISLE. 4.2.8.1 Effect of HMIRS Structural Changes Taking Effect in 2005 The major structural change made to HMIRS at the beginning of 2005 was to break up the three main tables, commonly labeled MAT, CON, and RMK into a series of smaller linked tables titled: IREPORT, IEVENT, IACTION, SHIPPER, MATERIAL, PACKAGE, PKGLAYER, and PKGFAIL. From an overall perspective, the material previously found in MAT is now found in IREPORT and SHIPPER. The material previously found in CON is now found in PACKAGE, PKGLAYER, and PKGFAIL. IEVENT appears to capture the descriptive text previously found in RMK, and IACTION is a new table that asks the carrier what actions have been taken to reduce the likelihood 54 Hazardous Materials Transportation Incident Data for Root Cause Analysis

or the consequences of this accident should the conditions that initiated the accident be present in the future. 4.2.8.2 Main Features of the Restructured HMIRS Database In the restructured database, the basic accident information is contained in the IREPORT Table. This table contains information regarding the date, time, and location of the incident, and the resulting consequences expressed in terms of hazmat- and non-hazmat-related fatalities and injuries, road closure durations, evacuations, and damage costs. The changes to IREPORT following the restructuring are described in Table 4-15. Similar tables could be prepared for the SHIPPER and PACKAGE Tables. In these cases, the data are restructured but there does not appear to be a significant expansion of the data fields. Overall, the restructured HMIRS provides a much-improved description of accident consequences. How- ever, there is limited information on the driver, description of the route, and conditions at the acci- dent scene. To obtain more accurate information on the accident scene, such as whether the acci- dent occurred on a curve or while turning at an intersection, one would have to be able to identify the location of the accident from the route and location fields and then refer to map software to determine the road geometry. The other alternatives would be to find the accident in MCMIS or access the specific PAR. Privacy concerns may limit the availability of some personal driver infor- mation. To make it publicly available, personal driver information could reside in a separate file that would be kept confidential. The packaging information provided before and after the restructuring of the database was extensive and has only improved. Capacity and quantity shipped is now requested as part of the report. 4.2.8.3 Relevant en Route Accident Statistics Table 4-16 shows the statistics for only those records with T_PHASE = 261, signifying the phase of operations is “en route” when ACCIDENT = T, signifying that the record is submitted because a reportable accident occurred. Tables 4-10 and 4-11 both summarized the records for the years 2005 and 2006, the first two years of reporting after the database reconstruction. Database Analysis 55 Table 4-15. Main features incorporated into the restructured database. General Topic Area Variables in Tables after Restructuring Status Prior to 2005 Report Referencing IREPORT contains several fields that could be used to relate incident reports to reports in other databases. There were no fields that would enable linking to records in other databases. Contact Information IREPORT contains numerous fields listing names, addresses, and phone numbers of the person filling out the report and provides entries to list the police accident report number. No information regarding person filling out the form recorded with the accident record. Deaths and Injuries IREPORT breaks down the deaths and injuries into public, employee, and emergency responders for both HM and for non-HM. Previously, only deaths and injuries from HM were listed and no breakdown into classes of individuals was possible. Evacuation Both the number and type of individuals evacuated is given, as is the duration of the evacuation. If a major road was closed, the duration of the closure is also given. Previously, just the number of individuals evaluated was listed. Conditions Weather condition is now listed. Previously no listing of weather conditions was recorded. Phase In addition to the transport phase, for air, additional information is given as to the step in the multimodal operation where the spill occurred. Only the phase is given, en route, loading, unloading, or en route storage.

Table 4-12 provided an expansion of part of Table 4-11, focusing on en route accidents and the reason for reporting as determined by the RPT_TYPE parameter. Any accident that results in a hazmat spill is classified as “A,” an undeclared hazmat spill is classified as a “B,” and new report- ing category “C” indicates that a cargo tank with a capacity of 1,000 gallons or greater was involved in an accident in which (1) there was damage to the lading or the safety system protecting the lading such that repairs had to be made and (2) there was no spill of hazardous material. The totals in Table 4-16 are mostly from spills, as can be seen by comparing the rows for RPT_TYPE = A with the totals row in Table 4-16. Note, that all the air and vessel records are con- sidered spills and not due to an accident. Of the 7,306 reports in Table 4-10, Table 4-11 shows that only 649 are accidents. Note that the absence of air and vessel columns in Table 4-16 indi- cates that all the air and vessel records shown in Table 4-10 are spills not associated with any en route accident. The new requirement to report non-spills associated with cargo tanks having capacities of greater than 1,000 gallons if the ladling or the system protecting the ladling is damaged is shown under Report Type C. There are a total of 33 reported accidents that were coded under this new classification. The remainder of this analysis will look at these 649 reports. The first evaluation will focus on whether the 649 accidents associated with the IREPORT records have corresponding records in the SHIPPER, MATERIAL, PACKAGE, and PKGFAIL Tables. By successively linking the tables to the IREPORT records it is possible to determine if there are accidents in IREPORT that do not have corresponding records in the linked tables. When SHIPPER is linked, the total number of records increases from 649 to 730. However, none of the IREPORT records are dropped. This emphasizes that one of the features of the restructured data- base is its ability to separately provide an accurate description of the spills of hazardous materials and their behavior for those offered for transport by several shippers. The 558 truck accidents rep- resent material from 596 shippers while the 91 rail accidents were associated with 134 shippers. The only limitation found in the SHIPPER Table is that the destination field has been left blank for 105 SHIPPER records. From the standpoint of identifying contributing and root causes, the lack of destination information for a significant fraction of the records is probably not a limitation. When the MATERIAL Table is added, the number of records increases to 750 records, indi- cating that several accident records in IREPORT involve multiple types of hazardous material. Once again, all 649 accidents described in IREPORT are represented. Adding the PACKAGE Table increases the number of number of records to 762, indicating that some materials in mul- tiple packages have been described accurately. The number of accidents associated with these 762 packages remains 649, indicating that package information is available for all 649 accidents listed in IREPORT. When the PKGFAIL Table is linked to the other tables, the situation changes significantly. There are 130 accidents described in IREPORT that have no PKGFAIL 56 Hazardous Materials Transportation Incident Data for Root Cause Analysis RPT_TYPE Year Truck Rail Totalsby Year Totals by Report Type A 2005 274 48 322 A 2006 247 39 286 608 B 2005 0 0 B 2006 4 8 8 C 2005 20 0 0 4 20 C 2006 13 0 13 33 Totals 558 91 649 Table 4-16. Summary of IREPORT records for en route accidents.

records and there are another 6 that have no failure cause for the package failure. Thus, the cause of package failure cannot be described for 78% of the accidents. A check was made to see if the package failure might be missing because these were Report Class C accidents (and thus no spill) but although 24 of the 33 Class C accidents are among the 136 with no PKGFAIL records, there are many where the information is missing. A query to look at the number of accidents with missing PKGFAIL records that were recorded as spills showed that all were flagged as spills. Clearly, the carrier should have provided a PKGFAIL record for all 130 of the accidents. 4.2.8.4 Significant Parameters HMIRS records request that the carrier reporting an accident fill out the PKGFAIL Table and from the WHY FAILED field, it is possible to identify why the package failed. Note that this is not why the accident occurred. A shipper could place some corrective actions in the IACTION Table that may be indicative of the contributing causes for the accident but the causes do not have to be listed in the database. HMIRS does not contain other driver infor- mation. Therefore, the only driver information in the database is that incidentally found in the IACTION Table. HMIRS also has very limited information on the location where the accident occurred. There are four relevant parameters that help identify the location of the accident: I_STATE, I_COUNTY, I_CITY, and I_ROUTE. The user guide requests that the I_ROUTE parameter specify the “street location on which incident occurred.” Based on the 558 truck accidents that occurred in 2005 and 2006, the carriers filing interpret I_ROUTE to be a street address, a route designation, and fre- quently a mile marker or mile-post designation, or an intersection of two named or numbered highways. The I_ROUTE listings that were not considered adequate to identify the location of the accident provided were a blank entry, just the route designation, or an incomplete phrase that des- ignates the route location as some distance from an undefined point. Of the 558 truck accidents, the actual location of the accident could only be identified for 347 cases, just over 60% of the acci- dents. Thus, if route characteristics were a contributing cause of the accident, it would be impos- sible to identify those causes for 40% of the accidents reported in HMIRS. HMIRS also contains limited information on the vehicle characteristics. If the package type is a C, a cargo tank, and the volume shipped is larger than 5,000 gallons, one can infer that the vehicle configuration is a semitrailer hauling a cargo tank. If the vehicle configuration is a contributing cause for the accident, the HMIRS record must be coupled with data in MCMIS. In MCMIS, if the data fields are fully populated, the vehicle, driver, and road characteristics documented in MCMIS can be coupled with the package information in HMIRS to get a comprehensive picture of the driver, vehicle, package, and route characteristics present at the time of the accident. The restructured HMIRS database provides fields to enter the carrier’s DOT number, a parameter not requested prior to the 2005 restructuring. If an accident meets the reporting cri- teria in both HMIRS and MCMIS, and the data fields are all filled out, then an analyst has a good description of the accident expressed in terms of driver, vehicle, package, and route characteris- tics. Coupling HMIRS with RAIRS and TIFA would provide a comprehensive picture of the acci- dent, just as is the case with coupling HMIRS with MCMIS. For a two-year period, there were no accidents involving hazardous materials that were reported on the water in both MISLE and HMIRS. The reporting criteria seem to be too different so for the few water spills of DOT-defined hazardous materials that occur as a result of ship or boat accidents, it is not possible to supple- ment the package data shown in HMIRS. For the DOT number field to be useful, it must be comparable with the corresponding field in MCMIS. HMIRS records the number as a number field and it is recorded in MCMIS as a text field. In MCMIS, sometimes there are letters like DOT and US preceding the number. If the let- Database Analysis 57

ters are ignored, it is easy to convert the text field into numbers and look for matches based on the date and state. The county and time can then be used to verify the matches. To begin with, the DOT number is not supplied for 60 of the 558 accidents, reducing the total number of pos- sible matches to 498. When the DOT number, date, and state of accident occurrence were used to match up the 498 accidents identified in HMIRS for years 2005 and 2006 with the much more extensive MCMIS file, a total of 110 matches were found. Four of the records were eliminated because the accident times were different by several hours and the counties were also different. Thus, for slightly more than 20% of the HMIRS records, the accident reporting criteria overlap suf- ficiently to require the accident to be reported in both databases. Of the 106 matched records, there were 8 records where the times were within 15 minutes but the counties were different. Using map- ping software, and the location field as a final determiner, it was concluded that in six of these eight differences, HMIRS recorded the wrong county and in the other two, the county reported in MCMIS was incorrect. In all of the cases, the accident location was very close to the county border, normally less than three miles. In the six cases where it is suspected that the county recorded in HMIRS is wrong, one possible cause is the extra step used in recording the county. Rather than entering the county name, the county FIPS code was used. Entering the actual name introduces the wide variety of misspellings possible. This impedes computer matches. The use of the FIPS code is much better. If the name was entered as a string, it could be used to cross-check with FIPS, which would be useful to guard against errors in entering the code. But the best approach may be to avoid having to use probabilistic matching and use a case identifier, such as police report number. Every time there is an extra step, an additional source of errors can arise. A single query was used to identify 110 accidents reported to both MCMIS and HMIRS that occurred for the same carrier on the same date, in the same state. These accidents were then screened to see if there were any cases where the MCMIS record should not have been joined with the HMIRS record. There were 12 records that were suspect and, in the end, 4 were believed to represent different accidents. One way to eliminate this 10% error rate would be to join additional fields. Requiring the county to be the same would have eliminated all 12 records, including the 8 where either HMIRS or MCMIS reported the wrong county. Time was more difficult to use because even for those believed to be reported in both data- bases, the times were often different by 15 minutes or more. Location was quickly rejected because of the vast difference in the types of information recorded for that field. Of the 106 MCMIS records that were joined to HMIRS, one-half were not flagged as being a hazmat shipment in any of the four hazmat descriptive fields in MCMIS. Thus, if the join between HMIRS and MCMIS had just looked at the MCMIS records that were flagged as being haz- mat, one-half of the incidents would have been missed. Information technology is rapidly advancing to the point that it is feasible to require the lon- gitude and latitude of the accident to be included as part of the accident record. Many states already have longitude/latitude as an entry on their police accident reports but few police offi- cers populate the field when filling out the form. It is not believed that the reason is their inabil- ity to know the coordinates of their position. Most police cars can be located geographically by their dispatcher and, for many companies, this capability also exists for trucks and trains. Thus, provided that a format for the longitude and latitude are specified—decimal degrees or degrees, minutes, and seconds—it should be possible to use these two fields to join records and thereby eliminate the screening step. Many handheld GPS devices will provide this information and, for those without a handheld device, the accident can be located on numerous free and commercial software packages. For these free packages, all that is needed is Internet access. 4.2.8.5 Class C Record Types In the restructured database, the 5800.1 form imposes a new requirement. If a cargo tank having a capacity of greater than 1,000 gallons is involved in an accident and the lading or the 58 Hazardous Materials Transportation Incident Data for Root Cause Analysis

system protecting the lading is damaged, the carrier is to report the accident even if there is no spill. The way these accidents are coded is to enter a “C” under the RPT_TYPE variable in the IREPORT Table. The other designations are “A” for a spill and “B” for an undeclared hazmat shipment. Of the 649 en route hazmat accidents, 33 shipments, all truck mode, were coded as RPT_TYPE = C for the years 2005 and 2006. The rate of carrier compliance with this new requirement to report some non-spill hazmat accidents for cargo tanks is difficult to determine. Would one expect more than 33 incidents in two years? Focusing just on the truck accidents, in 2005 and 2006, there are 558 records. A query for the cargo tank configuration shows that 442 of those accidents involved cargo tanks. One approach for investigating underreporting is to examine the rollovers reported as spills and determine if they are being over-represented. In the Hazardous Materials Serious Crash Analysis: Phase 2 (Battelle 2005) study, spills occurred in 66% of the rollovers. Of the 442 cargo tank accidents reported in HMIRS, 357 rolled over and of those rollovers, 291 were coded as spills. Thus spills occurred in 82% of all rollovers. Because rollovers are likely to damage the rollover protection system, spills from rollovers seem to be over represented in the HMIRS database. The number of cargo tank rollovers reported as Class C records was 23. The non- spill cargo tank rollovers would have to be increased from 23 in two years to 110, an increase of about 85 non-spill accidents to lower the spill rate to 66%. Is it reasonable that 110 cargo tanks would rollover in a two-year period and only 23 would experience damage to their rollover protection system that was serious enough to require repair? Given that cargo tank rollovers frequently result in a release, it might be anticipated that damage would occur in more than 21% (23/110) of the non-spill rollover accidents. Perhaps a more fundamental result is that if all cargo tank accidents that met the MCMIS definition of serious were required to be reported, the number of additional records being reported would increase by at most a few hundred. Given that there are more than 30,000 records added to HMIRS every two years, requiring all serious cargo tank accidents to be reported would increase the record load on PHMSA by less than 1%. By making the HMIRS reporting requirements have some of the same requirements as MCMIS, the added benefit is that more accidents would be reported in both databases. Since MCMIS has more information on the road configuration, environmen- tal parameters, and driver characteristics, it should be possible to perform more analyses that move toward identifying contributing and root causes of hazmat accidents. 4.2.8.6 Inclusion of IACTION Table One of the major changes in the restructuring of the HMIRS database was the addition of an action statement table that can be used by carriers to state the changes they propose to make to prevent or reduce the likelihood that such accidents would occur in the future. There are 649 accident records in IREPORT for 2005 and 2006 in which T_PHASE = 261 and ACCIDENT = T. These consisted of 558 and 91 truck and rail records, respectively. There are 453 records with action entries, 411 for truck and 42 for rail. Thus, the percentage of carriers providing action statements is 74% for truck and 46% for rail. In looking at the action statements, while the contributing cause is not given, it is clear that the contributing cause that resulted in the action statement is known by the carrier, but it just is not stated. Like the causes of failure that are used in the PKG_FAIL Table, a table of contributing causes could be developed and added to the IACTION Table that would provide the basis for the actions taken; they would not have to be assumed. The tables developed in the RAIRS might be a useful place to start when develop- ing this table of contributing causes. Even though no driver information is present in HMIRS, many of the actions focus on increased driver training, so it is clear that one of the contributing causes for many actions is inadequate driver training. Another goal would be to increase the number of carriers providing action statements, particularly for rail. A goal could be to have a compliance rate in excess of 90%. Database Analysis 59

4.2.8.7 Differences before and after Changes in HMIRS Structure Prior to the restructuring of HMIRS, hazmat records had to be duplicated to account for the presence of a carrier hauling hazardous material for several shippers in the same vehicle and for designating more than one destination for some of the packages. A query of the database for PHASE = 261 and ACCDR = True, shows that there are a total of 603 primary incident records for 2003 and 2004 and an additional 146 records to account for multiple shippers and multiple destinations. The total number of records is therefore 749 records, remarkably similar to the total of 747 records reported for 2005 and 2006 combined. Although not important when consider- ing root causes, it is easier to understand the shipment logistics after the restructuring. The total number of incidents is quite close as well, considering that in the 649 accidents reported for 2005 and 2006, 33 were associated with the new requirement to report non-spills if a cargo tank hav- ing a capacity of 1,000 gallons or greater was involved in an accident and there was damage to the cargo or the equipment protecting the cargo. If those were not considered, the number of crashes in 2005 and 2006 would total 616, again quite similar to the 603 reported in 2003 and 2004. A more detailed evaluation of the records is found in Appendix D (available on the TRB website at www.TRB.org by searching for HMCRP Report 1). 4.2.9 Summary and Potential Measures for Improving Root Cause Analysis 4.2.9.1 Summary of Database Evaluations The restructured HMIRS database can be considered to be a relational database and, except for the PKGFAIL Table, the record set for an en route accident is complete. Even in the case of the PKGFAIL Table, the data are available for about 80% of the HMIRS en route accidents. From the point of view of identifying route and contributing causes, this is not believed to be a signif- icant limitation. For a complete description of the package, vehicle, driver, and roadway characteristics asso- ciated with an accident, HMIRS would have to be joined with MCMIS for trucks and RAIRS for rail. Until the restructuring of HMIRS, the biggest detriment to joining the two databases was the lack of common fields. HMIRS now has a field to enter the DOT number, and this field is now being populated almost 90% of the time. The DOT number is also entered for about 90% of the MCMIS records designated as showing a hazmat placard. Assuming the non-reporting is random, the DOT number can be used to join about 80% of the accidents that meet both the HMIRS and MCMIS reporting criteria. Since all carriers of placarded quantities of hazardous material must register with both FMCSA and PHMSA, they must have a DOT number and there should be no blank entries in either database. Information technology has advanced to the point where both the carrier reporting in HMIRS and the police officer reporting to MCMIS have the capability to report the longitude and lati- tude of the accident. Providing that a common format is used in both databases, it is believed that one query could be used to identify accidents reported to both HMIRS and MCMIS and the data-scrubbing step could be eliminated or significantly reduced. The main reason why a fraction of the HMIRS and MCMIS records cannot be linked is the difference in reporting criteria. Some of the difference between the number of records in MCMIS that can be joined with HMIRS records can also be attributed to underreporting. It is suspected that carriers are not reporting all of their Report Type C accidents but that state- ment can not be made with certainty. Lastly, the carriers are providing action statements for 74% of the truck accidents and 46% of the rail accidents. An increase of the carrier reporting rate to at least 90% would be highly desirable. In addition, the action statements given are quite positive 60 Hazardous Materials Transportation Incident Data for Root Cause Analysis

and indicate that the carriers have done enough accident investigation to identify some changes that would decrease the frequency of similar accidents in the future. The usefulness of this infor- mation would be greatly improved if a cause table, similar to the WHAT FAILED Table, was devel- oped so the carrier could list some contributory causes from a pick list. Although there might be some resistance to adding that field because of liability issues, moving toward being able to rou- tinely list contributing causes would be helpful. 4.2.9.2 Potential Measures for Improving Root Cause Analysis The following potential measures would enhance the ability of HMIRS to identify the root causes of hazmat accidents. 1. Require that the DOT number be a mandatory input for all reports filed with PHMSA for en route incidents. 2. Perform an additional Q/A check on carrier names to verify that the name being entered cor- responds to the name provided on the annual PHMSA Registration form. 3. Require PKGFAIL entries to be filled out for all reports submitted to PHMSA. 4. Continue to emphasize the new requirement that carriers must file a 5800.1 form following an accident if there was damage to lading and lading protection systems on cargo tanks of 1,000 gal- lons or greater, even though there is no loss of hazardous material. This is the new requirement to report Class C accidents. Such a notice might be given to carriers when PHMSA notifies them that it has received and approved their annual hazmat registration application. 5. Capture driver condition information without compromising the confidentiality of the driver. The following design option from MCMIS can be enhanced for use in HMIRS. Based on analysis of the data, the list of options can be enhanced by using the following driver con- dition categories: 1 = Appeared Normal, 2 = Had Been Drinking, 3 = Illegal Drug Use, 4 = Sick, 5 = Fatigue, 6 = Asleep, 7 = Medication, and 8 = Unknown. The project team believes that adopting the potential measures above would decrease errors in data entry and make it easier to query the database for potential causes of accidents. 4.3 Fatality Analysis Reporting System (FARS) This section briefly describes the FARS file. Since the TIFA database incorporates the FARS records for trucks involved in fatal accidents, to avoid a fragmented analysis, much of the detailed evaluation is covered in Section 4.4, which describes TIFA. The FARS file is the primary national crash data file for fatal traffic accidents. It is a census of all fatal motor vehicle traffic crashes. The TIFA file covers all medium and heavy trucks involved in a fatal crash, and includes virtually all FARS variables for the crash, vehicle, and driver. TIFA survey data supplements FARS data for trucks (hereafter the word “trucks” will be used to refer to medium and heavy trucks, i.e., trucks with a gross vehicle weight rating [GVWR] over 10,000 lbs). The TIFA data include a more accurate identification and descrip- tion of trucks in fatal crashes, along with details about the cargo, configuration, motor carrier operating the vehicle, and crash type. Database Analysis 61

Both TIFA and FARS collect information about hazardous materials in the cargo. In the discussion of the variables that identify hazmat cargo in FARS, it will be shown that there are reasons to believe that the TIFA file identifies hazmat cargo more accurately. Since TIFA incorporates virtually all FARS variables, discussion of those variables and their usefulness will be discussed in Section 4.4, which focuses on the TIFA file. 4.3.1 Agencies/Organizations Responsible for Data Collection and Entry FARS is compiled by the National Center for Statistics and Analysis at NHTSA. 4.3.2 Database Years of Coverage The FARS file was initiated in 1975 and has been in continuous operation to the present time. 4.3.3 Criteria for Reporting and Inclusion of Data The FARS file includes all traffic crashes involving • A fatality that occurs as a result of a crash, or • A fatality that occurs within 30 days of a crash, and • At least one motor vehicle in transport on a public road. 4.3.4 Types of Hazmat Data Included The FARS crash data file includes limited information regarding hazardous materials. Since 2005, the vehicle-related variables (up to two responses allowed) include a level that captures haz- mat cargo releases as a result of a crash. The vehicle-related variables record pre-existing vehicle defects or special conditions related to the vehicle. The vehicle configuration variable, added in 2001, identifies light trucks or passenger cars that display a hazmat placard. The driver-related variable (up to four responses allowed) includes an entry for “carrying hazardous cargo improperly.” Finally, the hazardous cargo variable records if a vehicle was transporting hazardous material and if it was placarded. Comparison of the identification of hazmat cargo in FARS and TIFA over a five-year period showed a large discrepancy. The FARS file identified hazardous material in the cargo in 1,257 cases over that period, while hazardous material was identified in only 1,049 cases in TIFA (see Figure 4-3). Surprisingly, when the comparison was made on a case-by-case basis, there was a large amount of disagreement between the files. As shown in Table 4-17, hazmat cargo was identified in both FARS and TIFA in only 706 cases over the observation period. FARS coded 551 trucks with hazardous material when the TIFA survey did not identify hazmat cargo, but in 343 cases, the TIFA survey showed that the truck had hazmat cargo and FARS did not. 62 Hazardous Materials Transportation Incident Data for Root Cause Analysis 1, 257 in FA RS 1, 049 in TI FA Figure 4-3. Comparison of trucks with hazardous materials in FARS and TIFA, 1999–2004.

Only a small part of the discrepancy is explained by differences in the ability to determine if the vehicle was transporting hazardous materials. In eight of the cases marked in FARS as carry- ing hazardous material, the TIFA survey was unable to determine if the vehicle held hazmat, while there were 33 cases in FARS where the analyst left the hazmat variable unknown, but the TIFA survey showed that the vehicle was carrying hazardous material. In most of the cases (880), the coding of hazardous material was directly contradictory (i.e., coded as hazmat in one and as not hazmat in the other). Generally, there are a number of reasons to believe that the identification of hazardous material in the cargo is more accurate in TIFA than FARS. First, it is difficult in FARS to perform direct consistency checks on the hazmat variables, since FARS does not capture any other information about the cargo. Moreover, the FARS data collection protocol does not include direct contact with the carrier, driver, reporting officer, or other potential source, but relies primarily on the police report and other investigative documents. Moreover, the TIFA data collection protocol is based on a telephone survey of the motor carrier, driver, dispatcher, or safety director of the truck involved in the crash, as well as the reporting officer, so those sources are questioned directly. In addition, the TIFA data include other information about the cargo, so it is possible to perform basic checks on the accuracy of the hazmat coding, such as whether the truck was carrying cargo at all. The comparison of TIFA and FARS records showed that 264 of the cases coded in FARS were loaded with hazardous materials, while the TIFA survey showed that those trucks were empty at the time of the crash. Finally, it should be noted that the TIFA survey specializes in trucks, while the FARS file cov- ers all vehicle types. This focus on trucks allows the TIFA survey to go into greater depth and to develop more expertise in the details and varieties of truck operations. FARS analysts cover all vehicle types, and while the FARS file is a quality crash data source, it is not reasonable to expect FARS to have a higher degree of detail and accuracy than a file that has the advantage of narrowly focusing on only one vehicle type. In sum, while no doubt there are errors in the TIFA file, it is likely to be more reliable for analyzing truck crashes than FARS. 4.3.5 Usefulness of the Data for Determining Root Causes Because the TIFA file incorporates the relevant data, a discussion of usefulness is deferred to Section 4.4 on the TIFA file. 4.3.6 Data Quality FARS includes multiple layers of quality control. Cases are entered using computer software that includes validity and consistency checks. The validity checks ensure that the values entered are possible for the field. For example, if a field has valid values for one through seven, but an Database Analysis 63 Hazmat in cargo Number of cases Coded as Hazmat in FARS, not in TIFA 551 Coded as Hazmat in both FARS and TIFA 706 Coded as Hazmat in TIFA, not in FARS 343 Table 4-17. Hazardous materials in FARS and TIFA, 1999–2004.

eight is entered, the eight would be rejected; similarly, alphas would be rejected if the field were numeric. There are also extensive computer consistency checks for each field, in which the value in one field is compared with the values in other fields to ensure that the fields are consistent. For example, if the first harmful event is a collision with a non-motorist, the field for the number of non-motorists involved must not be zero. There are multiple consistency checks for each field. Some of the checks are prescriptive, that is, certain values must be registered, while others flag unusual situations that should be reviewed. The consistency checks are documented in the Cod- ing and Validation Manual (NHTSA 2002) for the FARS system. In addition, analysts receive annual training at a national meeting. Similar quality control procedures are implemented as the cases are aggregated into the final file. The records are reviewed for timeliness and completeness. Statistical control charts are used to monitor the coding of key data over time, to see if distributions are wandering according to past experience. Typically, three versions of the FARS file are released. The “early assessment” file is released as a partial file that provides an initial look at the data for a year. Next there is a “complete” version that is typically released in the fall of the year following the data year. A “final” version, which includes all corrections and additional cases that have accumulated, is sub- sequently released, typically 18 or more months after the data year. In some ways, the FARS file is the gold standard for data on fatal crashes. It is the product of considerable care over time, and is produced by a system that incorporates many checks for con- sistency and accuracy. On stable, well-understood data elements such as the environment of the crash, it is assumed to be of high quality and accurate. Items such as weather, time of day, light condition, and road type are coded from police reports and are dependent on the accuracy of the PAR. Although the accuracy of this information is unknown, these conditions are relatively sta- ble and should be identified on the PAR with acceptable accuracy. Similarly, it is assumed that the FARS file is acceptably complete (that is, virtually all vehicles involved in a fatal crash are included). However, given the sheer number of fatal crashes annu- ally (about 40,000) and vehicles involved (about 55,000), it is not possible that every crash and every vehicle involved is included. In the process of compiling the annual TIFA file, each year, a small number of trucks appear on a police report for a fatal crash, but the FARS file contains no record for that vehicle. It is also possible that some fatal crashes are missed because the fatality occurs toward the end of the 30-day window. These few omissions are, however, undoubtedly very small in number and inconsequential. The accuracy of the FARS data with respect to the main concerns of this project is a more com- plex matter. The TIFA file, indeed the entire TIFA protocol, allows an independent check for the accuracy of those data elements in common between it and the FARS file. The inconsistencies with respect to the hazmat variables have been noted above. There also are problems with the identification of trucks in FARS. Cases extracted for the TIFA survey include some categories of vehicles that are not identified as medium or heavy trucks in FARS. These include light vehicles coded with a GVWR of more than 10,000 pounds. Each year, the TIFA survey determines that 200 to 300 of these vehicles are, in fact, medium or heavy trucks. In addition, the TIFA survey determines that a number of the vehicles identified in FARS as trucks are actually light vehicles. Typically there are 60 to 70 vehicles identified in FARS as a medium or heavy truck that prove to be a light vehicle or something other than a truck. 4.3.7 Additional Fields The FARS system is quite complete and includes valuable fields. However, the addition of other data fields to the protocol would contribute to the ability to analyze crash causation. Some could be very easy to include, and require little modification of the program. Others would take 64 Hazardous Materials Transportation Incident Data for Root Cause Analysis

additional resources, but would fit well within NHTSA’s crash data program. Adding the follow- ing data fields is suggested: • Right of way. This data element would identify which vehicle, if any, within a crash had the right of way prior to the collision. This could be readily coded from the PAR in most cases. Some state crash reports include right of way on the report. Right of way would be very use- ful in most crashes in identifying the vehicle that primarily contributed to the crash. • Accident type. The General Estimates System (GES) file and Crashworthiness Data System (CDS) file both include an accident type variable coded at the vehicle level that captures the relative position and movement of the vehicle prior to its first harmful event. The TIFA data adds this to trucks in fatal crashes, but capturing this within the FARS system would be a valu- able addition. An accident type field can identify key relationships that describe how the crash occurred and suggest contribution (for example, by identifying the vehicle that crossed over the center line in a head-on collision). The following two fields would be useful although this would take additional resources and possibly require some change in the management of the FARS file: • Critical event is a field that would identify and describe the event that precipitated the vehicle crash. This field is included in both the GES and CDS files, so the agency is very familiar with (and, indeed, invented) its use. • Critical reason captures the “reason” for the critical event, classified broadly as driver, vehicle, or environment, with detailed levels under each. The variable is useful for identifying the immediate failure that led to the crash and would shed considerable light on crash causation. The field was used in the LTCCS, conducted jointly by the FMCSA and NHTSA, and in the National Motor Vehicle Crash Causation Survey (NMVCCS), conducted by NHTSA. Thus, the agency already has developed coding procedures for both variables. However, adding these fields might require some changes to the FARS protocol. Both are dif- ficult to code consistently and require a high level of focus and analysis. Currently, virtually all FARS fields are coded by analysts located off-site, that is in the 50 states and District of Colum- bia. But the coding of both GES and CDS is more centralized. In the LTCCS, both critical event and critical reason were coded by a small number of analysts in two locations. The National Cen- ter for Statistics and Analysis (NCSA) could adopt a similar method for the FARS file, if these data elements were added. 4.3.8 Potential Measures to Improve Data Quality The FARS quality control system is complete and mature. It is subject to annual review and adjustment, including continuous training of the coders. FARS might be improved if the system could be adapted to take advantage of the additional information provided through the TIFA system. FARS has not engaged TIFA in this regard, although one problem has been that information from TIFA has not been available in a timely fashion. However, greater cooperation between the systems would be valuable for both. 4.3.9 Compatibility with Other Databases The FARS file does not include case identifiers that can be used to uniquely link to other data systems, such as the PAR number. Including the PAR number would provide a hard link. (Note that the MCMIS Crash file report number field in the past was supposed to include the PAR number in one of the fields, and it is recommended that MCMIS require that again. Currently, many states use a random report number, rather than using the PAR number.) Database Analysis 65

FARS does include information about the time and geographic location of the crash, as well as vehicle descriptive information, that can be used to obtain a probable link to records in other files. This includes date, time, state, county, city (if applicable), and an alpha string with the road name and mile marker. The file also includes latitude and longitude, although its accuracy is unknown. FARS is highly compatible in another important way—capturing crash information using fields and code levels that are consistent with standard accepted practices. The code levels are almost always consistent with those available in other databases with elements in common. This includes the national crash files such as GES and CDS, as well as most state crash data systems. 4.3.10 Data Uses NCSA uses FARS data along with GES data to produce an annual publication called Traffic Safety Facts (NHTSA 2008). This publication tracks annual trends for many crash factors of interest, such as vehicle involvements, deaths, injuries, restraint use, and drug or alcohol use. NHTSA and other analysts use FARS data in virtually all traffic safety analyses that require data on fatal crashes. 4.4 Trucks Involved in Fatal Accidents (TIFA) 4.4.1 Agencies/Organizations Responsible for Collecting and Entering Data into Database The TIFA file is produced by the Center for National Truck and Bus Statistics at the Univer- sity of Michigan Transportation Research Institute (UMTRI). 4.4.2 Database Years of Coverage The TIFA file was initiated in 1980 and has been in continuous operation to the present. 4.4.3 Criteria for Reporting and Inclusion of Data All cases in TIFA are also found in the FARS file, so TIFA shares the FARS reporting thresh- old as follows for crashes in which: • A fatality occurs as a result of the crash, or • A fatality occurs within 30 days of the crash, and • At least one motor vehicle in transport on a public trafficway. Additionally, the TIFA file includes only medium or heavy trucks, defined as trucks with a GVWR of more than 10,000 pounds. 4.4.4 Types of Hazmat Data Included From its inception to 2004, TIFA data captured whether the cargo contained a quantity of haz- ardous material requiring a placard for each cargo body. Whether the hazmat cargo spilled as a consequence of the crash also was captured. The quantity of cargo is captured in terms of weight. Package type information is not collected in detail. Cargo in TIFA is classified into general cat- egories, such as general freight, solids in bulk, liquids in bulk, or gases in bulk. The specific cargo type, however, is recorded in an alphanumeric string. Examples of the type of information 66 Hazardous Materials Transportation Incident Data for Root Cause Analysis

recorded in this field include gasoline, anhydrous ammonia, or jet fuel. More detail, such as the Motor Carrier (MC) type of liquid or gas tank, is not considered. In the 2005 data year, the amount of data recorded for hazardous material was both expanded and restricted. Now, whether the truck was hauling hazardous cargo is recorded at the vehicle level, not separately for each cargo body. In addition, TIFA no longer collects cargo weight. Con- currently, more detail about the hazardous material is now recorded, including the hazmat class and UN number. 4.4.5 Usefulness of the Data for Determining Root Causes This section provides a discussion of the variables in the TIFA Crash file that relate to under- standing the factors that contributed to the crash or affected the severity of its consequences. Most of the variables are extracted from the FARS file and are added to the TIFA file without change. Of the contributing factors identified in Table 4-18, the TIFA file provides the informa- tion on truck configuration, cargo body, gross vehicle weight (GVW) (and gross cargo weight [GCW]), and accident type. Virtually all of the factors listed in Table 4-18 are captured in a single variable. For exam- ple, roadway surface condition is described in one variable with several possible levels. Driver condition, however, is captured across several variables. There is no single variable that con- solidates the information about the driver’s condition at the time of the crash. Instead there are separate variables that identify whether the driver had been drinking or using illegal drugs. Driver fatigue or illness is captured in a “driver-related-factors” variable that includes a vari- ety of driver state, action, and other conditions. In this discussion, it is useful to recognize that there are two types of parameters: conditions present and conditions contributing to producing the crash. Both types can be contributing or root causes of accidents. For example, the weather at the time of the accident is a condition and it may or may not be a contributing or root cause of an accident. Clearly, if it is not captured, there will be no information to implicate weather as a contributing or root cause. The parameters con- tributing to producing a crash include parameters like driver condition pre-crash, conditions such as a car cutting off a truck, forcing it to take evasive action to prevent the accident. The important point is that to identify contributing and root causes of accidents, which parameters serve a spe- cific function as to recognize that both types of parameters are needed. Descriptive information Database Analysis 67 Vehicle Driver Packaging Infrastructure Situational Configuration Age Package Type Road Surface Pre-Crash Condition Cargo Body Experience Quantity Shipped Road Condition Event Type GVW Condition Quantity Lost Road Type Vehicle Speed Valid License Age (Cargo Tank) Traffic Way Impact Location Citation Issued Rollover Protection Access Control Primary Reason Inspection History Speed Limit Accident Type Design Specification No. of Lanes Weather Condition Light Condition Time of Day Key: Factor obtained Partially met Not captured Table 4-18. Contributing factors present in TIFA/FARS.

characterizes conditions at the time of the crash. These include items such as light condition, road- way surface condition, time of day, and even road type. These pieces of data describe the circum- stances and many can be related to changes in crash risk. That is, the condition may have con- tributed to producing the crash, but coding the condition does not tell us whether it did, just that it was present. For example, wet or icy roads increase the possibility that a vehicle will skid while maneuvering, but the fact that the road was coded as wet does not imply a judgment that the lower road friction contributed to producing the crash. However, the TIFA file also includes a set of data elements that identify actions and conditions that are either identified as contributing to the crash or that strongly imply they contributed to producing the crash. There are separate sets of road- way, vehicle, and driver “related factors” that identify defects, conditions, and actions that con- tributed to the crash. Charged traffic violations are also coded for the driver. These two types of variables—conditions present and conditions producing—will be discussed separately, with the producing factors discussed first. There are four variables that directly identify factors that represent a judgment of contribu- tion to a crash: • Driver-related factors, • Vehicle-related factors, • Accident-related factors, and • Violations charged. Each of these factor variables are coded by the FARS analysts from the PAR or other investiga- tive reports that the analysts may be able to obtain. That is, the analysts capture the recorded judgment of the reporting officers, rather than applying their own judgment or inferring condi- tions and actions from the available evidence. Accordingly, information coded in these factor variables has been explicitly stated by the original crash investigators. The variables for driver-related factors record driver conditions or actions that may have con- tributed to producing the crash. Up to four responses may be recorded in these variables. The variables for driver-related factors that can be coded are divided into five different subsections that capture different types of influences on the driver. The first subsection includes driver con- ditions, such as fatigue, asleep, ill, blackout, or some other condition that impaired the ability of the driver to control the vehicle. The second subsection is called “miscellaneous factors” in the coding manual, but really each is a driver error of some sort. Examples of these errors are run- ning off the road, driving with a suspended license, speeding, following improperly, and failure to yield. The next subsection of variables for driver-related factors identifies different conditions that obscured the driver’s vision, including reflected glare, fog, trees, other vehicles, and build- ings. There also is a subsection for conditions or events that caused the vehicle to swerve or skid, including wind, slippery road, objects or holes in the road, and swerving to avoid animals or pedestrians. Finally, there is a subsection to capture the presence of possible distractions, includ- ing cell phones, computers, and navigation systems. With the exception of the codes that merely specify possible distractions, the information cap- tured in the variables for driver-related factors record the original police investigator’s judgment on conditions influencing, or errors made by, the driver that helped to produce the crash. All the others point to driving errors, unsafe driver conditions (e.g., fatigue), or conditions that con- tribute to a high-risk situation, such as view obstructions or causes of skidding or swerving. Each of these may be regarded as identifying a causal factor in the collision. Overall, the variables appear to be both useful and reasonably consistent. Missing data rates are low, with the variable left unknown in just over 1% of cases. Previous analysis has shown that the coding is reasonable. In two-vehicle collisions, typically only one driver will be coded with a factor, although there are some instances where both drivers are coded as having contributed to 68 Hazardous Materials Transportation Incident Data for Root Cause Analysis

the crash. This makes some intuitive sense, because we generally think of crashes being prima- rily due to poor actions by one of the drivers. In addition, the codes have been shown to be con- sistent with the general configuration of the crash. For more on this subject, see The Relative Contribution of Truck Drivers and Passenger Car Drivers to Two-Vehicle, Truck-Car Traffic Crashes (Blower 1998). The variables for vehicle-related factors capture vehicle defects that may have contributed to the crash. The FARS coding manual indicates that the variable is used to record pre-existing vehicle defects (i.e., defects not caused by the crash itself). The FARS manual (NHTSA 2002, p 344) also states that “The vehicle condition(s) noted only indicate the existence of the condition(s). They may or may not have played a role in the accident.” Thus, according to the manual, defects noted may not have contributed to the crash. However, in light of how the data are collected, it is likely that the defects played a role. The source for the data is ultimately the PAR or other crash investi- gation document. Crash investigators typically do not routinely perform a vehicle inspection, but primarily detect vehicle defects if they contributed to the crash. Reviews of the PARs of crashes in which tire defects were recorded in the vehicle-related factors variable showed that in every case the tire defect was mentioned as contributing to producing the crash. Missing data rates for the vehicle-related factors variable are low, averaging just under 2%. However, it is very likely, at least for trucks, that vehicle factors are greatly underreported. Typ- ically, the vehicle-related factors record about 2% of the trucks in fatal accidents with brake prob- lems, and much lower percentages with light system defects. However, roadside inspections of trucks—not crash-involved trucks but just trucks operating on the road—show much higher rates, with about 25% having one or more brake defects and similar percentages with light sys- tem problems. For additional discussion of this topic, see “Vehicle Condition and Heavy Truck Accident Involvement” (Blower 2002). It appears that the variables for vehicle-related factors significantly underreport the true incidence of vehicle defects in crashes. This is not surprising, since police officers are not trained to do vehicle inspections. Such underreporting, however, limits the use of the data. The variables for accident-related factors record conditions at the accident level that may have contributed to the crash. These include roadway design problems and roadway defects such as worn or missing pavement markings, inadequate warnings, and roadway washouts. Addition- ally, the variable captures special circumstances that may have affected the crash, such as previ- ous crashes nearby and police pursuits. Again, the variable is coded from the original PAR, with the requirement that the coded items were specifically mentioned. They are coded as factors that either existed or contributed to producing the crash. Overall, missing data rates for the variables for accident-related factors are also low, and are not coded in only 0.3% of cases. However, in more than 93% of the cases, no accident-related factor is coded. Given the very slight incidence of roadway defects recorded in the variables, road design and condition play a small role or the source of the information is unable to capture its contribution to crashes. Turning to the variables that capture conditions present or characteristics of the vehicle, driver, or environment, each may contribute to a specific crash, but the coding of the variable does not directly establish contribution. Instead, these factors are associated with crash risk— they may, in a sense, establish the preconditions for a crash. Table 4-19 lists all of the contribut- ing factors for each analytical level within an accident, and provides a brief comment character- izing how the information is available in the TIFA database. For the variables relating to infrastructure, all are taken from the FARS file and are coded in formats (i.e., code levels) that match national standards. For example, the same code levels are used in MCMIS, GES (General Estimates System), CDS (Crashworthiness Data System), and Database Analysis 69

most state files. Characteristics of the vehicle are available in greater detail than other crash files, but code levels could be aggregated to match other crash data files if necessary. With respect to drivers, all of the information is available, except driver experience. The factors listed as “situa- tional” also are available in formats that match national standards. Pre-crash condition identi- fies the vehicle’s maneuver prior to the initiation of the crash sequence, and the format used has been widely adopted. The accident-type variable, which captures the relative motion and posi- tion of the vehicles prior to the collision, is very useful in understanding how a crash occurred. The format used in the TIFA file is also used in GES and CDS. The primary area in which the TIFA file falls short relates to the details of packaging. TIFA includes only the general cargo body type with no information about the design specification, inspection history, or rollover protection. Moreover, quantity shipped is partially captured for 1999- 2004, but not captured after that. For all years prior to 2005, quantity shipped was partially captured in the TIFA file as cargo weight, in pounds, for all cargo, but capturing this information was aban- 70 Hazardous Materials Transportation Incident Data for Root Cause Analysis Level Variable Comment Configuration Captures the exact configuration of the vehicle Cargo Body Distinguishes 16 types of cargo bodies, including liquid, dry bulk, and gaseous tank types Vehicle GVW Prior to 2005, captures both gross vehicle weight and gross combination weight; weight variables dropped for 2005 and later Age Captured Experience Not captured Condition Captured in separate variables that identify alcohol use, drug use, and a set of multiple-response variables that code fatigue, asleep, ill, emotional, distracted, etc. Valid License Captured Driver Citation Issued Captured in a multiple-response variable; up to four citations may be recorded Package Type Captured only to the level of cargo body type Quantity Shipped Weight only, prior to 2005; not captured in 2005 and later Quantity Lost Spill/no spill captured only, not quantity spilled Tank Age Not captured Rollover Protection Not captured Inspection History Not captured Packaging Design Specification Not captured Road Surface Captured, in standard format Road Condition Captured, in standard format Road Type Captured, in standard format Traffic Way Captured, in standard format Access Control Not captured directly; instead, inferred from roadway function class Speed Limit Captured, in standard format Infrastructure No. of Lanes Captured, in standard format Pre-Crash Condition Captured as pre-crash maneuver; standard format Event Type Captured as first harmful event; standard format Vehicle Speed Captured in standard format Impact Location Captured as relation to roadway of the first harmful event; standard format Primary Reason Inferred from the vehicle-, crash-, and driver-related factors discussed above Accident Type Coded in the TIFA survey to match similar variable in GES and CDS Weather Condition Captured in standard format Light Condition Captured in standard format Situational Time of Day Captured in standard format Table 4-19. Coverage of contributing factors in TIFA.

doned after 2005. The cargo weight variable was known precisely for about 84%, unknown for about 4%, and partially known (light load or full load) for the remaining 12% of the cases. Generally, the fields derived from the TIFA survey or extracted from FARS are populated fairly completely. Table 4-20 shows that missing data rates for the fields captured are quite uniformly low. If the field is present, for the most part, it is complete in more than 95% of the cases. GVW and quantity shipped are exceptions, because it can be difficult to determine precise values after the fact, but even for those variables there is some information for 80% to 90% of the cases. Miss- ing data rates are somewhat higher for vehicle speed prior to the crash, as that is even more diffi- cult to determine. That information is taken from PAR and is available for about 75% of the cases. Complete rates of missing data, averaged over five years, are provided for all TIFA variables in Appendix D (available on the TRB website at www.TRB.org by searching for HMCRP Report 1). 4.4.6 Data Quality The TIFA system includes multiple layers of quality control. The survey is administered by means of a telephone interview. Each case record is reviewed by an editor for accuracy, con- sistency, and completeness. The vehicle identification number of the power unit is decoded, and the vehicle description from the survey is compared with the original specifications. Infor- mation about cargo and operations are similarly compared with a library of information that has been accumulated from over 25 years during which the TIFA survey has been conducted. Survey information also is compared with the information on the original PAR. Any discrep- ancies are discussed with the interviewer, who may be required to make additional calls for information. Once discrepancies are resolved, the data are then entered with verification. At that point, there is a computerized check of each batch of keypunched cases for consistency and to identify any invalid codes or responses that are outside of the usual range. Unusual responses are reviewed by the data editors. Finally, when a data year is complete, there is a computerized check of all cases for invalid codes, inconsistent data, or unusual responses. 4.4.7 Additional Fields Although the TIFA survey adds valuable detail to the FARS data, additional data fields could add important detail about hazmat packaging and also enhance the information available on Database Analysis 71 Vehicle Driver Packaging Infrastructure Situational Configuration Age Package Type Road Surface Pre-Crash Condition Cargo Body Experience Quantity Shipped Road Condition Event Type GVW Condition Quantity Lost Road Type Vehicle Speed Valid License Age (Cargo Tank) Traffic Way Impact Location Citation Issued Rollover Protection Access Control Primary Reason Inspection History Speed Limit Accident Type Design Specification No. of Lanes Weather Condition Light Condition Time of Day Key: > 95% 50% to 95% < 50% Not captured Table 4-20. Completeness of contributing factors in TIFA/FARS.

crash causation. Some data fields could be added very easily, with little modification of the program. Including other data fields would take additional resources. Recommended additional fields are as follows: • The right-of-way data element could identify which vehicle, if any, within a crash had the right of way prior to the collision. This could be coded readily from the PAR in almost all cases. Some state crash reports include right of way on the report. Right of way would be useful in most crashes in identifying the vehicle that primarily contributed to the crash. • Critical event is a field that identifies and describes the event that precipitated the crash for the vehicle. This field is included in both the GES and CDS files. Coding manuals are available and could be used to ensure that coding is consistent with NCSA standards. • The managers of the TIFA survey might also consider adding a field for critical reason. Criti- cal reason captures the “reason” for the critical event, classified broadly as driver, vehicle, or environment, with detailed levels under each. The variable is useful for identifying the imme- diate failure that led to the crash and would shed light on crash causation. The field was used in the Large Truck Crash Causation Study (LTCCS), conducted jointly by FMCSA and NHTSA, and in the National Motor Vehicle Crash Causation Survey (NMVCCS), conducted by NHTSA. Therefore, coding procedures are available. The suggestion to add a critical rea- son data field to TIFA, however, is subject to whether the information is available within the TIFA protocol, which relies primarily on PARs, to code this field. The TIFA program could add the following additional information about hazmat cargo: • MC number of the cargo tank, which has been collected in the past as part of a special data col- lection effort and, therefore, the feasibility of collecting this information has been demonstrated. • Quantity of hazardous material transported, which would entail adding cargo weight data fields back to the survey. The program could consider capturing the quantity in terms of liquid measure, where appropriate. 4.4.8 Potential Measures to Improve Data Quality The TIFA system is complete and mature. It is subject to annual review and adjustment, including continuing training of the coders. However, greater cooperation with the FARS pro- gram might help increase the accuracy with which trucks are identified. 4.4.9 Compatibility with Other Databases The comments in Section 4.3.9 on the FARS data apply equally to TIFA, since TIFA uses the same case number system. However, there is an additional constraint on linking the TIFA file to other data systems, if that raises a risk of identifying specific individuals or organizations. The TIFA program is bound by commitments to respondents to protect their identity, and by the terms of its operations under the University of Michigan’s Institutional Review Board (IRB). Thus, any effort to link the data to other data systems would be unlikely to be allowed. 4.4.10 Data Uses FMCSA uses TIFA data for a variety of research purposes. In addition, the TIFA data are used by researchers at UMTRI and other universities for traffic safety research. The research is designed to identify the scope of traffic safety problems related to trucks and to identify risk fac- tors in truck crashes, whether related to the vehicle, driver, operation, or environment. This information is used by government entities—including FMCSA, NHTSA, and certain states— for regulatory purposes. 72 Hazardous Materials Transportation Incident Data for Root Cause Analysis

In terms of uses relating specifically to the hazmat data in TIFA, over time, FMCSA has pub- lished data about the scope of hazmat crashes. In addition, FMCSA has requested data about fatal hazmat crashes to monitor trends. 4.5 Large Truck Crash Causation Study (LTCCS) LTCCS was designed as a one-time study to compile a comprehensive set of accident data for approximately 1,000 large truck accidents. The data compilation began in 2001 and was completed in 2003, although analysis of the data is still ongoing. 4.5.1 Database Description The LTCCS database is a series of tables that can be run in several relational databases. For pur- poses of this assessment, the tables were imported into the Microsoft Access database. The mas- ter table is the Crash file and contains 1,070 records representing the number of discrete accidents that were investigated. There were 107 of these accidents that were eliminated for one reason or another because they did not meet the analysis criteria. Thus, the LTCCS evaluates 963 discrete accidents. The next table is the CRASHASSESSMENT Table, containing a total of 2,284 records representing the data for the number of vehicles involved in the 1,070 accidents. Although all of these accidents involved at least one heavy truck, the second vehicle in many of the accidents was a passenger vehicle. Because the purpose of the study was to collect sufficient data to identify the causes of the accidents as well as to collect information on the drivers, vehicles, and environment present at the time of the accident, a lot of supporting information is provided. The dataset contains a total of 43 separate tables summarizing the many factors that, in total, represent a comprehensive set of data for each accident (see Table 4-21). Table 4-21 includes Database Analysis 73 Overall Driver Vehicle or Packaging Carrier Environment or Situational Crash Driver Assessment Air Bags SAFER Authority Status Environment Crash Assessment Driver Decision Aggression Brakes SAFER Carrier Factor Assessment Events Driver Drugs Cargo Shift SAFER Insurance Injuries Health Crush SAFER Review SAFER Inspection Summary Driver Recognition Distraction General Vehicle SAFESTAT Rating Jackknife PAR Violations Hazardous Material Carrier Interviews (1 Table) Non-Motorist Sleep Hazmat Inspection Occupants Driver Interviews (14 Tables) Truck Exterior SAFER Crash Report Truck Inspection SAFER Driver Inspections Overview SAFER Driver Violations Truck Units MCMIS Driver Vehicle Events MCMIS Violations Vehicle Exterior Note: Safety and Fitness Electronic Records (SAFER) System; Safety Status (SAFESTAT) Management System. Table 4-21. LTCCS tabular structure arranged by category.

references to tables that summarize the results of driver and carrier interviews. Interview tables contain data on driver condition (aggressive driving, attention, condition, license status, fatigue, health, perception, and sleep). Interview tables also are provided for cargo shift, fire, jackknife, rollover, and trip. Most of the database tables are related using two parameters, the case number and the vehicle number. Driver information is collected for all 2,284 vehicles, but it was possible to collect infor- mation on driver health for only 1,839 vehicles, about 80% of all vehicles involved in the acci- dents. Other significant data are contained in the ENVIRONMENT and FACTOR_ASSESSMENT Tables, conveying information on the GENERAL_VEHICLE, VEHICLE, and TRUCK_UNIT Tables that provide additional details. As might be expected, although there are driver interview data for most of the 2,284 vehicles, the cargo shift data are not listed for all vehicles because cargo shift is not applicable to passenger vehicles. In total, cargo shift data are provided for 1,071 vehicles. Although very close to the num- ber of accidents investigated, 1,070, there are multiple trucks involved in many of these accidents, so the data are actually provided for about 90% of the heavy trucks involved in the 1,070 accidents. The database contains descriptions for 1,207 heavy trucks and 29 bobtails (power units without a semitrailer). The PAR_VIOLATION and BRAKES Tables list defects found at the accident scene. The latter comes from a Commercial Vehicle Safety Alliance (CVSA) Level 1 inspection of the vehi- cles following the crash. The CDCRUSH Table lists the type and extent of vehicle damage, some coming from accident reconstruction analyses. Although not many vehicles were carrying haz- ardous material, there are two tables, HAZMAT and HAZMAT_INSP, that provide information specific to the packaging and hazardous material involved in the accident. The LTCCS included 57 vehicles carrying hazardous material with 77 material types, and provided detailed event descriptions for 30 of the vehicles. 4.5.2 Purpose and Function The purpose of the LTCCS was to determine the causes of, and contributing factors to, crashes involving commercial motor vehicles. The study was mandated by the Motor Carrier Safety Improvement Act of 1999, P.L. 106-159. 4.5.3 Data Collection A nationally representative sample of large-truck fatal and injury crashes was investigated dur- ing 2001 to 2003 at 24 sites in 17 states. Each crash involved at least one large truck and resulted in at least one injury or fatality. Data were collected on up to 1,000 parameters in each crash. The total sample, after non-qualifying accidents were eliminated, involved 967 crashes, which included 1,127 large trucks, 959 non-truck motor vehicles, 251 fatalities, and 1,408 injuries. The data for each accident were collected from a wide variety of sources. These included the General Vehicle Form, the police accident report, medical reports, scene photographs, and post- accident inspections, including the CVSA Level 1 inspection of the truck involved in the accident. In addition, interviews were conducted with the drivers, witnesses, vehicle occupants, and carrier personnel. Local weather station data was used to describe weather conditions and the driver’s log was used to determine hours of service. AASHTO documents provided the criteria used to deter- mine the values that should be assigned to parameters such as the driver’s line of sight at the time of the accident. Time-stamped toll and fuel receipts were also collected. Relevant data were also obtained from federally maintained databases. These databases included the MCMIS Registration File, and the Safety and Fitness Electronic Records (SAFER) system and Safety Status (SAFESTAT) Management System databases. Although the data collection involved using trained investigators 74 Hazardous Materials Transportation Incident Data for Root Cause Analysis

to visit the site of each accident, the data collection was performed in phases so the highway where the crash occurred would not be unduly blocked while all the data were collected. The CVSA Level 1 inspection was conducted at the repair facility, not at the accident scene. Advanced photographic techniques were used to enable the compilation of scaled schematics and scene measurement logs. The typical approach was to mark key points in the accident progression while the vehicles were present and then go back later and take more extensive measurements. In addi- tion, the accidents that were selected occurred close to the 17 locations where trained investiga- tors resided. No effort was made to require that the investigators travel great distances, thereby forcing long-term closure of the highway. Simply stated, the data collection and compilation were designed to minimize disruption yet, at the same time, collect data on many relevant parameters. 4.5.4 Data Compilation The data from interviews, photographs, accident scene measurements, and vehicle inspections were used to populate many of the parameters in the database. For example, the interviews with the carrier and driver were used to compile data on the driver’s previous sleep interval, the hours of service recorded on the log, as well as data on the driver’s mental and physical state. Similarly, the measurements taken at the scene were used to generate scaled schematics and the scene meas- urement log. Data from the interviews of the driver and carrier were used to identify the driver aggression and driver distraction factor number. Photographs and on-scene measurements were compiled into deformation logs to be placed in the database as deformation codes. 4.5.5 Accuracy and Completeness of Data Every effort was made to obtain a comprehensive set of data for more than 1,000 parameters. Many of the parameters were estimated from multiple sources of data and apparent or real incon- sistencies could be resolved normally, thereby producing a consistent dataset for each vehicle involved in the accident. In addition to the 43 tables in the database, there are numerous support- ing tables that define code numbers to be used instead of phrases or words. This increased the accuracy of entry among data compilers. In this regard, the LTCCS project generated a 512-page Analytical Users Manual to ensure that all of the parameters’ codes were consistently entered into the database tables. The lengthiness of the manual is due primarily because a definition, source, cross reference, variable name, and attribute code ID was provided for each parameter. 4.5.6 Quality Control Extensive quality control checks were performed to ensure the accuracy of the data put into the database. The use of attribute codes that are defined in the Analytical Users Manual greatly enhances the quality of the data. Based on the manual, codes are used for most parameters and since the manual defines the meaning for each of the numerical codes, there is little room for ambiguity. This minimizes the inconsistencies in the dataset. 4.5.7 Interconnectivity with Other Databases It is not possible to connect the data in the LTCCS with other databases because the loca- tion and day of the month in which the accident occurred has been removed from the pub- lic version. The carrier is not named, the DOT number of the carrier is not given, and the vehicle registration number has been shortened. Any of these parameters would enable the datasets to be joined. There are only 57 crashes that involved hazardous material and, since HMIRS mainly reports spills, there are probably fewer than 10 crashes that might be reported in both datasets. Database Analysis 75

4.5.8 Analyses Using Database The LTCCS raw dataset was presented to the analytical community in 2006 and numerous analyses have been performed on the dataset. The analyses presented here will focus on the haz- mat truck accidents. Table 4-22 shows the breakdown of hazmat shipments included in the LTCCS. If a reportable quantity was being shipped, then the shipment would have to be plac- arded. There are only 40 placarded vehicles analyzed in the LTCCS. Table 4-22 shows that slightly more than 40% of the reportable shipments, 17, were Class 3 materials. Class 2 was the next most common, with 8 vehicles out of the 40, about 20% of the total. Thus, Class 3 and 2 shipments make up more than 60% of the total hazmat vehicles included in the LTCCS. For the 40 reportable accidents, the database can be queried to look at health factors, as shown in Table 4-23. Of the health factors listed in Table 4-23, other than requiring corrective lenses, almost all of the entries identify no health factors that might have contributed to the accident. To get better statistics for health issues that could affect safety would require a health assessment to be col- lected for at least 400 or more hazmat incidents. With 400 drivers, it might be possible to address the contribution from heart attacks. More than 1,000 would be required to get valid statistics on less common health conditions. This implies that if driver health is a contributing cause, it prob- ably has to be captured in all hazmat truck accident records, as it was a few years ago in MCMIS, and for some reason has been left blank in MCMIS beginning in CY 2002. Drug use by the driver also was tabulated. In the 40 drivers hauling hazardous materials, there were 10 drivers taking prescription drugs and 6 taking over-the-counter drugs. There were no instances where the driver was taking illegal drugs. In the 40 accidents involving a hazmat vehi- cle, in three of the accidents, the driver of the other vehicle was believed to be taking illegal drugs. A drug test verified the presence of the drug in one case and in the two others, the results of the drug test was unknown. 76 Hazardous Materials Transportation Incident Data for Root Cause Analysis Reportable Quantity Specified in 172.101 Hazardous Material Table [49 CFR 172.101] Material Yes No Unknown 2.1 Flammable Gas 5 1 2.1 LPG 1 2.2 Nonflammable Gas 2 2 3 Combustible Liquid 5 3 Flammable 12 3 2 4.1 Flammable Solid 1 4.3 Dangerous When Wet 1 5.1 Oxidizer 1 6.1 (Liquids) 1 1 6.1 Zone A 1 8 (PIH) Zone A 1 8 Corrosive Material 6 1 9 (Elev Temp Materl) 1 9 (Hazardous Waste) 1 9 Miscellaneous HM 1 1 Total 40 9 2 Table 4-22. Types of hazardous materials included in vehicles in LTCCS.

The ENVIRONMENT Table provides a lot of information that is useful for defining the char- acteristics of the accident location. Table 4-24 shows the relationship between the JUNCTION and INTERCHANGE parameters for the 40 placarded trucks included in the LTCCS dataset. As shown in Table 4-24, the data are internally consistent. There are no entries with INTERCHANGE = Yes that are not entered under a JUNCTION category related to an inter- change. The data also show that of the hazmat truck accidents, more than 25% (11/40) occur at interchanges. Table 4-25 looks at the same data using the parameters INTERCHANGE and FUNCTIONAL_CLASS. Table 4-24 shows that the 11 hazmat trucks in the LTCCS with INTERCHANGE = Yes are all associated with Interstate highways. Table 4-25 shows that there are also 16 additional accidents on Interstate highways not associated with interchanges. This means that of the 27 crashes involving hazmat trucks, 40% of the accidents are at interchanges. Given that interchanges occur only every few miles, on a per mile basis, accidents at interchanges on Interstate highways clearly dominate. Although one might be tempted to look at the ratio of urban and rural freeway accidents from these data (2 rural and 25 urban), the fraction of the miles driven in urban and rural areas is not known. Thus, it is difficult to infer an accident rate from these data alone. In HMIRS, the origin and destination of the shipments is shown and if these data could be matched with HMIRS records, it would be possible to derive an estimate of these rates. There is another factor that could enter into the analysis as well. In the LTCCS, the accidents were not selected randomly; they had to be close to the location of 17 accident investigation teams and, as a result, the acci- dents selected could be biased toward urban accidents. Database Analysis 77 Health Factor Total Health Factor Total Illness Factor Count 1 Astigmatic 2 Heart Attack 1 Other Vision 6 Epileptic Seizure 0 Unknown Vision 7 Diabetic Blackout 0 Other Factor Count 1 Other Blackout 0 No Factors 32 Cold Flu 0 Hearing Impairment 0 Other Illness 1 Prosthesis 0 Normal Vision 17 Paraplegia 0 Legally Blind 0 Strenuous Recreation 0 Myopic 5 Strenuous Non-Work 1 Hyperopic 4 Sleep Apnea 0 Glaucoma 0 Other Factor Physical 0 Color Blind 0 Table 4-23. Listing of health factors present for vehicles in LTCCS containing hazardous materials. Interchange Junction No Yes Entrance/exit ramp related 1 7 Intersection 2 Intersection related 2 Non-junction 22 Other location in interchange 4 Rail grade crossing 2 Table 4-24. Relationship between junction and interchange in LTCCS.

In a typical data analysis, it is difficult to analyze just a few accidents. Thus, while it is possible to look at the decision factors associated with those 11 interchange accidents, the statistical uncertainty regarding the conclusion will be very high. Clearly, to identify significant differences would require more hazmat truck accidents in the dataset. Just as the LTCCS targeted large truck accidents involving a serious injury or fatality, a com- parable study that focused on hazmat accidents would provide a similar benefit. Rather than doing a two-year study of 1,000 truck accidents, there appears to be merit to doing a continual study of fewer truck accidents, perhaps 100 to 200 per year. To look for differences between haz- mat truck accidents and regular truck accidents, it would be important to have data for both, perhaps a sample of 100 regular truck accidents and 100 hazmat truck accidents. If such a study were performed on an annual basis, it is important to have weighting factors to enable the find- ings from a limited sample of accidents to be related to the universe of accidents occurring annu- ally. These can be developed as part of the sampling methodology or come from other databases such as MCMIS and HMIRS. 4.5.9 Summary and Potential Measures to Improve Root Cause Analysis The analysis of the data from the LTCCS is still ongoing, so the following summary is based on its status as of the time of this report. The potential measures are prepared to focus on the objectives of this project. 4.5.9.1 Summary The LTCCS represents a comprehensive analysis of serious, large truck crashes. The variables captured in the 967 accidents investigated by contributing cause category are shown in Table 4-26. As shown, all of the contributing factors listed under the categories for Vehicle and Situational and most of the contributing factors under the categories for Driver and Infrastructure are covered. The Infrastructure category’s factors are actually known by the LTCCS analysts, but have been coded to prevent these data from being known by those outside the LTCCS program. Thus, the training and experience of the driver were the only contributing causes that are not captured under the Driver category. The Packaging category is not well captured, since package behavior was not the focus of the LTCCS. 4.5.9.2 Potential Improvements Based on the LTCCS Experience Comprehensive studies, such as the LTCCS, are needed to obtain contributing and root causes of accidents. Similar to the LTCCS, these detailed analyses can be focused on a sample of all the accidents occurring in the United States, provided that the weighting of the sampling is known. 78 Hazardous Materials Transportation Incident Data for Root Cause Analysis Interchange Functional Class No Yes Rural local 2 Rural minor arterial 4 Rural principal arterial – Interstate 1 1 Rural principal arterial – other 3 Urban minor arterial 2 Urban principal arterial – Interstate 15 10 Urban principal arterial – other 2 Table 4-25. Relationship between functional class and interchange in LTCCS.

If a comprehensive study of selected classes of accidents is performed, there are significant advantages to performing a selected number of accident investigations annually rather than per- forming a larger intensive study over a one- to two-year period as was done for the LTCCS. The advantage of continuing studies is that the investment in training can be realized over many years, and trends in the data also can be evaluated. Any program for identifying root and contributing causes could, for selected classes of accidents 1. Obtain data taken from interviews with drivers and other witnesses with information about the crash. There are many factors that cannot be obtained unless post-vehicle-inspections and reviews of driver qualifications are conducted. 2. Obtain data collected from SAFER, SAFESTAT, and the MCMIS Registration file. Such data would provide information on the programs to ensure that hazmat is being transported using well-trained drivers in safe vehicles. 3. Visit the scene of the accident to obtain first-hand observations, including photographs of the scene, vehicle damage, scaled measurements, and a scene measurement log. Such informa- tion is critical for accident reconstruction and also to resolve any data inconsistencies. 4. Consider the techniques developed in the LTCCS. It is not considered feasible to perform the level of analysis performed in the LTCCS on all hazmat accidents. However, just as the LTCCS looked at injury and fatality accidents involving heavy trucks in a selected area, so it would be possible to perform the LTCCS level of analysis for perhaps 50 to 100 hazmat accidents annu- ally, perhaps those involving a specific type of hazardous material (e.g., TIH [toxic inhalation hazard] cargo tank shipments). Such analyses could be multimodal if accident investigations were coordinated. 4.6 Railroad Accident/Incident Reporting System (RAIRS) Prevention of hazmat releases caused by railroad accidents differs from other modes in a vari- ety of respects due to physical, operational, and institutional factors. Some of the important dif- ferences include Database Analysis 79 Vehicle Driver Packaging Infrastructure Situational Configuration Age Package Type Road Surface Pre-Crash Condition Cargo Body Experience Quantity Shipped Road Condition Dangerous Event GVW Condition Quantity Lost Road Type Vehicle Speed Vehicle Defect Valid License Age (Cargo Tank) Traffic Way Impact Location Vehicle Response Citation Issued Rollover Protection Access Control Primary Reason Driver Response Inspection History Speed Limit Accident Type Training Design Specification No. of Lanes Weather Condition Location Light Condition Time of Day Health Consequences Key: Variable obtained Partially met Not captured Table 4-26. Summary of variables captured in LTCCS.

• Railroads typically operate trains rather than single vehicles; • Operation is restricted to a fixed guideway or “single-degree-of-freedom” system; • Railroad track age and other infrastructure is generally privately owned and maintained; • Much of the operational control is either automated or controlled by individuals other than the train operating crew; • Many railcars spend a substantial fraction of the time operating on and under the control of a railroad other than its owner; and • These factors all have the effect of reducing certain general types of failure that can lead to a hazmat release, while elevating the importance of certain others. Although human error is an important cause of railroad accidents, failure of infrastructure or vehicle components comprise a much larger percentage of hazmat accidents. Additionally, fail- ures in the traffic control system sometimes cause accidents. These may be mechanical or electri- cal in nature, or they could be caused by human errors committed by personnel other than those operating the train, and who may be located hundreds of miles from the scene of the accident. To support this type of approach, FRA and the railroads have a comprehensive accident reporting system that has roots dating back to 1910 and was implemented in its present form beginning in 1975. FRA regulations require that all accidents in which damage to track and equipment exceeds a specified monetary threshold (adjusted periodically for inflation) must be reported using Form FRA F 6180.54, the Rail Equipment Accident/Incident Report, which records 52 different variables regarding the circumstances and cause of the accident. Beyond this, major railroads maintain their own internal databases. These typically contain all of the infor- mation necessary to comply with FRA reporting requirements, and may contain additional data that individual railroads believe is useful for their own safety analysis purposes. These efforts are significant to root cause analysis in several respects. At the most proximate level, the FRA reporting requirements ensure that all accidents of consequence are subjected to an analysis of the circumstances of the accident, and that both primary (and if applicable, secondary) causes of the accident be determined and reported to FRA. In some cases, these may require fairly intensive analysis of the accident scene if there is some uncertainty about the cause, and major railroads employ specially trained individuals responsible for this func- tion. Understanding all of these aspects is pertinent to root cause analysis of railroad-accident- caused hazmat releases. In the following section, the nature and character of basic elements are described. FRA divides accident causes into five broad categories: • Track, roadbed, and structures, • Signal and communication, • Mechanical and electrical failures, • Train operation—human factors, and • Miscellaneous causes not otherwise listed. Within each of these categories, there may be sub-categories and then, at the most detailed level, specific cause codes. 4.6.1 Track, Roadbed, and Structures The most frequent cause of railroad accidents, and accidents resulting in a hazmat release are failures of the track system, especially rail failure due to various forms of fatigue-induced frac- ture. Railroads conduct frequent inspection of rails to find and remove defects; however, certain types are difficult or impossible to detect using existing technology. A number of other infra- 80 Hazardous Materials Transportation Incident Data for Root Cause Analysis

structure failures also can cause serious accidents. The second most common track-caused haz- mat release accident is track geometry, followed by roadbed problems, and switch and frog prob- lems. Railroads use a combination of manual and automated inspection technologies to detect problems before they become critical, but some are not found and derailments can occur as a result. Overall, the FRA has more than 65 different cause codes for railroad-track-caused acci- dents. This enables a very fine-grained ability to analyze which causes are the most important contributors to hazmat accidents. For both rolling stock and infrastructure, the American Association of Railroads (AAR), FRA, and Class 1 railroads are conducting or sponsoring research and development of better designs, materials, and operational practices that will be more resistant to failure. In parallel, they are also conducting research and development on an array of technologies intended to improve the inspection capability for a wide range of possible defects. 4.6.2 Signal and Communication Accidents caused by signal and communication failure rank last among major categories of accidents and as a cause of hazmat releases. Unlike highways, virtually all railroad operations take place in a highly controlled environment. Specific rules and protocols apply to operation on all portions of the railroad. Communications and signals (C&S) are an essential element of these systems whose purpose is to ensure safe and efficient operation of the railroad. If some element of these systems malfunctions, it may result in incorrect or incomplete information being trans- ferred to or from the train, thereby creating the potential for conflicting track occupancy author- ities or excessive speed. Under these conditions, the consequences may often be a collision or derailment. Railroad C&S systems are thus designed to be extremely robust and embody exten- sive fail-safe elements in their design (i.e., if they fail, it results in a “safe” condition, indication, or message). Consequently, railroad accidents attributable to C&S failures are rare. In a recent study, they accounted for only 3/10ths of 1% of all the U.S. railroad mainline accidents. Never- theless, when such failures do occur, the resultant accidents tend to have high consequences because the outcome will often be a collision or overspeed derailment, thereby resulting in rela- tively large impact forces. If hazardous materials are involved, there is a reasonably high poten- tial to breach the car transporting them and cause a release. 4.6.3 Mechanical and Electrical Failures Accidents caused by mechanical and electrical failure are the second most common major cat- egories of accident cause, and third overall in causing hazmat releases, However, when one con- siders only mainline-accident-caused hazmat releases, they rank second. Railroads operate trains with hazardous materials in the consist, which ranges greatly in number of vehicles and length. The consist is defined as the group of rail vehicles that make up a train. These trains may have less than a dozen cars or more than 150, ranging in length from a few hundred feet to nearly two miles. This has a variety of implications in terms of the occurrence of accidents, and the conse- quent approaches to root cause analysis. With approximately 1.5 million railcars and approximately 800 different owners, railcars spend a great deal of time operating on railroads and by companies other than their owners. Fre- quently, repairs must be conducted on the road by someone other than the railcar’s owner. Rail- cars have not generally been subject to programmed maintenance in North America. Instead, railroads and car owners have operated under a philosophy of run to (near) failure. The objec- tive is to obtain as much life as possible from components without suffering failure. Due to the frequent and redundant inspections railcars receive as they move from terminal to terminal dur- Database Analysis 81

ing their journeys, most failing components are indeed found before they cause a problem, but a few are missed and these sometimes result in an accident. Each of the railcars in the train is subject to dozens of different failure modes with the poten- tial to cause an accident. Although terminal personnel and the train crew are responsible for inspecting key attributes of the cars before departure and at certain intervals during a journey, many problems are difficult to detect or can become critical en route. The nature of trains is that crew members are separated from most vehicles by a considerable distance and will often be unaware of an incipient failure until it has already occurred. Consequently, railcars are necessarily robust and, wherever possible, designed in a fail-safe manner. As mentioned above, railcars undergo human inspection in terminals and, in addition, railroads rely heavily on a variety of automated technologies to detect certain types of vehicle failure. The industry is aggressively developing new technologies to expand this capability to other components and failure modes. Overall, FRA has more than 140 different cause codes for railroad-equipment-caused accidents. 4.6.4 Train Operation—Human Factors Accidents caused by human-factors rank third among the major accident categories in terms of frequency but second in terms of causing hazmat releases. The majority of these are accidents in yards and industry tracks, although mainline collisions are a particularly important cause for the reasons discussed below. Human-factors accidents vary widely in their severity and poten- tial to cause serious hazmat releases. Among the most common human-factors-caused accidents are various errors committed during switching, such as run through switches. These are gener- ally low-speed incidents with little potential to cause sufficient damage to a hazmat car to pro- duce a serious release. On the other hand, accidents caused by failure to obey signals on the main- line or other operating instructions can result in high-speed collisions with substantial potential to breach one or more hazmat cars. Several of the most serious hazmat release accidents in the past few years have been due to such failures. Both FRA and NTSB have placed a high priority on developing technology and implementing requirements for adoption that are intended to prevent certain types of human-factors accidents from happening. Notable among these is the recent rulemaking requiring implementation of Positive Train Control (PTC) on all rail lines handling toxic inhalation hazard (TIH) materials. 4.6.5 Summary of Causes and Impact The points discussed are pertinent to the root cause analysis objectives of this project because there are a wide variety of possible causes, any one of which occurs relatively infrequently. These incidents are distributed over 150,000 miles of railroad lines and 1.5 million freight cars. Con- sequently, the rate of failure for any particular component or system is relatively low and dis- persed across a large system. In order to understand the principal factors most likely to result in conditions that can lead to a hazmat release, a statistical approach is needed. In the context of understanding the contribution of the current FRA Guide for Preparing Accident/Incident Reports (FRA 2003) to root cause analysis, it is worth reviewing a few of the cat- egories that railroads are required to provide: Item 38—Primary Cause Code, Item 39—Contribut- ing Cause Code, and Item 52—Narrative Description. Item 38—Primary Cause Code Proper entry of the correct primary cause code is of critical importance, not only for the accident being reported, but also for FRA’s analyses conducted for accident prevention 82 Hazardous Materials Transportation Incident Data for Root Cause Analysis

purposes. Because of the extensive use made of primary cause code entries, careful attention must be given to making correct entry for all accidents. (FRA 2003a, p 11) As stated by FRA, this code is critically important to “accident prevention analysis,” which is implicit in root cause analysis. There are several additional paragraphs providing more detail about the factors railroads should consider when identifying and possibly updating the Primary Cause Code as more information becomes available. Item 39—Contributing Cause Code If there were one or more contributing causes, enter the code for the foremost contributing cause. Otherwise, enter “N/A.” An accident is frequently the culmination of a sequence of related events, and a variety of conditions or circumstances may contribute to its occurrence. A complete record of all of these would be beneficial in accident prevention analysis. However, it is not practical, even if it were possible, to develop forms and codes that would capture every detail that may be associated with the causes and resulting consequences of each accident. Therefore, the most appropriate combination of available codes that best identifies the likely primary and any contributing cause, and other factors, is to be used. Railroads are encouraged to use the Contributing Cause Code. When the events cannot be adequately describe[d] using the Primary and Contributing Cause, the railroad must use the Narrative Block to complete the causes of the accident. (FRA 2003a, p 13) As discussed in this report, accidents are often the result of more than one factor. FRA explic- itly recognizes this elsewhere in their discussion of the accident reporting and analysis process and, by providing Item 39—Contributing Cause Factor, they allow for one contributing cause to be identified. In the context of root cause analysis, this may be one area for improvement. FRA states that more than one contributing cause may be a factor and asks the railroad to enter the “foremost contributing cause,” implying that only one be identified. It seems feasible that the process could be modified to allow for multiple (perhaps up to three) contributing causes to be identified, with a requirement that they be rank-ordered in importance (i.e., Contributing Cause 1, 2, and 3). However, this would often require a certain amount of subjectivity in ranking the causes and different individuals or railroads might use different criteria, thereby introducing additional variability. Furthermore, the notion of primary and contributing causes may have a tendency to over-simplify the process. It may be worthwhile to consider proximate and ultimate causes with more sophistication. Another potential measure could be to evaluate whether this aspect of the reporting system can be modified to enhance the value of the information. In recognition of the potential complexity of factors contributing to an accident, FRA requires that railroads provide additional details in Item 52—Narrative Description. FRA’s general instructions are as follows (FRA 2003a, p 15): A detailed narrative is basic to FRA’s understanding of the factors leading to, and the consequences aris- ing from, an accident. While many minor accidents can be described in a few brief comments, others are more complicated and require further clarification. In addition, FRA specifically requests that information be provided on drug/alcohol involve- ment, cause, diesel fuel tank, hazardous materials, and other railroads. Of these, the following are of particular relevance to hazmat root cause analysis and the text for each is as follows (FRA 2003a, pp 15–16): Drug/alcohol involvement—Include a discussion of any drug/alcohol use connected with this accident. If positive tests were made, but usage/impairment was not determined to be a causal factor, explain the basis of this determination. Cause—Discuss any event(s) or circumstance(s) occurring prior to the accident that has relevance to the accident. Provide additional information concerning the reasons(s) for the accident when the causes found in Appendix C do not sufficiently explain why the accident occurred. Hazardous Materials—Identify the initial and number of any car releasing hazardous material. List the name and indicate the quantity of hazardous material released. Report the number of fatalities and injuries resulting from a direct exposure to the released substance. If there was an evacuation, estimate the size of the affected area and the length of the evacuation. Database Analysis 83

4.7 Marine Information for Safety and Law Enforcement (MISLE) The MISLE database supports the Marine Safety and Operations Programs. MISLE contains vast amounts of data, including detailed vessel characteristics, cargo carriage authorities, involved party identities, bridges, facilities and waterways, and records of related Coast Guard activities. MISLE activities include law enforcement boardings and sightings, marine inspections and inves- tigations, pollution and response incidents, and search and rescue operations. In addition, MISLE manages the information flow involving the administration of all of these activities, from the ini- tial triggering event, to incident management and response, and the resulting follow-on actions. Its development was initiated in 1992 and it became fully operational in January 2002 when the Coast Guard transitioned from the Marine Safety Information Reporting System. 4.7.1 Database Description The database is logically broken into a relational table structure that contains an activity table that includes all of the incidents reported to MISLE. As the example in Figure 4-4 shows, there are tables presenting an inventory of facilities and vessels that can be tied to the Activity and Events Tables. The activities are joined to the Facility Events and Vessel Events Tables, which provide additional information on the activity reported to MISLE. These, in turn, are joined to Facility and Vessel Pollution Tables that are also joined to an Injury Table that lists all of the reported injuries and fatalities associated with the activities. The pollution activities are a very small portion of the activities reported to MISLE. Commer- cial, as well as pleasure boat, collisions and groundings are reported. If a reportable quantity of a hazardous substance is released (40 CFR Part 302), the National Response Center (NRC), which is administered by the Coast Guard, must be notified promptly, and the vessel operator must fill out Form CG-2692 and submit it to the Coast Guard to document the event. Note that the reportable quantity is determined using the EPA list of hazardous substances, which also includes marine pollutants. 4.7.2 Purpose and Function The purpose of the MISLE database is to maintain a comprehensive record of vessel, facility, and Coast Guard activities related to commercial shipping. Incidents resulting in the loss of life to the public from private boating activities are also included in the database. The information system contains links to other resources so that Coast Guard personnel can respond quickly to any major incident. The database part, which is the focus of this discussion, reports all vessel or facility incidents related to commercial shipping. The documentation of pollution events, while significant, represents only a small fraction of the documented incidents. The MISLE system maintains a log documenting the status of all judicial activities associated with the documented incidents. The record of any incident with an ongoing judicial action is not available publicly until the case is closed. Since cases are commonly kept open for several years, a comprehensive picture of the number of pollution events occurring in a given year is difficult to identify from the publicly available file. The focus of much of the monitoring activities relates to efforts to speed up judicial actions so that cases can be closed more rapidly. 4.7.3 Data Collection If a reportable event occurs, the vessel operator must fill out Form CG-2692 and submit it to the Coast Guard to document the event. There also are cases where Coast Guard personnel file an event report using CG-2692. Once filed, the Coast Guard accident investigators update the file as the investigation proceeds. 84 Hazardous Materials Transportation Incident Data for Root Cause Analysis

4.7.4 Data Compilation Data from the accident form are transferred to the appropriate fields in the MISLE database. As the status of the investigation proceeds, the data are updated. 4.7.5 Accuracy and Completeness Assessing the accuracy of the entries in the MISLE database is difficult for outside investiga- tors. Since the majority of the events entered into MISLE involve legal action taken on the part of the Coast Guard, the accuracy of the entered data is likely to be very high. The completeness of the data entry is difficult to judge as well. Although all of the fields have entries, in many cases Database Analysis 85 Figure 4-4. Example of tabular relationships in MISLE.

they are filled out with a standard term, such as “not noted” and it is not known if an entry should have been made in the field. 4.7.6 Quality Control No assessment was made as to the use and effectiveness of MISLE quality control procedures. 4.7.7 Interconnectivity with Other Databases There are no common fields that would enable this database to be connected easily to other databases. The MISLE record does contain the date, time, and location of the accident, expressed as the latitude and longitude, so if another database like HMIRS reported the same information, linkages could be made. Since the carrier name is a difficult field to join on (because the name must be identical right down to the spelling of the name, abbreviations, spaces, and periods) automatically connecting database fields would be very difficult. A brief attempt to connect the HMIRS reports listing mode as water, identified no events reported to both databases during a one-year period. 4.7.8 Analyses Using Database Since the MISLE events could not be compared with the hazmat incidents reported in HMIRS with water as the mode, no analyses were performed. 4.7.9 Summary and Potential Measures for Improving Root Cause Analysis Much of the MISLE database is accessible only to Coast Guard staff. Furthermore, the MISLE data become available to the general public only for closed cases and it can take several years to close many of the MISLE-reported incidents. This might be one of the reasons why it was not possible to find common events reported to both HMIRS and MISLE. Lack of timeliness, access, and interconnectivity are considered insurmountable barriers for MISLE use. Any database used for root cause analysis should provide timely accident reports, have unrestricted access, and be able to easily connect reports made to other databases. 4.8 NTSB Accident Investigations and Reports 4.8.1 Scope of Investigations There are numerous NTSB investigations of individual accidents. While all commercial aircraft crashes are included, there are certain rail and truck accidents that are also selected for investiga- tion by NTSB. This section will focus on one particular type of NTSB investigation—passive pri- vate grade-crossing accidents. A study (NTSB 1998) summarized the investigation of 60 passive grade-crossing accidents that occurred between December 1995 and August 1996. The accidents investigated were not selected on the basis of establishing statistical confidence. The criteria were that damage to the motor vehicle had to be serious enough for the vehicle to require towing from the scene and that it had to occur close enough to an NTSB regional office for an investigator to travel to the site before the vehicle was towed from the scene. The NTSB investigator recorded the types of signage present, as well as the characteristics of the grade crossing, and obtained witness statements. 86 Hazardous Materials Transportation Incident Data for Root Cause Analysis

4.8.2 Approach to Identifying Root Causes The approach taken in all NTSB investigations is similar, although the scope of the individual pieces may vary. In all cases, a team of NTSB accident investigators is dispatched rapidly to the scene of the accident to collect evidence that might be usable in determining contributing causes. The evidence collected includes data describing the accident scene, the amount of damage to equipment, and the extent of injuries to individuals involved in the accident. Witness statements are always collected and it is pointed out in several investigations that it is important to get those statements quickly because witness memories fade. The NTSB then goes through an extensive analysis of the collected data and will frequently follow up with requests for additional informa- tion as the analysis proceeds. By going to the scene, the NTSB has all the contact information needed for follow-up purposes. The NTSB investigator collects the following information: • Location, • Date and time, • Lighting conditions, • Type of motor vehicle (year and type), • Train action reported (horn sounded/auxiliary lights on), • Signs present (crossbuck, advance warning, and/or stop sign), • Physical characteristics (limited sight distance, angle of intersection, road or track curve, and presence of a nearby road intersection), and • Number of injuries and fatalities. All of these items, except some of the physical characteristics, are included in the grade-crossing incident report submitted to FRA. The FRA accident database does not document the proximity of the grade crossing to other roads nor does it document whether the road or track is curved. The curvature of the road or rail track was not listed as a probable or contributing cause in any of the 60 cases (NTSB 1998), and there was only one case where the presence of traffic plus a nearby intersection was listed as a contributing cause. Thus, the absence of this information being cap- tured in the FRA database is not considered to be a significant weakness. Whereas the NTSB find- ings are based on a site visit and witness statement, FRA does not require either of these. While there are narrative fields provided in the database, during the time period of the NTSB study, the narrative fields were left blank for all grade-crossing accidents for both active and passive grade crossings—more than 3,000 reports. The analysis performed here included an additional task that was not performed by NTSB— a comparison of the data reported to FRA and the data reported by NTSB for the same accident. Although reports for all 60 of the accidents were found, the matching task, initially thought to be easy, turned out to be a challenge for the following reasons: • NTSB did not include the FRA incident number, so it had to be discovered. • NTSB listed the closest town to the grade crossing whereas FRA reports the nearest timetable station, county, and city (if the accident occurred within city boundaries). For almost one-third of the cases for which matches could be found, it was necessary to refer to a map to find the location of the grade crossing, a time-consuming process. If both databases had provided the GPS coordinates (an option in the rail equipment accident database), then the month, day, and GPS coordinates would have made it much easier to match the accidents in the two databases. Database Analysis 87

4.8.3 Insights for Analyzing Root Cause The insights are divided into three categories: (1) data quality, (2) probable cause findings, and (3) summary. These categories are described in the following subsections. 4.8.4 Data Quality A comparison of the NTSB (1998) and FRA (2003) data revealed many significant differences. In five cases, there were differences in the number of fatal and non-fatal injuries. FRA establishes explicit reporting requirements and it is not known if NTSB followed the same requirements. For example, if a medical examiner determined that the driver committed suicide, the railroad would not have been required to report the fatality to FRA. Although FRA requires that the rail- road file an amended report if the person dies within 180 days of the crash, if the railroad does not follow the progress of the injured person, then they would not know if the person died within that time period and this would not be reported to FRA. Whatever the reasons, the differences were not small. For Case 3, the FRA database stated no fatalities and NTSB reported 3 fatalities. For Case 7, the FRA database stated 1 fatality and 3 injuries whereas NTSB reported 3 fatalities. In Case 26, the FRA reported 2 injuries and NTSB reported 12 injuries. In Case 55, the FRA data- base stated no injuries and NTSB reported 6 injuries. In Case 60, the FRA database reported 1 injury and 1 fatality and NTSB reported 2 fatalities. Since the NTSB finding is always more severe, one can assume that NTSB did more follow-up in the period after the accident to more accurately reflect the number of fatal and non-fatal injuries. There were 4 of the 60 cases where no corresponding FRA report could be found at the date and location listed in the NTSB report. In one case, the accident was found at the same time and location, three days later. For the other three, no report could be found. In one case, the vehicle attempted to cross at an abandoned grade crossing and that accident, while classi- fied as a grade-crossing accident by the NTSB is classified as an obstruction accident by FRA for which a grade-crossing accident report need not be filed. For the remaining two, no FRA report corresponding to the NTSB report could be found. There were large differences reported for other parameters as well. The view of oncoming trains in the vicinity of the grade crossing showed the largest discrepancy (see Table 4-27). As shown in Table 4-27, NTSB separated its assessment of obstructed view into two parame- ters, the view as the motor vehicle approached the crossing and the view of the oncoming train 88 Hazardous Materials Transportation Incident Data for Root Cause Analysis NTSB Findings Sight Distance Limited on Approach Sight Distance Limited at Stop Line FRA Database View Parameter Number of Cases Yes Yes Topography 1 Yes Yes Vegetation 1 5Yes Yes No Obstruction Yes No Permanent Structure Yes No Vegetation 1 1 Yes No No Obstruction 24 No Yes No Obstruction No No Passing Train 1 1 No No No Obstruction 22 Table 4-27. Query on obstructed vision.

at the stop line. The FRA report has only one parameter, limited sight distance. Based on the def- inition in the FRA reporting guide (FRA 2003), the view as the driver approaches the crossing, the first NTSB parameter, should be reported in the FRA database. The two assessments agreed in 27 cases, less than half the 57 for which a comparison was possible. In 22 of these cases, both assessments agreed that there were no obstructions. In one of five remaining cases, the obstruc- tion was a passing train, a factor not considered to be an obstruction by NTSB. In the four remaining cases, both recorded that an obstruction to the driver’s vision was present on approach. That leaves 30 cases where the two assessments disagree. In five of the cases, the NTSB investigators concluded that the driver’s view was obstructed both on approach and at the stop line—at the stop line for one case, and for the remaining four on approach. It seems clear that the NTSB investigators and the employees filling out the FRA grade-crossing report differ on the definition of what constitutes an obstruction as a driver approaches the grade crossing. Given that the NTSB investigator visited the scene, the NTSB assessment is thought to be a more accu- rate assessment. Differences also were observed in the types of signage at the grade crossings. In 10 of the 57 cases, the FRA database stated that no stop sign was present whereas the NTSB report stated one was present. In 4 of the 57 cases, the FRA database shows there was a stop sign when the NTSB investigator reported none was present. Being inaccurate in 25% of the cases makes it dif- ficult to draw conclusions regarding the effectiveness of stop signs at passive grade crossings. Because the NTSB investigators take statements from witnesses and observe the conditions at the accident scene at the time of the accident, they clearly have an advantage regarding reporting accuracy. There were cases where the NTSB reported a building, a large pile of rocks, overgrown vegetation, and cars on the tracks as blocking the vision of vehicles approaching the crossing. Clearly, conditions change at grade crossings and if there is no way to capture those changes in the data being used by the railroad employee filling out the FRA grade-crossing report, the FRA report will always be less accurate. 4.8.5 Probable Cause Findings The NTSB (1998) report recognizes that determining the probable and contributing causes of passive grade-crossing accidents is a challenging task. It is even more difficult to summarize the findings based on only 60 cases. The following discussion summarizes the findings, recognizing that no finding is likely to be statistically significant. There were 14 incidents at private grade crossings and 46 at public crossings. In 33 of the 60 cases, more than one-half, limited sight distance was listed as the primary cause. In 28% of the private crossing incidents, there was limited sight distance, while 63% of the public crossing incidents cited this as a factor. Stop signs were present at 3 of the 4 private grade-crossing incidents with limited site distance and at 15 of the 29 public grade-crossing incidents with limited sight distance. Since one of the rec- ommendations of an earlier NTSB report was that stop signs be installed at all passive grade cross- ings, it is interesting to note that stop signs were present at one-half of the private crossings and about one-third of the public crossings. Regarding the primary cause of these accidents at cross- ings with stop signs, the driver ran the stop sign in 3 of the 7 incidents at passive grade crossings (about half) and 13 of the 15 public grade crossings with stop signs (almost 90%). Clearly the rec- ommendation to place stop signs at all grade crossings with limited sight distances will have lim- ited effect until the compliance rate with the stop signs is improved. Regarding injuries and fatalities, there were two cases where a person on the train was injured. Two crew members and 12 passengers were injured in these accidents, respectively. Fatalities or Database Analysis 89

injuries occurred in 47 of the 60 incidents. Because the 60 cases were not chosen at random, no significance can be made regarding this finding. For the 14 private grade-crossing incidents in the database, there were 7 incidents in which there was an injury or fatality and in only 1 of those incidents was there limited visibility at the crossing. At the 46 public grade crossings in the data- base, there were 30 incidents where limited visibility was present and of those, 28 incidents resulted in one or more injuries or fatalities. The situation is similar with the 16 cases in which limited visibility was not an issue. In 12 of the 16, there was one or more injury or fatality. One can conclude from these data that while more than one-half of the passive grade-crossing acci- dents occurred at crossings with limited visibility, whether or not there is an injury or fatality is not a function of limited visibility. Rather, a large fraction of grade-crossing accidents result in one or more injury or fatality to the occupants of motor vehicles; this is just associated with the seriousness of the incident when it does occur. The NTSB study clearly demonstrates the importance of site visits and driver and witness interviews if probable and contributing causes are to be identified. Witness statements appear to have the greatest value. Otherwise, it would be difficult to obtain data, such as whether the driver was talking to an occupant and never looked, or the driver was talking on a cell phone. It is clear that the NTSB conclusions were based on these interviews. Even in the seven cases where it could not be determined if the findings were based on witness statements, it is likely that the witness statements either influenced the determination of probable cause or validated the conclusion by providing supporting information. Two cases reported that an event data recorder was read to verify that the train engineer sounded the horn on approach of the crossing. Such collaborating evidence is also useful when attempting to identify probable and contributing causes. In the FRA train accident database, there are no fields specifically designated for probable and contributing causes. There are narrative fields that could be used for statements provided by the railroad crew, motor vehicle occupants, and/or witnesses. The railroad train crew could be a valu- able source of information. They could document contact information for witnesses and fill out a form describing the conditions at the crossing at the time of the accident. For example, they could note if the view of the train was blocked by overgrown vegetation, a recently constructed building, or something temporary (e.g., a pile of rocks or train equipment). This would enable the railroad officials to more accurately complete the FRA form. The railroad official could also obtain state- ments from the train crew and witnesses, placing their statements in the narrative fields of the cur- rent reporting form. The person filling out the FRA accident report would also have access to the readings from the train’s event recorder at the time of the accident. A query of the FRA railroad crossing accident database for the period beginning in November 1995 and ending in September 1996 revealed that the narrative fields were left blank in all cases—nearly 4,000 records. The other alternative is to add fields where the person filling out the grade-crossing accident report can specif- ically list causes. This means that if the same person fills out both forms, he or she is already trained in determining accident causation. 4.8.6 Summary The following potential measures apply to capturing root causes routinely in grade-crossing accidents: 1. Require the information that was captured by the NTSB investigator to be incorporated in the FRA grade-crossing accident form. While the railroad can be left to determine how this recommendation is to be implemented, it can probably be accomplished by either a site visit or by the train crew documenting the conditions at the accident scene and obtaining contact information for witnesses, including law enforcement personnel. It would be possible for the 90 Hazardous Materials Transportation Incident Data for Root Cause Analysis

railroad official filling out the grade-crossing incident form to follow up with these individ- uals as the form is being completed. 2. Require the narrative sections of the FRA grade-crossing accident form to be filled out with witness statements. Police officials, if present at the scene, are trained to provide accident details and their findings should be included in the narrative fields. 3. Readings from the train’s event recorder at the time of the accident could be obtained and stored to verify some of the statements made by the train engineer regarding speed and whether or not the horn was sounded at the proper time. 4. The rail equipment incident database has fields entitled primary cause and contributing cause. Such fields could also be included in the grade-crossing accident reporting form. 5. Evaluate the need to provide additional guidance on the definition of restricted view of the railroad tracks. While both NTSB and FRA emphasize the importance of seeing the train as the motor vehicle approaches the crossing, it does not appear that the railroads were using this definition when filling out the FRA grade-crossing accident report. Railroads also could place maintenance cars, supplies, and equipment in a location where the view of oncoming trains is not obstructed at passive grade crossings. Although this occurred in only 1 of the 60 cases, it is a condition that can be addressed easily. 6. If the GPS location of all accidents was recorded in the database, by using these data along with month, day, and year, it could be possible to display the frequency of grade-crossing acci- dents in a region and also couple records in multiple databases, thereby expanding the amount of information available regarding an accident with no increase in the number of forms that have to be filed. 4.9 The Hazmat Serious Truck Crash Project Database 4.9.1 Introduction The Hazardous Materials Serious Crash Analysis: Phase 2 (Battelle 2005), a project conducted from 2002 to 2005 for FMCSA, demonstrated two methods for improving the usefulness of a database for identifying the root causes of hazmat crashes. The project achieved this result by improving data quality (including comprehensiveness) and augmenting the database with addi- tional fields. The project demonstrates how, by adding specific data fields, checking data from the original source, and supplementing data with telephone calls, the user is able to develop insights into root cause analysis that could not be obtained without the application of these techniques. This project had the following three basic objectives: • Enhance the current methodology for identifying and characterizing serious hazmat truck crashes in the United States. • Improve the capability to analyze causes and effects of selected serious hazmat crashes. • Support the implementation of hazmat truck transportation risk reduction strategies for pack- agings, vehicles, and drivers. The Hazmat Serious Truck Crash Project used the MCMIS Crash file for serious crashes occurring in 2002, extracted the crashes that involved hazardous materials and, for a sample of 1,000 hazmat crashes, supplemented the data in MCMIS with information from other sources. These sources included HMIRS, PARs filed by police from individual states, and direct cor- respondence with the involved carriers. Database Analysis 91

Sample crash information was input and stored in the Hazmat Accident Database, a specially designed database that enabled the aforementioned information sources concerning a particular crash to be assembled into as complete a record as possible. This included characteristics describ- ing the crash event, as well as the accuracy of the information itself. Extensive database protocols and quality control checks were employed to accomplish this objective. Once database develop- ment was complete, analyses were performed on the database for the purpose of providing infor- mation that might support the development of more rigorous hazmat truck safety policy. 4.9.2 Adding Explanatory Variables to the Hazmat Accident Database For the Hazmat Accident Database, the Hazmat Serious Truck Crash Project added a num- ber of fields designed to capture the actions of the vehicle(s) before the crash. In addition, fields were added to provide more detail on the type and quantity of the hazardous materials, hazmat packaging, infrastructure, and such driver characteristics as age and experience. Explanatory variables are crash characteristics that help explain cause and effect. Table 4-28 shows the five types of explanatory variables. The crash analysis process involved associating explanatory variables with impacts to determine how vehicle, driver, packaging, infrastructure, and situational characteristics influence crash occurrences in general, as well as those that result in spills. Appendix E (available on the TRB website at www.TRB.org by searching for HMCRP Report 1) describes selected analytical results from the project. 4.9.3 Crash Records Selection Of the approximately 2,000 hazmat crashes in the MCMIS database in 2002 (out of a total 105,000 records), the Hazmat Serious Truck Crash Project selected about one-half of the hazmat crashes for a more in-depth analysis. These records were primarily selected on a random basis, with the exception that less common accidents involving hazmat Classes 1, 4, 5, 6, and 7 were all selected to obtain a large enough population. Before performing data analysis, weighting factors were applied to compensate for the non-random aspects of the selection. Table 4-29 shows the hazmat crash classes selected for more detailed analysis. 4.9.4 Populating Records and Improving Data Quality After the records were imported into the Hazmat Accident Database, HMIRS data were used to both fill in data for fields not included in MCMIS and for quality checking the existing data. Because the HMIRS fields are more fully populated, any fields in the database that were com- mon to HMIRS and MCMIS were overwritten by the HMIRS information. The remaining HMIRS information was also incorporated into the database. 92 Hazardous Materials Transportation Incident Data for Root Cause Analysis Vehicle Driver Packaging Infrastructure Situational Configuration Age Package Type Road Surface Pre-Crash Condition Cargo Body Experience Quantity Shipped Road Condition Dangerous Event GVW Condition Quantity Lost Road Type Vehicle Speed Age (Cargo Tank) Trafficway Impact Location Rollover Protection Access Control Primary Reason Inspection History Speed Limit Accident Type Design Specification No. of Lanes Weather Condition Table 4-28. Explanatory variables used in the Hazmat Accident Database.

A similar process was used to input PAR data. As the information was being filled in from the PAR, the data entry form showed the default values for any parameters that were previously entered based on information supplied by MCMIS and HMIRS. Any inconsistencies were changed to reflect the information contained in the PAR. Frequently, the changes were not inconsistencies, but expansions of the data. For example, many PARs list the actual GVW of the vehicle and, in those cases, that number was input in place of a broad weight category. The final step in populating the Hazmat Accident Database involved entering information that was obtained through direct telephone conversations with the involved carrier. These calls verified the accuracy of the entered information and provided specific information only the carrier could supply. This included such information as the amount of material being shipped; whether there was a spill and, if so, how much; the manufacturer and specification number of the packaging; and the year the packaging was fabricated. Valuable information on packaging characteristics was obtained from carriers who provided the DOT specification number for the tank, the year it was manufactured, the manufacturer, type of rollover protection on the cargo tank, and the inspection history. Many carriers could estimate the amount of material being shipped and if any was spilled. The type of damage to the cargo tank could sometimes be recalled, usually only if there was a spill. Most carriers also were willing to provide information on the driver’s experience. 4.9.5 Quality Control Checks Several quality control checks were built into the data collection process. Accuracy checks were performed at three critical junctures: (1) after the data from the PAR were entered for the crash, (2) after the carrier calls were completed, and (3) whenever a reviewer changed a pre-existing database entry. Special efforts also were made to identify and reconcile blank fields. In addition, error-trapping queries were run to identify reporting inconsistencies (e.g., Interstate highways that were not flagged as limited/controlled access). Finally, summary reports were generated of each recorded crash to use in reviewing the entered information or to use as a reference during carrier correspondence. 4.9.6 Database Enhancements and Limitations The fields in the Hazmat Accident Database reflect a list of parameters that are considered pertinent for safety analysis. While every effort was made to obtain relevant information, it was not expected that it would be possible to populate all of the fields. Nevertheless, significant Database Analysis 93 Analyzed Crashes Estimated 2002 Totals Hazmat Group Description Crashes Spills Crashes Spills 1.1 – 1.6 Explosives 19 2 21 2 2.1 Flammable gases 148 14 256 21 2.2 Non-flammable gases 60 8 102 12 2.3 Gaseous poisons 11 1 18 2 3.0 Flammable liquids 544 125 914 182 4.1 – 4.3 Flammable and reactive solids 7 2 8 2 5.1 – 5.2 Oxidizing materials 31 9 36 10 6.1 – 6.2 Poisonous and infectious substances 14 2 16 2 7.0 Radioactive materials 4 2 4 2 8.0 Corrosive liquids 75 16 139 23 9.0 Miscellaneous hazardous materials 57 23 86 27 Unknown Hazmat group could not be determined 17 5 28 9 Table 4-29. Sampled crashes by hazmat group.

improvements were made in the breadth and accuracy of hazmat crash information from which safety assessments and root cause analyses can be performed. These improvements are evident by comparing initial MCMIS tables with the completed Haz- mat Accident Database. In addition to broadening the selection of eligible entries to many of the descriptive tables, new tables also were created that are not present in either MCMIS or HMIRS, such as Pre-Crash Events, Primary Reasons, and Impact Location. Moreover, data collected from PARs and from carrier correspondence for nearly 1,000 MCMIS crash records enabled many MCMIS data fields that were initially blank to be populated. Despite these improvements, some fields were largely blank. For example, no PAR captured information on evacuations. Only one state, Kentucky, captured information on road closures. The vehicle speed was captured in roughly 50% of the PARs, and the trailer dimensions, length, and width could be obtained in only one-quarter of the cases. The other fields were filled out for more than 80% of the selected crashes and in some states that figure was 100%. Some states, such as California, have extensive PARs that provide information on all of the key parameters as well as other parameters that might be of future interest. Roughly 60% of the states use a commer- cial vehicle supplement, designed to capture data required for the MCMIS Crash file. These sup- plements tend to have a uniform hazmat section that provides all of the information needed to fill out the five hazmat entries in MCMIS. Unfortunately, about 25% of the states that are known to have commercial vehicle supplements did not provide the supplemental form. Appendix E (available on the TRB website at www.TRB.org by searching for HMCRP Report 1) presents rep- resentative analyses that were conducted using the Hazmat Accident Database. 4.9.7 Summary The Hazardous Materials Serious Crash Analysis: Phase 2 (Battelle 2005) convincingly demon- strated that by adding explanatory fields to MCMIS, selecting a sample of crashes for more detailed investigation, matching the same crash in HMIRS with the one in MCMIS, using PARs to check data quality and complete added data fields, and telephoning carriers to collect data on such elements as hazmat type and quantity, root cause analyses could be made more accurately and thoroughly. Consequently, the project team suggests that the following lessons learned from the Hazmat Serious Truck Crash Project be applied to the data collection process: • Add selected explanatory fields to increase the type of data available for analysis. • Select an annual population of hazmat crashes for more detailed investigation. This popula- tion could be selected based on a number of criteria such as type of crash or type of hazardous material. For example, for a particular year (or years), all rollover crashes resulting in a spill could be selected or all crashes involving a spill of Class 3 hazardous materials could be selected. The number of years selected would relate to the number of accidents available for analysis. For rarer events and hazmat classes, a larger number of years would be chosen. • Match all applicable crashes to HMIRS. Use HMIRS data to supplement MCMIS data wher- ever possible. • Use the PARs to supplement and check crash data. Collect PARs from the states and use them to ensure data quality and to complete data in the added fields. • Telephone key contacts such as carriers to collect unique data. Carriers should be called to col- lect data on such characteristics as driver age, packaging type, and type and quantity of haz- ardous material shipped and/or spilled. 94 Hazardous Materials Transportation Incident Data for Root Cause Analysis

Next: Chapter 5 - Potential Measures for Improving the Identification of Root Causes for Hazardous Materials Crashes »
Hazardous Materials Transportation Incident Data for Root Cause Analysis Get This Book
×
MyNAP members save 10% online.
Login or Register to save!
Download Free PDF

TRB’s Hazardous Materials Cooperative Research Program (HMCRP) Report 1: Hazardous Materials Transportation Incident Data for Root Cause Analysis examines potential technical improvements to hazardous materials accident databases that are collected and managed by various agencies. The report explores gaps and redundancies in reporting requirements and attempts to estimate the extent of the under-reporting of serious incidents.

Appendixes A through E to HMCRP 1 are available online.

Appendix A: Questionnaires

Appendix B: Questionnaire Results for Carriers and Database Administrators

Appendix C: Brief Summary of the 2005 MCMIS Crash Records

Appendix D: The Percent of Missing Data for Variables from TIFA/FARS, 1999–2004

Appendix E: Selected Analyses Performed with the Hazmat Accident Database

  1. ×

    Welcome to OpenBook!

    You're looking at OpenBook, NAP.edu's online reading room since 1999. Based on feedback from you, our users, we've made some improvements that make it easier than ever to read thousands of publications on our website.

    Do you want to take a quick tour of the OpenBook's features?

    No Thanks Take a Tour »
  2. ×

    Show this book's table of contents, where you can jump to any chapter by name.

    « Back Next »
  3. ×

    ...or use these buttons to go back to the previous chapter or skip to the next one.

    « Back Next »
  4. ×

    Jump up to the previous page or down to the next one. Also, you can type in a page number and press Enter to go directly to that page in the book.

    « Back Next »
  5. ×

    To search the entire text of this book, type in your search term here and press Enter.

    « Back Next »
  6. ×

    Share a link to this book page on your preferred social network or via email.

    « Back Next »
  7. ×

    View our suggested citation for this chapter.

    « Back Next »
  8. ×

    Ready to take your reading offline? Click here to buy this book in print or download it as a free PDF, if available.

    « Back Next »
Stay Connected!