03-24-PK

8/10/2019 03-24-PK

1/17

Risk Analysis, Vol. 23, No. 5, 2003

Accident Epidemiology and the U.S. Chemical Industry:Accident History and Worst-Case Data from RMP*Info

Paul R. Kleindorfer, 1 James C. Belke, 2 Michael R. Elliott, 3 Kiwan Lee, 4

Robert A. Lowe, 5 and Harold I. Feldman 6

This article reports on the data collected on one of the most ambitious government-sponsoredenvironmentaldataacquisitionprojectsofalltime,theRiskManagementPlan(RMP)datacol-lected under section 112(r) of the Clean Air Act Amendments of 1990. This RMP Rule 112(r)was triggered by the Bhopal accident in 1984 and led to the requirement that each qualifyingfacility develop and le with the U.S. Environmental Protection Agency a Risk ManagementPlan (RMP) as well as accident history data for the ve-year period preceding the ling of the RMP. These data were collected in 19992001 on more than 15,000 facilities in the UnitedStates that store or use listed toxic or ammable chemicals believed to be a hazard to the en-vironment or to human health of facility employees or off-site residents of host communities.The resulting database, RMP*Info, has become a key resource for regulators and researchersconcernedwith the frequency andseverity of accidents, andthe underlying facility-specic fac-tors that are statistically associated with accident and injury rates. This article analyzes whichfacilities actually led under the Rule and presents results on accident frequencies and sever-ities available from the RMP*Info database. This article also presents summaries of relatedresults from RMP*Info on Offsite Consequence Analysis (OCA), an analytical estimate of

the potential consequences of hypothetical worst-case and alternative accidental releases onthe public and environment around the facility. The OCA data have become a key input in theevaluation of site security assessment and mitigation policies for both government plannersas well as facility managers and their insurers. Following the survey of the RMP*Info data, wediscuss the rich set of policy decisions that may be informed by research based on these data.

KEY WORDS: Accident epidemiology; chemical accident history; Clean Air Act; risk management

1 Risk Management and Decision Processes Center, The WhartonSchool, University of Pennsylvania.

2 Ofce of Emergency Prevention, Preparedness, and Response

(OEPPR), United States Environmental Protection Agency.3 Department of Biostatistics and Epidemiology, Center for Clin-

ical Epidemiology and Biostatistics, University of PennsylvaniaSchool of Medicine.

4 Risk Management and Decision Processes Center, The WhartonSchool, University of Pennsylvania.

5 Departments of Emergency Medicine and Public Health & Pre-ventive Medicine,OregonHealth& Science University; LeonardDavisInstitute of HealthEconomics, Universityof Pennsylvania.

6 Department of Biostatistics and Epidemiology, Center for Clin-ical Epidemiology and Biostatistics, University of PennsylvaniaSchool of Medicine.

1. INTRODUCTION

The tragedy at Bhopal in December 1984, fol-lowed by a subsequent release of aldicarb oxime froma facility in Institute, West Virginia, resulted in greatpublic concern in the United States about the poten-tial danger posed by major chemical accidents. Thispublic concern was translated into law in three differ-ent legislativeprograms. The rstlawintendedspeci-cally to address the problem of chemical accidents inthe United States was the Emergency Planning andCommunity Right-to-Know Act of 1986 (EPCRA;Public Law 99-499), which required state and localgovernments to plan for emergencies and required

865 0272-4332/03/1200-0865$22.00/1 C 2003 Society for Risk Analysis

8/10/2019 03-24-PK

2/17

866 Kleindorfer et al .

chemical facilities to report information about chemi-cal hazards to government and the public. The othertwo legislative programs addressing chemical acci-dents were enacted under sections 304 and 112(r), re-spectively, of the Clean Air Act Amendments of 1990

(Public Law101-549).The rst of these was the Occu-pationalSafety andHealth Administration s (OSHA)ProcessSafety Management (PSM) standard(29 CFRPart 1910), which required facilities having speci edhazardous chemicals to implement accident preven-tion measures designed to protect workers. The mostrecent program, arising from section 112(r) of theClean Air Act Amendments, was the Environmen-tal Protection Agency s (EPA s) Risk ManagementProgram (40 CFR Part 68). The EPA rule RiskMan-agement Programs for Chemical Accidental ReleasePrevention (hereafter the Rule), sets forth a seriesof requirements aimed at preventing and minimizingthe consequences associated with accidental chemi-cal releases. The Rule applies to facilities (both pub-lic and private) that manufacture, process, use, store,or otherwise handle regulated substances at or abovespecied thresholdquantities (whichrange from 500 20,000 pounds). 7

The U.S. Environmental Protection Agency(EPA) estimated in its 1996 economic impact analysis justication study (USEPA/CEPPO, 1996) to theOf ce of Management and Budget (OMB) thatabout 66,000 facilities nationwide would be regulatedunder the Rule, including many facilities not coveredunder OSHA s PSM standard. With some exceptions,the Rule requires all regulated facilities to prepareand execute a Risk Management Program, whichcontains the following elements:

1. A Risk Management Plan (RMP), a reportcapturing certain details about the facility saccident prevention program, emergency re-sponse program, and hazard assessment alongwith administrative information about thefacility.

2. A hazard assessment to determine the con-

sequences of a speci ed worst-case scenarioand other accidental release scenarioson pub-lic and environmental receptors and providea summary of the facility s ve-year history of accidental releases.

7 This is not meant to imply that any facility that does not meetthreshold inventory requirements is completelyexemptfrom theaccidentalreleasepreventionprovisionsof 112(r) sincea GeneralDuty provision may still apply if the facility poses a potentialhazard to the public.

3. An accidental release prevention program de-signed to detect, prevent, and minimize acci-dental releases.

4. An emergency response program designed toprotect human health and the environment in

the event of an accidental release.BeginningJune 21, 1999, theRule (section68.42) alsorequires that regulatedfacilities submit their ve-yearhistory of accidental releases to the EPA as part of their Risk Management Plan and periodically updatetheir submission every ve years or when signi cantchangesoccur. Thedata reportedhere re ectthestateof the RMP*Info database as of June 29, 2001, cor-rected for a few known errors (as described in thenext section).

The purpose of this article is to describe the re-porting facilities, the accidents that have occurred

at these facilities in the rst ve years covered bythe Rule, and the potential off-site consequences inthe event that worst-case scenarios occurred at thesefacilities. The basic approach followed in this studyhas been the epidemiological methodology known asretrospective cohort study design. Epidemiology isthe study of predictors and causes of illness in hu-mans. Its use in studying industrial accidents has beenproposed in a number of quarters (e.g., Saari, 1986;Rosenthal, 1997). The motivating idea is to studythe demographic and organizational factors of thosefacilities whose accident histories are captured in

RMP*Info to determine whether any of these fac-tors have signi cant statistical associations with re-ported accident outcomes, positive or negative, just asone might use demographic or life-style data for hu-man populations to determine factors that might beassociated with the origin and spread of speci c ill-nesses. This article provides the descriptive statisticson RMP*Info and serves as a foundation for analyt-ical work (e.g., Elliott, Kleindorfer, and Lowe, 2003;Elliott, Wang, Lowe, andKleindorfer,2003)exploringthe structure of statistical relationships between acci-dent outcomes, facility characteristics, regulations inforce, and the demographics of the communities inwhich facilities are located.

As several commentators have already noted,RMP*Inforepresentsa signi cantstepinunderstand-ingthe scope of accidentsin thechemical industryandin promoting more effective accident prevention andmitigation. 8 New business models have emphasized

8 See, for example, Rosenthal (1997) and Mannan and O Connor(1999) for discussions of the promise of using large-scale com-parative data to determine robust predictors of accidents in thechemical industry.

8/10/2019 03-24-PK

3/17

Accident Epidemiology and the U.S. Chemical Industry 867

the importance of learning across facilities, basedon benchmarking and best practices. Using data inRMP*Info, together with other organizational and -nancial data on the facilities and companies involved,is taking this approach to another level. Indeed, look-

ing across the entire U.S. chemical industry, as wellas across speci c segments, technologies, and chemi-cals therein, clearly holds the potential for detectingand validating factors predictive of severity and fre-quency of accidents. These models can then provideinput for rational prioritization of risk managementand regulatory policy initiatives designed to preventfuture accidents.

The article proceeds as follows. In Section 2 wedescribe the nature of RMP*Info and the prelimi-nary data screening undertaken to assure data qual-ity for RMP*Info. Section 3 describes the nature of the facilities that led, with the top 20 by chemicaluse and by the North American Industry Classi ca-tion System (NAICS) Code listed explicitly. Section 4presentsresultsonaccidentfrequency andseverity, in-cluding details by chemical andNAICS Code.Section5 presents some simple analyses on timing and loca-tion of accidents as well as on the size of plants (mea-sured by the number of full-time equivalent (FTE)employees) involved in these accidents. Section 6 fo-cuses on the Offsite Consequence Analysis (OCA)based on worst-case and alternative release scenariosin theRMP*Infodatabase. Conclusions areoffered inSection 7.

2. INTRODUCTION TO RMP*INFO ANDPRELIMINARY DATA SCREENING

This section describes the information collectedunder the Rule. We also discuss data quality issueshere as a necessary precursor to our analysis in therest of thearticle.As promulgated in theRule, thefol-lowing are the data elements required to be reportedin RMP*Info for each covered facility:

1. Executive Summary: This must cover the na-

ture of the facility and its policies for preven-tion and emergency response, as well as a ver-balsummaryof thefacility s ve-year accidenthistory.

2. Section 1: Facility identi cation informationand basic demographics on the facility, its par-ent company, and its covered processes, in-cluding a listing of regulated chemicals abovethreshold quantities at the facility and indica-tions of whether the source is covered by var-ious other regulatory processes (e.g., OSHA

Process Safety Management (PSM) Standard,Emergency Planning and Community Right-to-Know Act (EPCRA) Section 302, CleanAir Act (CAA) Title V).

3. Sections 2 and4: Description of worst-case re-

lease scenarios for regulated toxic (Section 2)and ammable (Section 4) substances abovethreshold quantities at the facility.

4. Sections 3 and5: Descriptionof alternative re-lease scenarios for regulated toxic (Section 3)and ammable (Section 5) substances abovethreshold quantities at the facility.

5. Section 6: Five-year accident history for thefacility, including a separate record for eachaccidentalrelease fromcoveredprocesses thatoccurredduring the ve-year reportingperiodfor the facility.

6. Sections 7 and8: Prevention Program descrip-tions for Program 3 processes (Section 7) andProgram 2 processes(Section8), includingde-tails on hazard analysis methods, operatingprocedures, training, and other related infor-mation, together with a list of the major haz-ards identi ed for these processes.

7. Section 9: Details on the emergency responseplan at the facility, including indications of which of several federal and state regulationson emergency response apply to the facility.

As of June 29, 2001, 15,430 facilities had led withthe RMP*Info database. The vast majority (14,500of 15,430) of facilities covered in this analysis ledtheir RMP data by the initial deadline of June 21,1999, and some corrections were made on additionsof new lers or exemptions of existing lers untilJune 29, 2001. Certain other facilities producing orutilizing ammable fuels such as propane sought leg-islative and judicial relief from the RMP require-ments, and were granted a temporary judicial stayfrom RMP compliance and reporting requirements.Indeed, many of these facilities were eventually ex-cluded from the Rule under the Chemical Safety In-

formation, Site Security and Fuels Regulatory Relief Act (Public Law 106-40) passed in August 1999. InJanuary 2000, the judicial stay was lifted and an addi-tional 930 facilities, primarily propane-producing fa-cilities and fuel wholesalers, led risk managementplans under the Rule.

The result of all of these developments in the im-plementation of the Rule is that the time window rep-resented by RMP*Info is not uniform for all facilities.A facility, for example, that led its RMP on May10, 1999 could have interpreted the ve-year history

8/10/2019 03-24-PK

4/17


covered by the Rule to be May 11, 1994 through May10, 1999. Other facility owners interpreted more pre-cisely as given above, and anticipated ling updatedRMPs if their facility had an accident between thetime they led and June 20, 1999. In addition, as will

be discussed below, some facilities were initially ex-empt from ling, but eventually held to be covered bythe Rule. These facilities then led after the initial l-ing date of June 21, 1999, and their RMP reportedaccidents for the ve-year period preceding theirlater ling date. Notwithstanding changes, the vastmajority of the data represents accident histories forthe period mid 1994 to mid 1999.

While, as notedabove, theoriginalestimate of thecovered facilities expected to le under the Rule was66,000, we will see in the data reported below that thenumber of facilities actually ling was, in fact, 15,430(23.2% of the original estimate), with 1,205 of thesefacilities (7.8% of 15,430) reporting some 1,970 acci-dents over the ve-year period of interest. A furtherrestrictionininformationavailabletothepublicisthatworst-case data, in the form of the required OffsiteConsequenceAnalysis(OCA)noted under (1)above,have not been made available to other than coveredpersons in order to reduce the possibility that thesedata might be used by terrorists to target speci c fa-cilities. Members of the public can get limited accessto OCA data for individual facilities by visiting anyof 50 federal reading rooms. The database itself hasbeen named RMP*Info and has been available to thepublic after August 1999. 9

Concerning accuracy and consistency, a rst stepin any epidemiologic study is the screening of data,and we therefore note some of the steps taken withrespect to this critical issue in data quality assurance.In this regard, it is important to note that nearlyall submissions under the Rule were electronic, with97% of the nal RMP submissions made using stan-dardized software, entered on diskette, and mailedto the EPA. While manual submissions using a stan-dard paper form were allowed, these accounted foronly 3% of the total. 10 Electronic submission is criti-

cal to data quality since the data submission system,called RMP*Submit, used a standard data entry tem-plate and had a number of self-correcting and error-

9 Since September 11, 2001, the RMP*Info data remain avail-able to the public only on a more limited base through 50EPA reading rooms set up for this purpose. For details onpublic access, refer to 40 CFR Chapter 4, and to the websitehttp://www.epa.gov/ceppo/.

10 Personal communication of 01/24/00 from KarenSchneider, whoguided much of CEPPO s effort in data input and the qualityassurance program surrounding RMP*Submit.

checking mechanisms built into it to assure that thedata submitted were in a standard format and metother consistency checks (such as range checks). 11Notwithstanding the signi cant effort undertaken byEPA to assure the overall quality of the data, the re-

searchteam also undertook itsowndata-cleaning andscreening checks. Inparticular, thefollowingtwo stepswere undertaken by the research team:

1. Extensive interviews with plant-level and cor-porate managers responsible for submittingthe RMP data were undertaken during theperiod November 1998 through June 1999 todetermine whether there were ambiguities inthe minds of facility managers as to what datawere required. The primary dif culties werewith understanding the requirements for theOCA, both worst-case and alternative scenar-ios. The managers at both large and small fa-cilities generally exhibiteda clearunderstand-ing of the requirements of the Rule and theyshoweda positiveandconstructiveattitude to-ward the RMP process, where smaller com-panies typically relied on trade associationsand consultants to assist them in this pro-cess. The effort expended on complying withthe Rule was considerable. Indeed, data onsome 10 companies collected as part of thispre-screening process indicated that, includ-ing internal and external consultants time,

person-hours dedicated to putting thedata to-gether forRMP*Info rangedfrom50 hoursforsome small companies to nearly 3,000 hoursfor some large facilities.

2. Standard approaches for quality assurance of data, commonly employed in epidemiologicstudies, were employed to look for data er-rors. For all variables included in this re-port, frequency distributionswerereviewedtolook for unusual or unexpected values ( out-liers). Where appropriate, cross-tabulationswere performed to look for internal inconsis-tencies in the data. Outliers were discussedwith EPA staff, who reviewed these cases todetermine their validity. 12

11 Itis notour purpose to reviewor comment onthe extensive effortundertaken to assure data quality in the RMP process and thedetails of the software developed to assure data quality underthe RMP*Submit system. The details of this can be found byconsulting the extensive documentation provided by CEPPO attheir website http://www.epa.gov/ceppo/.

12 An example of thisquality assuranceprocessmaybe informative.A frequency distribution of the number of full-time equivalent

8/10/2019 03-24-PK

5/17


Because the number of reported deaths is such animportant data element, further checking was doneof each accident in which nonemployee deaths oc-curred. This led to the nal result (reported inTables VII and VIII below) that while there were 32

deaths among employees at reporting facilities, all of the originally reported 45 public responders deathsand 11 public deaths were data errors. 13 We haveincorporated these corrections on deaths to publicresponders and other nonemployees into our anal-ysis. However, there may be corrections and revi-sions to RMP*Info at any time via the submission of a corrected RMP by any facility; other, less obviouschanges to thedatabaseafter June 29,2001,will not bereected in this report. In particular, in interpretingresults from RMP*Info, it is critical to know the dateof thelast update incorporated in theanalysis andanynotable revisions, such as those above, undertaken tothe data.

3. PLANT DEMOGRAPHICS FOR FACILITIESREPORTING IN RMP*INFO

This sectionreports thebasic demographicsof thefacilities that led under RMP*Info. There are 15,430

employees(FTEs) reported at each facility revealeda range from0 to 48,000 FTEs. Eight hundred eighty-eight plants reported 0FTEs and 14 plants reported over 15,000 FTEs. The authors of this report queried EPA staff about these outliers. EPA staff

noted that all 14 of the facilities with over 15,000 FTEs weremilitary bases and con rmed that these values were plausible.EPA staff hypothesized that the facilities with 0 FTEs might berelated to speci c industries. That led the authors to determinethe NAICS codes of the facilities reporting 0 FTEs. The mostcommon processes were Water Supply and Irrigation Systems(246facilities), FarmSuppliesWholesalers (229),and FarmProd-uct Warehousing and Storage Facilities (186). EPA investigatedwhether it is plausible for such facilities to report 0 FTEs. EPAstaff responded, in part, to this question as follows: Coops areusually large organizations, frequently covering several states,but certainly serving many communities with individual outlets.They reported having zero FTEs because they are reporting on astorage facility that is unmanned except for certain seasons. Ac-cording to the way FTEs are calculated, if they have one person

there for ve months, they have less than 0.5 FTE and reportzero employees. (Breeda Reilly, CEPPO, personal communi-cation, December 14, 1999.) Further discussion with EPA staff addressed other categories of processes associated with 0 FTEs,until theresearch teamand EPA staff were satis edthatthe datawere accurate.

13 Breeda Reilly at EPA/CEPPO con rmed such errors. Four facil-ities reported total 45 public responders fatalities but they werereporting errors. There was no public responders death. In ad-dition, twofacilities reported a total of 11 on-site publicfatalitiesand two facilities reported total 68 on-site public injuries. Butthey turned out to be errors. In fact, there were zero incidentsfor on-site public fatalities and injuries.

facilities in RMP*Info and there are 1,970 reportedaccidents in RMP*Info, with 1,205 facilities reportingat least one accident. However, the sample size forvarious analyses will not remain constant at 15,430and 1,970, since some sites have multiple processes

and some processes use multiple listed chemicals.Tables I III list various characteristics of lersunder the Rule. Table I lists the 20 most commonlyreported chemicals, along with the number of plantsusing each chemical and the number of FTE employ-eesat these facilities. Also listedarethetotal numbersof facilities reporting use of at least one listed toxicor one ammable chemical. The average facility em-ployees reported158FTE,rangingfromfacilitieswithless than 0.5 FTE (recorded as 0 FTE in RMP*Info)to 48,000 FTE. Half of the facilities have 11 FTEs orfewer. Of the top 20 chemicals in terms of reportingfacilities, note that 11 are toxics and 9 are ammables.

Table II lists the 20 most commonly reportedindustrial sectors, along with the number of plantsreporting each process and the number of FTE em-ployees at these facilities. The industrial process isspecied by the NAICS code of the facility report-ing. The most common sector reported is farm supplywholesalers (4,357 or 29% of facilities), followed bywater treatment and irrigation systems (2,000 or 13%of facilities) and sewage treatment (1,421 or 9% of facilities).

Table III lists the numbers and percentages of reporting facilities, which indicated that they werecovered under various state and federal regula-tory programs covering process safety, noti cationrequirements, and emergency response regulations.Table III also lists the maximum Prevention ProgramLevel of any process at reporting facilities (this wascomputedby considering all processesat each report-ingfacility andtaking themaximum of thePreventionProgram Levels across all processes at a given facil-ity).14 We note that 7,209 (or 47%) of the reportingfacilities had at least one process at level 3, thereforerequiringa full Process Hazards Analysis to be under-taken and reported in the facility s RMP.

Asnotedin theintroduction, there isa substantialdifference between the number of facilities that were

14 EPA has de ned three different Prevention Program Levels toreect the potential for public impacts and the level of effortneeded to prevent accidents. Only minimal requirements areimposed on Program Level 1 processes, while Program Level 3processes are subject to much higher compliance requirements;Program Level 2 processes face intermediate requirements. Pro-gram Level 3 processesare those processesthatare eithersubjectto OSHA s PSM standard or belong to nine speci c SIC codesplaced in Program Level 3 by the EPA.

8/10/2019 03-24-PK

6/17

8/10/2019 03-24-PK

7/17


Table II. Twenty Most CommonlyReported NAICS Codes and

Characteristics of the Facilities ReportingThem 16

Filers with Avg. FTEs Std. Dev. of NAICS the Speci ed of Filing FTEs of Code NAICS Description NAICS Code Facilities Filing Facilities

42291 Farm supplies wholesalers 4357 7 11

22131 Water supply and irrigation systems 2000 206 233722132 Sewage treatment facilities 1421 216 215349312 Refrigerated warehousing and storage

facilities576 196 303

211112 Natural gas liquid extraction 482 14 2342269 Other chemical and allied products

wholesalers371 26 38

49313 Farm product warehousing and storagefacilities

342 5 17

11511 Support activities for crop production 305 7 7325211 Plastics material and resin manufacturing 255 263 519325199 All other basic organic chemical

manufacturing252 248 500

454312 Liqueed petroleum gas (bottled gas)dealers

242 16 94

311615 Poultry processing 226 805 510115112 S oil preparation, planting, and cultivating 194 10 10325188 All other basic inorganic chemical

manufacturing193 243 577

32411 Petroleum re neries 168 370 396221112 Fossil fuel electric power generation 140 86 11332512 Industrial gas manufacturing 135 58 16349311 General warehousing and storage

facilities131 600 4273

42271 Petroleum bulk stations and terminals 128 17 86311612 Meat processed from carcasses 124 424 411

Tables VII and VIII report the number of injuries and deaths for employees/contractors and non-employees, respectively. There were a total of 1,987injuriesand32 deaths among workers/employees, andthere were 167 injuries and 0 deaths among non-employees, including public responders. There were215 total hospitalizations and 6,057 individuals givenother medical treatments.

Table IX notes the damage to property andthe nonmedical off-site consequences resulting fromaccidents during the reporting period. Property dam-ages alone were inexcessof$1billion,and they donotinclude business interruptioncosts, including losses inshareholder value and lost business associated with

appear more than once in this table if more than one of the top20 chemicals are present at the facility. For the same reason, thenumber of facilities indicating the use of at least one toxic orammable material will exceed the total number of lers sincesome facilities have both toxic and ammables on site.

16 In Table II, if a facility has multiple processes with the sameNAICS code, it is reported only once. However, the same facilitymay appear more than once if it supports processes in more thanone NAICS code.

accidents. 17 In addition, large numbers of community

residents were affected by accidents (over 200,000 in-volved in evacuations and shelter-in-place incidents).The ecological consequences of the accidents are alsoreported in Table IX. Four hundred and eight ac-cidents (21%) resulted in on-site property damage;51 (2.6%) resulted in off-site property damage; 174(8.8%) resulted in evacuations; and 97 (5.0%) re-sulted in individuals being sheltered in place.

5. BASIC DESCRIPTIVE ANALYSISOF ACCIDENT HISTORY

Analytic studies are concerned with establishingstatistical associations between predictor variablessuch as facility characteristics and outcome variablessuchas frequency andseverityofaccidentsof facilitieshaving various characteristics. We willonlypursue thesimplest of such studies here, in the spirit of merely

17 These latter costs are likely to be larger, and perhaps muchlarger, than losses due to property damage. For a study of thefull shareholder costs of environmental accidents, see Klassenand McLaughlin (1996).

8/10/2019 03-24-PK

8/17


Table III. Reporting Facilities Coveredby Various Regulatory Programs

Number of Facilities Percent of Total FacilitiesCovered (from a Reporting under theTotal of 15,219 Rule Covered by Each

Name of Regulatory Program Reporting) Speci c Program

Process safety and hazardspermitting programsOSHA-PSM 7,600 49CAA-Title V 2,267 15EPCRA-302 12,689 82

Emergency response programsOSHA 1910.38 12,98 84OSHA 1910.12 9,190 60RCRA (40 CFR 264, 265, 279.52) 3,176 21OPA 90 (40 CFR 112, 33 CFR 1,424 9

154, 49 CFR 194, 30CFR 254)

State EPCRA rules/law 11,215 73Prevention program level

Level 1 647 4

Level 2 7,574 49Level 3 7,209 47

describing the basic characteristics of RMP*Info inthis article. 18

Table X displays the incidence of reportedaccidents over the initial reporting period forRMP*Info. 19 Several artifacts of the data collectionprocess confuse the results shownin Table X. The rstartifact is that the lower numbers for 1994 and 2001are because these years were only partly within theve-year reporting window for most companies. The

second artifact is that RMP*Info lacks data on theageof the facilities reporting to it. We know that the re-porting facilities existed as of their initial reportingdate but we do not know for what part of the pre-ceding ve years they existed or conducted processescovered bytheRule. Without this information, it isnotpossible to compute therate of accidentsperplantperyear todeduce anythingabout changes over time inthe accident rate. 20 The thirdartifact is that plantsthat

18 For a more detailed analysis of RMP*Info data relating facilitycharacteristics and regulations to accident frequency and sever-

ity, see Elliott, Kleindorfer, and Lowe (2003).19 Three accidents are omitted from this table because they werereported to occur in 1992 (2 cases) or 1993 (1 case). It is unclearif these represent data entry errors in the submissions, with thewrong date reported, or unnecessary reporting of accidents thatoccurred prior to requirements of the Rule. One accident wasreported to occur in 1966 (1 case) but it was a data entry er-ror. Breeda Reilly at EPA/CEPPO con rmed that the accidenthappened in 1996. We included the accident in our analysis.

20 Of course, this might be done for particular sectors or technolo-gies if plant ages for these sectors or technologies can all bereliably determined.

Table IV. Frequency of Accidents at Individual Facilities

Number of Facilitiesin RMP*Info with

the Indicated NumberNumber of Accidents of Accidents in the Total Accidentsat Facility Reporting Period Represented

1 850 8502 197 3943 69 207

4 31 1245 25 1256 12 727 7 498 4 329 1 9

10 3 3011 2 2213 1 1314 2 2815 1 15

Total 1,205 1,970

ceased operation before June 21, 1999 (or that ceasedactivities covered by the Rule) are not included inthe RMP*Info database and any accidents occurringat these plants are not recorded. If plants with highaccident rates were disproportionately likely to ceaseoperation early in the reporting period, there couldactually have been more accidents in the early years,in contrast to the apparent ndings shown in Table X.Given these uncertainties, we cannot state whether

8/10/2019 03-24-PK

9/17


Table V. Accidents Reported in RMP*Info by Chemical Involvedin the Accident for the Entire Period 1994 2001 (25 Chemicals

Most Frequently Involved)

Number of Chemical Name Chemical ID Accidents

Ammonia (anhydrous) 56 696Chlorine 62 534Flammable mixture 155 100Hydrogen uoride/hydro uoric 55 98

acid (conc. 50% or greater)[hydro uoric acid]

Chlorine dioxide 71 59[chlorine oxide (ClO 2)]

Propane 98 51Sulfur dioxide (anhydrous) 49 46Ammonia (conc. 20% or greater) 57 46Hydrogen chloride (anhydrous) 54 33

[hydrochloric acid]Hydrogen 149 31

Methane 93 27Formaldehyde (solution) 1 21Hydrogen sul de 63 21Butane 118 21Ethylene oxide [oxirane] 9 19Pentane 125 16Titanium tetrachloride 51 15

[titanium chloride (TiCl4) (T-4)-]Ethylene [ethene] 95 14Isobutane [propane, 2-methyl] 107 14Ethane 94 13Trichlorosilane [silane, trichloro-] 153 13Nitric acid (conc. 80% or greater) 58 12Oleum (fuming sulfuric acid) 69 12

[sulfuric acid, mixture with

sulfur trioxide]Toluene diisocyanate 77 11

(unspeci ed isomer)[Benzene, 1,3-diisocyanatomethyl-]

Vinyl chloride [ethene, chloro-] 101 11

the incidence rate of accidents has increased or de-creased over the last ve years.

Table XI reports the day of the week on whichaccidents in RMP*Info took place. A small peak inaccident rates is noticeable in mid-week. Again, one

should not infer from this anything about safe week-end operations since we do not know how manyof the facilities in RMP*Info operated as intensivelyon weekends as they did during weekdays. Similarly,we do not know whether the lower number of acci-dents on Mondays and Fridays is a result of shorterperiods of operation on these days, different workattitudes on these days, or other factors. Additionaldata would be required in order to study this issue.A number of other factors should also be consid-

Table VI. Accidents Reported in RMP*Info by NAICS Code of the Process Involved in the Accident for the Entire Period19942001 (Top 25 Processes Most Frequently Involved)

NAICS Number of NAICS Description Code Accidents

Petroleum re neries 32411 182Water supply and irrigation systems 22131 118Sewage treatment facilities 22132 112All other basic inorganic chemical

manufacturing325188 93

Farm supplies wholesalers 42291 93Other chemical and allied products

wholesalers42269 89

All other basic organic chemicalmanufacturing

325199 81

Alkalies and chlorine manufacturing 325181 79Poultry processing 311615 71Nitrogenous fertilizer manufacturing 325311 70Pulp mills 32211 55

Refrigerated warehousing andstorage facilities 49312 55

Petrochemical manufacturing 32511 52Animal (except poultry) slaughtering 311611 47Plastics material and resin


Natural gas liquid extraction 211112 35Frozen fruit, juice, and vegetable


Paper (except newsprint) mills 322121 30Meat processed from carcasses 311612 29Industrial gas manufacturing 32512 25Other basic organic chemical


Pesticide and other agricultural

chemical manufacturing

32532 21

Other basic inorganic chemicalmanufacturing

32518 20

Ice cream and frozen dessertmanufacturing

31152 19

Frozen food manufacturing 31141 18

ered in analyzing the temporal pattern of accidents,including seasonal manufacturing facilities, contin-uous versus batch operations, and speci c processcharacteristics.

Next, we report results related to thesize of plant,

as measured by FTEs at the plant, and accident ratesduring the reporting period. Several caveats must bekept in mind in reviewing these data. First, these datado not account for many possible confounders withplant size. For example, we do not control for the in-herent hazards in the processes in question and thiscould be a signi cant confounding in uence on thestatistical association of plant size and accident fre-quency andseverity. Generally, a much more detailedanalysis controlling for suchfactors as process hazard,

8/10/2019 03-24-PK

10/17


Table VII. On-Site Injuries and Deaths Resulting from AccidentsDuring Reporting Period

Mean or Std. Number of Total Dev. Min Max Observations

On-site injuries to workers/contractorsTotal on-site injuries 1,987 1,969Injuries per accident 1.0091 2.828 0 67 1,969Injuries per FTE 0.0207 0.0783 0 1 1,951

per accident

On-site deaths to workers/contractorsTotal on-site deaths 32 1,968Deaths per accident 0.0163 0.218 0 6 1,968Deaths per FTE 0.0003 0.0070 0 0.25 1,950

per accident

OSHA PSM implementation, and so forth, would be

required in order to understand the nature of the as-sociation of plant size with accident frequency andseverity.

With these cautions in mind, Table XII shows theassociation of increasing plant size with a higher fre-quency of accidents. We separated the data into thosefacilities reporting less than 0.5 FTEs, between 0.5and 10 FTEs, and more than 10 FTEs. As explainedearlier (see footnote 13), the FTE category of lessthan 0.5 represents mostly seasonal or part-time farmoperations that have less than 0.5 FTEs and, there-fore, report 0 FTE. Plants with more employees weresignicantly more likely to have accidents ( p < 0.001,chi-square for trend). 21

6. OFFSITE CONSEQUENCE ANALYSIS:INFORMATION AND ANALYSIS

Perhaps the most interesting information in theRMP*Infodatabaseis theOffsite ConsequenceAnal-ysis (OCA) information. 22 OCA information consistsof data related to worst-case and alternative releasescenarios. These scenarios represent hypothetical es-

21 Note that the FTE number includes all reported employees atthe facility, regardless of occupational class. Hence facilities witha large number of employees whose work does not involve reg-ulated substances (e.g., military bases) will have a lower per-employee risk from chemical exposures than facilities wherenearly all employees would be exposed to the consequences of accident releases (e.g., petroleum re neries).

22 Because of security concerns, the OCA data are currentlyonly available to members of the general public by appear-ing in person at one of the 50 reading rooms set up for ac-cess to these data. In accordance with these restrictions, thestatistical analysis reported here was accomplished internal toEPA/CEPPO.

timates of the potential consequences of accidentalchemical releases occurring under speci ed atmo-spheric and topographic conditions. The OCA datareported in the RMP include the following:

1. Name, physical state, and percent weight (if amixture) of chemical involved in the release

2. Analytical modelused to perform the analysis(i.e., scienti c technique used to estimate thedistance to which a toxic vapor cloud, over-pressure blast wave,or radiant heat effectswilltravel)

3. Type of scenario (e.g., gas release, explosion,re, etc.)

4. Quantity released5. Release rate and duration6. Atmospheric conditions and topography7. Distance to toxic or ammable endpoint

8. Residential population living within the end-point distance.

9. Other public or environmental receptorswithin the endpoint distance (e.g., schools,hospitals, churches, state or national parks,etc.)

10. Mitigation measures accounted for in con-ducting the analysis

OCA information does not include any estimate of the probability of a scenario actually occurring. How-ever, OCA scenarios are considered to be unlikely.Worst-case scenarios in particular are considered tobe very unlikely. This is because they are based onthe assumption of a very large accidental release (anunlikely event under any conditions) occurring undera combination of atmospheric conditions (low windspeed and stable atmosphere) that occurs rarely anddoes not persist for very long. Further, the regulatoryrequirements for conducting the worst-case scenarioanalysis prohibit facilities from accounting for anyactive release mitigation features such as water del-uge systems and automatic shutoff valves that mightsignicantly reduce the effects of an actual release.Facilities may, however, account for passive mitiga-

tion features such as containment dikes and buildingenclosures.Note that each facility usingat least one regulated

toxic chemical was required to provide a toxic worst-case and alternative release scenario; similarly, eachfacility using at least one regulated ammable chem-ical was required to provide a ammable worst-caseand alternative release scenario. The unit of analysisin the remainder of Section 6 is the scenario, not thefacility.

8/10/2019 03-24-PK

11/17


Table VIII. Nonemployee Injuries andDeaths Resulting from Accidents During

Reporting Period


Nonemployee injuriesTotal injuries to public responders for

all accidents63 1,968

Injuries to public responders peraccident

0.032 .5537 0 21 1,968

Total on-site injuries to other membersof the public for all accidents

104 1,968

On-site injuries to other members of the public per accident

0.0528 1.390 0 59 1,968

Total hospitalizations for all accidents 215 1,968Hospitalizations per accident 0.109 1.931 0 80 1,968Total other medical treatment for all

accidents6,057 1,968

Other medical treatment/accident 3.078 104.51 0 4,624 1,968

Nonemployee deathsTotal public responder deaths 0 1,968Total on-site deaths by other members

of the public

0 1,968

Overall nonemployee deaths/accident 0 1,968

6.1. Worst-Case Scenarios

EPA de ned the worst-case scenario as the re-lease of the largest quantity of a regulated substancefrom a single vessel or process line failure that resultsin the greatest distance to an endpoint (for most facil-ities, this is the amount contained in the largest vesselor pipe in the process). 23 In broad terms, the distanceto the endpoint is the distance, based on a releaseof the speci ed quantity of material, that a toxic va-por cloud, heat from a re, or blast waves from anexplosion will travel before dissipating to the pointthat serious injuries from short-term exposures willno longer occur. For toxic worst-case scenarios, EPAspecied certain input parameters for conducting theanalysis, such as wind speed and atmospheric stabil-ity. For ammableworst-case scenarios, EPA speci edthat thescenario consistedof a vaporcloudexplosion.

EPA placed numerous speci cations on worst-case scenarios in order to simplify the analysis and toensure comparabilityamong facilities. However,EPAdid not specify that any particular analytical modelbe used to conduct the analysis. When comparing

23 Usually, each facility has a single worst-case scenario, but thereare approximately 2,000 facilities that must report more thanone worst-case scenario, for either of two reasons. First, facili-ties that have both toxic and ammable substances must reportone worst-case scenario for each class of substance. Second, theRule requires facilities to report more than one worst-case sce-nario when the facility has multiple processes that could affectsignicantly different off-site populations.

worst-case scenarios, this is a potentially confoundingvariable, since the same scenario analyzed using twodifferent analytical models can sometimes producesignicantly different results. However, most worst-case scenarios were conducted using EPA OCAmodeling. 24

6.1.1. Endpoint Distances

Ingeneral, toxicreleasescenariosresult ingreaterendpointdistancesthan ammable worst-casescenar-ios. This is mainly due to the fact that for ammablesubstances, EPA speci ed the endpoint distance to bethe distance from the source of a vapor cloud explo-sion to the point where the overpressure from the ex-plosion falls to 1 psi. For most regulated ammablesubstances, this distance tends to be signi cantlyshorter than the toxic endpoint distance resulting

24 EPA published several guidance documents and one computersoftware program to assist facilities in conducting OCA model-ing. Foremost among these is Risk Management Program Guid-ance for Offsite Consequence Analysis , which contains genericOCA lookup tables and modeling equations for all RMP-regulatedchemicals. EPA alsopublished several industry-speci cguidance documents that contain lookup tables for regulatedchemicals of particular concern to certain large industry sectorsregulated under the RMP Rule. Additionally, EPA and the Na-tional Oceanic and Atmospheric Administration together pro-duced a software program, called RMP*Comp, which conductsOCA modeling according to the same methodologies containedin the EPA guidance documents. OCAresults achieved usinganyof these sources are derived from the same set of models.

8/10/2019 03-24-PK

12/17


Table IX. Property Damage andNonmedical Off-Site Consequences

Resulting from Accidents DuringReporting Period


On-site property damage ($ millions)Total on-site damage $1,041 1,966Damage per accident $0.529 $6.641 $0 $219 1,966

Off-site property damage ($ millions)Total off-site damage $11.7 1,967Damage per accident $0.006 $0.108 $0 $3.8 1,967

Off-site consequencesTotal number of evacuations 174 1,968Total number of evacuees in all

accidents31,921 1,968

Number of evacuees per accident 16.22 140.97 0 3,000 1,968Total number of accidents involving

shelter in place97 1,968

Total number of individualsconned to shelter in place inall accidents

184,839 1,968

Number of individuals con ned toshelter in place per instance

93.9 1,913.4 0 55,000 1,968

Number of accidents with effectson the ecosystemFish or animal kills 19 1,970Minor defoliation 55 1,970Water contamination 25 1,970Soil contamination 29 1,970Any environmental damage 104 1,970

from the release of a similar quantity of the mostprevalent RMP toxic chemicals.

Figs. 1 and 2 are frequency histograms of end-point distance forRMP toxic and ammable chemicalprocess worst-case scenarios, respectively. Each barrepresents scenarios having endpoint distances in aparticular distance interval. Relatively few processesof either type result in extremely long endpoint dis-tances. However, while the shapes of the two distri-butions are similar, ammable scenarios are differen-tiated from toxics by their shorter endpoint distances.

Table X. Pattern of Accidents over the Five-Year Period

Number of Percent of Accidents in Total

Year the Year Accidents1994 141 7.21995 319 16.11996 393 201997 436 22.11998 440 22.31999 204 10.42000 34 1.72001 3 0.2

Total 1970 100.0

The median endpoint distance for toxic worst-casescenarios is 1.6 miles, while the median endpoint dis-tance for ammable worst-case scenarios is 0.4 miles.This re ects the differences in the physical nature of the two hazard classes and their worst-case scenarios,as described above.

6.1.2. Potentially Affected Population

Under the RMP Rule, the population potentiallyaffected by a release is de ned as the residentialpopulation inside a circle with radius equal to the

Table XI. Day-of-the-Week Pattern of Accidents

Percent of Day of the Number of TotalWeek Accidents Accidents

Sunday 157 8.0Monday 310 15.7Tuesday 308 15.6Wednesday 355 18.0Thursday 351 17.8Friday 281 14.3Saturday 208 10.6

Total 1970 100.0

8/10/2019 03-24-PK

13/17


Table XII. Plant Size vs. Accident Frequency

Proportion of FTEs at Facilities with Number of Facility Accidents (%) Facilities

10 12.9 7,654Total 7.8 15,424

endpoint distance. Therefore, for a given populationdensity, the population inside the worst-case cir-cle will increase according to the area of the circle,or proportionally to the square of the endpoint dis-tance. Naturally, population density is not constant,and other factors such as terrain, geography, zoning,

etc., also affect this correlation. But, in general, onewouldexpect to see population increase as the squareof endpoint distance.

Figs. 3 and 4 are histograms of the potentially af-fected population for toxic and ammable worst-casescenarios. The distribution of the potentially affectedpopulation among the toxic worst-case scenarios ishighly right-skewed: a mean of over 40,000 peoplewould be affected per scenario, whereas the medianscenario would affect 1,500. The distribution of thepotentially affected population among the ammableworst-case scenarios is also highly right-skewed, al-though the estimated number affected is smaller: amean of668perscenario, with themedian scenarioaf-fecting 15.In evaluating these results, it isagain impor-tant to consider the physical difference between toxicand ammable worst-case scenarios. Toxic chemicalreleases generally result in a plume that travels in thedownwind direction. 25 Should an accidental releaseoccur, only the portion of the population covered bythe plume could feel its effects. This population usu-ally represents only a minor fractionof thepopulationinside the worst-case circle. Therefore, the OCA gen-erally overestimates the impact of a toxic release.

Flammable worst-case scenarios, on the other

hand, consist of an overpressure blast wave that gen-erally travels in all directions from the source. Whileterrain andobstructions will affect the propagation of the blast wave to some degree, in general, everyonewithin the worst-case circle would feel the effects of a vapor cloud explosion resulting from a ammable

25 Under certain conditions, the direction that a toxic gas plumetravels may be dictated more by the elevation of surroundingterrain than by wind direction.

substance release. So, while Figs. 3 and 4 indicate avery largedisparitybetween potentially affected pop-ulation for toxic and ammable worst-case scenarios,the disparity is, in fact, not as great as these guresindicate.

It is interesting to note that the distribution of residential population potentially affected by toxicworst-case scenarios appears to be log-normal inshape but that the ammable worst-case scenario dis-tribution is clearly not log-normal. 26 It is unclear whythe two distributions have such markedly differentshapes, but the difference may be due partly becauseeach distribution is actually a collection of underlyingdistributions, one for each different chemical repre-sented in the database. It is also possible that facilitieswith toxic chemicals tend to be located in areas withdifferent population densities than do facilities withammable chemicals, thereby affecting the popula-tions at risk. Further, while EPA modeling (i.e., EPAlookup tables and RMP*Comp software) was used toobtain the majority of OCA results in the database,thefact that several otheranalytical models were usedto obtain theremaining resultsprobably inducessomearti cial variations in these distributions.

6.2. Alternative Release Scenarios

The RMP regulation provides much greater ex-ibility in de ning alternative release scenarios thanworst-case scenarios. The only hard requirementsfor alternative release scenarios are that the scenariomust be more likely to occur than the worst-case sce-nario andthat it reaches an endpointoffsite,unless nosuch scenario exists. Facilities may account for bothpassive and active mitigation measures that may bein place when calculating the potential consequencesfrom an alternative release scenario. Alternative sce-narios are generally considered to be more repre-sentative of actual emergency scenarios that mightoccur.

Since there are no objective criteria for develop-ing alternative scenarios, the results vary widely, even

among similar facilities. Except for including the ba-sic parameters of the data distribution in Table XIII,this study has not attempted any in-depth analysis of alternative scenario data.

Table XIII indicates basic descriptive statisticsfor endpoint distances and populations for toxic and

26 Due to the extremely wide range of residential populations (0 to12 million for toxic worst-case scenarios) both distributions areplotted on a logarithmic scale.

8/10/2019 03-24-PK

14/17


Frequency Histogram - OCA Toxic Endpoint Distance

0

5 0 0

1 0 0 0

1 5 0 0

2 0 0 0

2 5 0 0

3 0 0 0

3 5 0 0

4 0 0 0

4 5 0 0

5 0 0 0

1 2 3 4 5 6 7 8 9 1 0 1 1 1 2 1 3 1 4 1 5 1 6 1 7 1 8 1 9 2 0 2 1 2 2 2 3 2 4 2 5 > 2 5

Endpoint Distance (miles)

N u m b e r o f F a c i l i t i e s

Fig. 1. Frequency histogram endpoint distance for toxic worst-case scenario.

ammable worst-case and alternative release scenar-ios. As expected, alternative release scenarios forboth toxic and ammable substances have, in gen-

eral, shorter endpoint distances and smaller popula-tions than do the worst-case scenarios for the same

Frequency Histogram - OCA Flammable Endpoint Distance

0

2 0 0

4 0 0

6 0 0

8 0 0

1 0 0 0

1 2 0 0

1 4 0 0

1 6 0 0

1 8 0 0

2 0 0 0

< 0 .2 5 0 .2 5-

0 .5

0.5-

1.75

1 .7 5 -1 1 -1 .2 5 1 .2 5 -

1 .5

1.5-

1.75

1 .7 5-2 > 2

Endpoint Distance (miles)

N u m b e r o f F a c i l i t i e s

Fig. 2. Frequency histogram endpointdistance for ammable worst-casescenarios.

hazard class. Similarly, as ammable worst-case sce-narios are generally less severe than toxic worst-casescenarios, so are ammable alternative scenarios less

severe thantoxic alternative scenarios. TableXIII alsoeffectively highlights the much larger scale of toxic

8/10/2019 03-24-PK

15/17

8/10/2019 03-24-PK

16/17


Table XIII. Descriptive Statistics forWorst-Case and Alternative Release

Scenarios

Type of Scenario

Toxic Flammable FlammableToxic Alternative Worst Alternative

Distance or Population Worst Case Release Case Release

Endpoint distance (miles)Mean 2.9 0.45 0.44 0.13Median 1.6 0.22 0.4 0.1Mode 1.3 0.1 0.4 0.1Std. Dev. 4 0.66 0.39 0.15Range 40 18 6.4 4.4

Residential populationMean 38,161 937 713 66Median 1,410 40 12 0Mode 0 0 0 0Std. Dev. 2.8 105 1.4 104 4.7 103 4.5 102

Range 1.2 107 1.6 106 1.5 105 1.1 104

Rule? Another important area, going forward, willbe to evaluate desirable changes in RMP*Info for thenext reporting of accident history data, scheduled totake place in 2004. Some questions for further studyinclude:

1. Do the data reveal the need for any policy,practice, or regulatory changes with regard toparticular chemicals, industrial sectors, pro-cesses, or equipment?

2. Docorrelationsexistbetweenaccidenthistorydata and other data elements (in RMP*Infoor other databases, e.g., nancial data, OSHAdata on safety and health statistics, data onacute health effects as in the HazardousSubstances Emergency Events Surveillance(HSEES) database, TRI data, etc.) that mightserve as predictors of accident-prone oraccident-free performance?

3. Does the database constitute a large enoughsample of chemical facilities to determinerisk distributions with suf cient con denceto make decisions about low-frequency, high-

consequence events at the tail end of the dis-tribution?4. Comparing the combined accident history

data from the 1999 ling with the 2004 l-ing, what trends or patterns in accidents areevident?

5. What changes to the database or RMP regu-lation might be necessary to correct de cien-cies in this important database to make the

data more meaningful and useful for accidentprevention?

ACKNOWLEDGMENTS

This work is part of on-going work by theWharton Risk Management and Decision ProcessesCenter under a Cooperative Agreement with U.S.EPA/CEPPO on risk management in the chemical in-dustry and, speci cally, on the implementation of sec-tion 112(r) of the Clean Air Act Amendments. Cen-ter Co-Director Howard Kunreuther and the Cen-ter s EPA Cooperative Agreement Project ManagerPatrick McNulty have played important roles in shap-ing and guiding this research. This report has bene-ted greatly from discussions with and comments onan earlier draft by Russell Localio of the Center forClinical Epidemiology and Biostatistics (CCEB) atthe University of Pennsylvania School of Medicine,and from the assistance of statistician Yanlin Wang of CCEB. The authors are particularly grateful for theadvice of IrvRosenthal of the Chemical Safety Board

for his early leadership in launching this project andfor the advice andassistance of James Makris, BreedaReilly, and Karen Schneider of U.S. EPA/CEPPO.None of the above individuals should bear the blamefor any errors or omissions in this report. Com-ments on this report may be sent to Paul Kleindor-fer ([email protected]) or Kiwan Lee([email protected]). Readers who wish tohave access to other materials on the Wharton Risk

8/10/2019 03-24-PK

17/17


Center s work on accident prevention in the chem-ical industry should consult the Center s website athttp://opim.wharton.upenn.edu/risk/.

REFERENCES

29 CFR Part 1910, Process Safety Management of Highly Haz-ardous Chemicals;Explosivesand Blasting Agents, Final Rule,57 FR 6356, February 24, 1992.

40 CFR Chapter IV, AccidentalRelease PreventionRequirements;Risk Management Programs Under the Clean Air Act Section112(r)(7); Distribution of Off-Site Consequence Analysis In-formation; Final Rule, 65 FR 48108, August 4, 2000.

40 CFR Part 68, Accidental Release Prevention Requirements:Risk Management Programs Under the Clean Air Act, Sec-tion 112(r)(7); List of Regulated Substances and Thresholdsfor Accidental Release Prevention, Stay of Effectiveness; andAccidental Release Prevention Requirements: Risk Manage-ment Programs Under Section 112(r)(7) of the Clean Air Actas Amended, Guidelines; FinalRulesand Notice, 61 FR 31668,June 20, 1996.

Belke, J. (2001). Chemical accident risks in U.S. industry A pre-liminary analysis of accident risk data from U.S. hazardouschemical facilities. In H. J. Pasman, O. Fredholm, and A.Jacobsson (Eds.), Proceedings of the 10thInternational Sympo- sium on Loss Prevention and Safety Promotion in the Process Industries , Stockholm, Sweden, Amsterdam: Elsevier ScienceB.V.

Elliott, M. R., Kleindorfer, P. R., & Lowe, R. (2003). The role of hazardousness and regulatory practices in the accidental re-lease of chemicals at U.S. industrial facilities. Risk Analysis , 23(5), 883896.

Elliott, M. R., Wang, Y., Lowe, R. A., & Kleindorfer, P. R. (Inpress). Environmental justice: Frequency and severity of U.S.chemical industry accidents and the socio-economic status of surrounding communities. Journal of Epidemiology and Com-munity Health .

Klassen, R. D., & McLaughlin, C. P. (1996). The impact of environ-mental management on rm performance. Management Sci-ence , 42(8), 11991214.

Mannan, H. S., & O Connor, T. M. (1999). Accident historydatabase: An opportunity. Environmental Progress , 18(1), 16.

Public Law 99-499, Superfund Amendments and ReauthorizationAct of 1986, Title III, Emergency Planning and CommunityRight-to-Know Act.

Public Law 101-549, Clean Air Act Amendments of 1990, Title III,Sections 304, 301, November 15, 1990.

Public Law 106-40,Chemical Safety Information, Site Security, andFuels Regulatory Relief Act, August 5, 1999.

Rosenthal, I. (1997). Investigating organizational factors re-lated to the occurrence and prevention of acciden-tal chemical releases. In A. Hale, B. Wilpert, & M.Freitag (Eds.), After the Event: From Accident to Organisa-tional Learning (pp. 4162). New York: Pergamon, Elsevier

Science.Saari, J. (1986). Accident epidemiology. In M. Karvonen & M. I.Mikheev (Eds.), Epidemiology of Occupational Health , Euro-pean Series No. 20 (pp. 300 320). Copenhagen: World HealthOrganizations Regional Publications.

U.S. Environmental Protection Agency, Chemical Emergency Pre-paredness and Prevention Of ce (USEPA/CEPPO). (1996).Economic Analysis in Support of Final Rule on Risk Manage-ment Program Regulations for Chemical Accident Release Pre-vention, as Required by Section 112(r) of the Clean Air Act .Washington, D.C.: EPA.

03-24-PK

Documents

Transcript of 03-24-PK