Will Additional Federal Enforcement Improve

the Performance of Pipelines in the U.S.?

Sarah Stafford College of William and Mary

College of William and Mary Department of Economics

Working Paper Number 144

Current Version: August 2013

COLLEGE OF WILLIAM AND MARY DEPARTMENT OF ECONOMICS WORKING PAPER # 144 August 2013

Will Additional Federal Enforcement Improve the Performance of Pipelines in the U.S.?

Abstract

This paper provides the first empirical analysis of the effectiveness of regulatory enforcement in increasing the environmental and safety performance of U.S. natural gas and hazardous liquid pipeline operators. The analysis combines data on federal regulatory inspections, enforcement actions, and penalties with data on injuries, fatalities, property damage, and barrels of product lost through pipeline “incidents” for 2006-2011 for the 344 largest pipeline operators in the U.S. The results of the analysis do not provide compelling evidence that either federal inspections or civil penalties are particularly effective in increasing performance; however, the number of federal cases initiated against an operator does have a significant effect on many forms of performance, although not for incidents in general. The results also suggest that some targeting of federal enforcement resources is based on past performance, but there may be room for even more effective targeting. Finally, the analysis reveals interesting patterns between state and federal enforcement efforts. Keywords: Regulation, Environmental Performance, Enforcement, Pipelines JEL Classification: K42, L95 Sarah Stafford Department of Economics College of William and Mary Williamsburg, VA 23187-8795 slstaf@wm.edu

2

WillAdditionalFederalEnforcementImprovethePerformanceofPipelinesintheU.S.?

1. Introduction

Overthepastseveralyears,therolethatoilandnaturalgaspipelinesmightplayin

increasingtheU.S.’senergyindependencehasgainedsignificantattention.Inparticular,

TransCanada’sproposedKeystoneXLPipelinehasbeenthesubjectofheateddebate

betweenthosethatbelievetheprojectisacriticalpartoftheU.S.’senergysecuritystrategy

andwillhaveapositiveeffectonthecountry’seconomyandthosethatbelievetheproject

imposesunacceptablerisksforthenaturalenvironmentincludingdevastatingsensitive

environmentsandpollutingimportantwatersources.Anumberofrelativelyrecentevents

havereinforcedtheargumentsthatpipelinesposeseriousthreatstohumanhealthandthe

environment:inSeptemberof2010anaturalgaspipelineexplosioninSanBruno,

Californiaresultedinamassivefirethatkilledeightpeople,injureddozensofothers,and

destroyedover100homesandinJulyof2011anExxonMobilpipelinerupturespilledover

1,000barrelsofoilintothescenicYellowstoneRiver.

Inlate2011,theU.S.CongressapprovedandPresidentObamasignedthePipeline

Safety,RegulatoryCertainty,andJobCreationActtoimprovetheperformanceofpipelines.

Theactwaspassedduringthe112thCongress,oneoftheleastproductive–ifnottheleast

productive–legislativesessioninrecenthistory[1].Theactdrewunanimoussupportfrom

bothpartiesinpartbecauseofpublicoutcryovertheSanBrunoexplosionandthe

YellowstoneRiverspill.However,theactwasacompromiseanddidnotincludeallofthe

recommendedpolicychangesthatwereproposedbytheNationalTransportationSafety

Boardforincreasingpipelinesafety[2].Themainprovisionsoftheactareanincreasein

3

fundingforfederalinspectionsofpipelines(the“JobCreation”partoftheact)aswellasan

increaseinthefinesassociatedwithviolationsofpipelineregulations.Inaccordancewith

theact,theadministration’s2013fiscalyearbudgetincreasedfundingforthePipelineand

HazardousSafetyMaterialsAdministrationby60percentandadded120newfederal

inspectors.

Whilenumerousstudieshaveassessedtheeffectivenessoffederalenforcementin

improvingcompliancewithgeneralenvironmentalregulations,tomyknowledgetherehas

neverbeenasystematicevaluationoftheeffectoffederalenforcementeffortsonpipeline

performance.ThusitisnotclearwhetherthePipelineSafety,RegulatoryCertainty,andJob

CreationActwillactuallyaccomplishitsstatedgoalofincreasingpipelineperformance.In

particular,becausetheactwaspromptedbypublicpressuretodosomethingabout

pipelineperformance,asMay[3]pointsout,thecompromisesolutionmaynotfully

addresstheunderlyingregulatoryfailure.Thegoalofthispaperistoprovidethefirst

empiricalanalysisoftheeffectthatfederalpipelineenforcementonpipelineperformance.

Theresultsofthisanalysisshouldprovideinsightintowhetherthechangesmandated

underthePipelineSafety,RegulatoryCertainty,andJobCreationActarelikelytoachieve

theirgoalofimprovingpipelinesafety.

2. BackgroundonthePipelineIndustry

Manyliquidproductsaremostcost‐effectivelytransportedviapipelines.However,

manyoftheproductstransportedbypipelinecanposesignificantthreatstohumanhealth

andtheenvironmentifleakedorreleasedfromthepipeline.Althoughpipelinesare

designedandconstructedtomaintainstructuralintegritysincethetransportedmaterials

4

haveintrinsicvalue(unlikemanyeffluentsubstances,suchashazardouswastesorby‐

products),manyfactorsmakeitdifficulttoavoidleaksandotherreleasesduringa

pipeline’slifetime.Naturaldisasters,suchasflooding,earthquakes,andstorms,canresult

inpipelinefailures,ascanaccidentalhuman,machine,andanimalintrusions.Additionally,

pipelinesmaydevelopleaksorrupturesduetocorrosionfromthematerialsbeing

transportedormaterialfatiguefromfluctuatingtemperatureandpressureconditions.

IntheU.S.over2.5millionmilesofpipelinestransportnaturalgas,petroleum

productsandotherhazardousliquids.Overall,pipelinesarearelativelysafemodeof

transportationcomparedtoalternativessuchastankersandrailcars,andthepipeline

transmissionsafetyrecordhasimprovedsignificantlyovertime.However,morethan100

significantpipelinereleasesoccureachyear,anddeathsfrompipelineaccidentsare,

unfortunately,notrareoccurrences.

Priorto1968,pipelineswerenotsubjecttosafetyorenvironmentalregulations.In

1968,CongressestablishedtheOfficeofPipelineSafety(OPS),adivisionoftheDepartment

ofTransportation(DOT),todevelopandimplementsafetyregulationsfornaturalgas

pipelines.HazardousliquidpipelineswereaddedtoOPS’sportfolioin1979,butuntil2002

OPSwasgenerallyseenasineffectual,withweakenforcementandineffectiverules[4].In

2002,CongresspassedthePipelineSafetyImprovementAct,whichincreasedpenaltiesand

enforcementauthority,andlimitedOPSdiscretion.

OPSsetsthefederalstandardswithwhichallpipelineoperatorsmustcomply.Asis

truewithmanyotherregulations,statescananddopasssupplementalregulations.

Additionally,pipelinesin“highconsequence”areasaresubjecttoastrictersetofcontrols

duetotheincreasedriskfordamagetohumanhealthortheenvironment.Bothfederaland

5

stateregulatorsenforceOPSregulations.Intheory,standardinspectionsareconducted

everycoupleofyearsonallpipelinesandmoreoftenonpipelineswithhigherpotential

risks.Ifapipelinecrossesstateborders,enforcementgenerallyfallstoOPS,whilestates

inspectmostintrastatelines.However,notallstateshavebeencertifiedorapprovedto

conductintrastateinspections;inunapprovedstatesfederalregulatorsconductallpipeline

inspections.Conversely,OPShasauthorizedsomestatestoactasitsagentandinspectthe

sectionsofinterstatepipelinesthatrunthroughthestateinadditiontointrastatepipelines.

Tocomplementformalenforcement,regulatedpipelinesmustalsoself‐inspectandreport

anyviolationsdiscoveredduringthecourseofrequiredinspections.

OPSisarelativelysmallagency.In2011priortothepassageofthePipelineSafety,

RegulatoryCertainty,andJobCreationAct,therewereunder120inspectorsworkingfor

OPSoutoffiveregionaloffices(Trenton,NJ;Atlanta,GA;KansasCity,MO;Houston,TX;and

Denver,CO)[5].Anadditional300stateinspectorscarryoutthemajorityofpipeline

inspections.Standardinspectionsaredesignedtoensurethatoperationandmaintenance

procedures,abnormalandemergencyoperatingprocedures,damagepreventionand

publiceducationprocedures,andpipelineinstallation,connection,repair,andoperations

areincompliancewiththerelevantregulations.Constructioninspectionsincludeareview

ofmaterialandcomponentdesignspecifications,weldingproceduresandwelder

qualifications,corrosionprotection,andinstallationaswellaspost‐constructiontesting.

Integritymanagementinspectionsaredesignedtodeterminewhetheranoperatorusesall

availableinformationaboutitspipelinesystemtoassessrisksandtakesappropriateaction

tomitigatethoserisks.

6

OPScaninitiateanenforcementcasewhenaninspectionidentifiesaviolationof

pipelineregulationsorinresponsetoanaccident.Thetypeofenforcementactiontaken

dependsonthesignificanceoftheviolation.Minorproblemsoccurringforthefirsttime

mayonlyreceiveawarningletter,whilemoresignificantviolationsmayrequirea

complianceorderthatspecifiesactionstheoperatormusttaketocomeintocompliance

(e.g.,requiringoperatorstoreplacepipelinesectionsorimplementcorrosioncontroland

remediationstrategies)oracivilpenalty.Civilpenaltiesaregenerallyreservedforserious

violationsleadingtodeaths,injuries,orsignificantenvironmentaldamage.Regulatorsmay

imposecivilpenaltiesassevereas$100,000foreachdayaviolationexisted,uptoa

maximumof$1,000,000.Since2008,OPShasproposedover$21millionincivilpenalties

[6].

Therearecurrently2,705regulatedpipelineoperatorsintheU.S.Ofthese,1,921

operatelessthan10milesofpipeline,440operatebetween10and100milesofpipeline,

and344operate100milesormoreofpipeline.In2010,22fatalitiesand109injurieswere

attributedtopipelineincidents.Ofcoursethesenumbersarequitevariable–overthelast

20years,thenumberoffatalitieshasrangedfromalowof7in2001toahighof53in

1996.Similarlythenumberofinjurieshasrangedfromalowof36in2006toahighof127

in1996.Ofcourseinjuriesanddeathsarenottheonlydamagesthatresultfrompoor

pipelineperformance.In2010,pipelineincidentsresultedinalmost$1.4billiondollarsof

propertydamageandalmost175,000barrelsofspilledhazardousliquids.Onthe

enforcementside,in2010federalregulatorsconductedaround600pipelineinspections,

initiatedjustover200enforcementactionsandassessedover$4.5milliondollarsin

penalties.Duringthesametimeperiodstateregulatorsloggedalmost38,000inspection

7

days,discoveredalmost14,000violations,initiatedover4,000enforcementactions,and

assessedover$13milliondollarsinpenalties.

3. RelatedLiterature

Theobjectiveofthispaperistobetterunderstandtherolethatfederalinspections

andenforcementactionsplayinincreasingpipelineperformanceandcompliance.Tomy

knowledge,therearenoexistingpapersthatexplicitlymodelcompliancewithpipeline

regulations,eithertheoreticallyorempirically.1However,thereisalargeliterature

examiningcompliancewithenvironmentalregulationsmorebroadly,andIusethisasa

startingpointfortheanalysis.

Thetraditionaleconomicviewofenvironmentalcomplianceandperformance

assumesthataregulatedentity’sdecisiontocomplywithenvironmentalregulationsisa

rationalonebasedontheobjectiveofprofitmaximization.Thebasicframeworkforthese

modelsisBecker's[8]paperontheeconomicsofcrime,whichwasadaptedbyRussell,

Harrington,andVaughan[9]toprovideacomprehensiveapplicationtoenvironmental

regulation.Whileanumberofinterestingvariationsonthesemodelshavebeendeveloped

overthepasttwodecadestoallowforvariouscomplexitiessuchasimperfectinformation,

self‐reporting,principal‐agentrelationships,anddynamicsettings,inallofthese

deterrence‐basedmodelscomplianceandperformanceareultimatelyimprovedby

increasingtheexpectedcostofnoncompliance–eitherbyincreasingthelikelihoodthata

violatorgetscaughtorbyincreasingthelevelofsanctionsassociatedwithviolations.

1Thereareanumberofpapersthatanalyzepipelineincidentsfromanengineeringperspectivetobetterunderstandthedistributionofpipelinefailures(see,forexampleSosaandAlvarez‐Ramirez[7]).Thesepapersdonotexamineregulatorystructuresorpolicies.

8

Whiledeterrence‐basedmodelsdominatetheeconomicsliteratureon

environmentalenforcement,anumberofpapersinotherfieldshaverecognizedalternative

motivationsforcomplianceorreasonsfornonperformance.Forexample,aregulatedentity

maycomplywithregulationsoutofaninherentsenseofdutytoobeyrulesorbecauseof

socialpressure,eveniftheprobabilityofdetectionisveryloworthepunishmentfora

violationisnegligible.Alternatively,evenwithseveresanctionsorahighprobabilityof

detection,ifaregulatedentity’smanagersdonotunderstandtheregulatoryrequirements,

orhavepoorinternalcontrols,evenwellintentionedregulatedentitiesmaystillviolate

regulations.Finally,someviolations,suchasthosetriggeredbyextremeweather,may

occurdespitearegulatedentity’sfullycompliantoperations.Insuchcases,deterrence‐

basedmeasureswouldprovegenerallyineffectiveatincreasingperformance.Ofcourse,

whilesometheoreticalmodelsfocusonaparticularmotiveunderlyingthecompliance

decision,inpracticethecompliancedecisionislikelytodependonanumberofdifferent

objectivesandfactorsthatdifferacrossfacilities.

AccordingtoGrayandShimshack[10],mostpolicy‐makersandscholarsbelieve

thatanenforcementregimeofinspectionsandsanctionsisgenerallyeffectiveatincreasing

compliancewithenvironmentalregulations,andmostregulatedentitiesciterigorous

monitoringandenforcementasaprimarymotivatoroftheirenvironmentalcompliance

decisions.Anumberofempiricalanalysesconfirmthesebeliefs.Forexample,Grayand

Deily[11]andGrayandShadbegian[12]examineairpollutioncomplianceforsteelmills

andpulpandpapermillsintheU.S.,respectively,andfindthatbothinspectionsand

enforcementactionshaveastatisticallysignificantpositiveimpactoncompliance.Looking

atcompliancewithU.S.waterregulations,Earnhart[13]andGlicksmanandEarnhart[14]

9

similarlyfindthatinspectionsandsanctionsdeterviolationsandreduceemissionsatwater

treatmentplantsandchemicalfacilities,respectively.Stafford[15]showsthatcompliance

inspectionsandpenaltiesforviolationshaveasignificantdeterrenteffectonviolationsat

facilitiessubjecttohazardouswasteregulations.2

Theseresultsfromtheenvironmentalcomplianceliteratureechofindingsinother

regulatoryareas.Inparticular,anumberofpapersexaminethedeterrenteffectof

OccupationalSafetyandHealthAdministration(OSHA)inspectionsandsanctionson

workplaceinjuries.Manyofthesepapersfindthatinspectionsandsanctionsdodeter

injuries,althoughtheeffectsofdeterrencedependsignificantlyonthecharacteristicsofthe

regulatedentitybeinginspectedorsanctionedandwhethertheinspectionresultsina

sanction[16,17,18].Thegoalofthispaperistoaddevidencefromanothercloselyrelated

regulatorysector,pipelines,onthedeterrenteffectoffederalinspectionsandenforcement

actionsinincreasingcomplianceandperformance.

4. FrameworkfortheAnalysisandDescriptionoftheData

Whilepipelinesarefixedstructures,theyarenotconstrainedwithinaparticular

geographicarealikemostentitiessubjecttoenvironmentalandsafetyregulations.While

manypipelinearerelativelyshort,therearealsooperatorsthathavethousandsofmilesof

pipelinecrossingnumerousstateborders.Federalandstateregulatorsdodividepipelines

into'inspectionunits'.Foroperatorswithshortpipelines,theentirecompanymay

constituteoneinspectionunitwhilelargeroperatorsmaybedividedbasedonoperating

areas(e.g.,citiesormetropolitanareas)orcompanyorganization(e.g.,allelements2SeeGrayandShimshack[10]foracomprehensivesurveyoftheempiricalliteratureonenvironmentalmonitoringandenforcement.

10

reportingtoasinglevicepresident).Unfortunately,dataonpipelineperformanceand

enforcementisnotavailableattheinspection‐unitlevel.Thusthisanalysisfocusesonthe

aggregateperformanceofindividualpipelineoperators,ratherthantheperformanceofa

particularsectionofapipeline.Thisanalysisismostanalogoustofirm‐levelstudiesof

complianceandenvironmentalperformance,suchasKhannaandAnton[19]andThornton,

Gunningham,andKagan[20],althoughitisbasedondatareportedtothefederal

governmentratherthandatacollectedthroughavoluntarysurvey.

Asdiscussedearlier,thereare2,705regulatedpipelineoperatorsintheU.S,over

two‐thirdsofwhooperatelessthan10milesofpipeline.Thisanalysisfocusesonthe344

operatorsthatoperate100milesormoreofpipeline.Theseoperatorsrepresentover90%

ofallpipelineincidentsthatoccurredbetween2006and2011and80%ofallfederal

inspectionsduringthatsametimeframe.OPSdefinesanincidentasanyeventthatresults

inadeathorpersonalinjurynecessitatingin‐patienthospitalization;anexplosionor

unintentionalfire;anyeventthatresultsinpropertydamageof$50,000ormore(excluding

costofmateriallost);anyeventthatresultsinunintentionallossoffivegallonsormoreof

hazardousliquidorcarbondioxideorthreemillioncubicfeetofgas;anyemergencythat

resultsinanemergencyshutdownofafacility;oranyothereventthatissignificantinthe

judgmentoftheoperator.

ThePipelineandHazardousMaterialsSafetyAdministration(PHMSA)provides

dataontheperformanceofpipelineoperatorsstartingin2006.3Theperformance

measuresincludethetotalnumberofreportedincidents,fatalities,andinjurieseachyear;

thetotaldollaramountofpropertydamagereportedeachyear;andthereportedtotal

3TheOPSisanofficewithinthePHMSA.

11

barrelsofproductspilledandthenetbarrelsofproductlosteachyear.Whilethesedataare

self‐reported,thecivilpenaltiesfornotreportinganincidentwithin30dayscanbeupto

$1milliondollarsandanyindividualthat“willfullyandknowinglyviolates”the

requirementscanfaceacriminalfineofupto$25,000andbeimprisonedforuptofive

years.Additionally,manyofthepipelineincidentsdirectlyaffectorinvolvethirdparties,

makingitmuchlesslikelythatoperatorscouldunder‐reportthoseincidents.Forexample,

almostthree‐quartersofthepipelinefatalitiesandthree‐quartersofthepipelineinjuries

reportedduringthe2008to2012periodinvolvedthird‐party(non‐industry)individuals.4

Table1presentsasummaryoftheperformancemeasuresfor2010forthe

operatorsinthisstudy.First,notethatforallofthesemeasures,themajorityofoperators

havenothingtoreport.Themostwidelyreportedmeasureispropertydamage,followed

closelybyincidents.Propertydamageisreportedmoreoftenthanincidentsbecauseevents

thatcauselessthan$50,000inpropertydamagearenotconsideredincidentsiftheydonot

alsoresultinfatalities,significantinjuries,orsufficientlossofmaterial.Giventherelatively

smallnumberofoperatorsthatreportinagivenyear,Iaggregateperformancedatafor

2009and2010toincreasethenumberofoperatorsreporting.Themeanandstandard

deviationsfortheaggregateddataarepresentedinTable2whichincludessummary

statisticsforallvariablesusedintheanalysis.Notethatthesummarystatisticsareforall

operatorsinthestudy,notjustthosereporting.

Oneoftheprincipalchallengesthatcanarisewhentryingtoestimatethe

effectivenessofinspectionsandenforcementonperformanceisthatofendogeneityor

reversecausality,whichcanoccurifthereareomittedexplanatoryvariablesorthe4See“ConsequencestothePublicandthePipelineIndustry”availableathttp://primis.phmsa.dot.gov/comm/reports/safety/cpi.html(lastaccessedJuly11,2013).

12

complianceandenforcementdecisionsaremadesimultaneously.Withrespecttothe

omittedvariablesconcern,duetodatalimitationstheanalysismaynotincludesome

factorsthataffectboththeoperator’senvironmentalperformanceaswellastheregulator’s

decisiontoconductinspections.Forexample,significantfloodinginanareamaycause

pipelinestorupture,butmightalsobringincreasedinspectionstothatarea.Withrespect

tothesimultaneityconcern,contemporaneousinspectionsmaybeendogenoustothe

numberofincidentsreportedifinspectionsserveasasignificantmechanismthrough

whichincidentsarediscoveredorreported.Similarly,thenumberofenforcementcases

andamountofproposedpenaltiesinaparticularperiodarelikelytodependonthe

numberofincidentsandfatalitiesthatoccurinthatsametimeperiod.Toaddressthis

concernIlagtheenforcementvariables,whichmaybeendogenous,andIalsoincludethe

laggeddependentvariableasanexplanatoryvariable.IdeallyIwouldalsousean

instrumentalvariablesapproachtocontrolforendogeneity,butduetothelimited

informationavailableaboutpipelineinspectionsandenforcement,Ihavenotbeenableto

findanyvalidinstrumentstouseforsuchanapproach.

ThefirstsetofexplanatoryvariablespresentedinTable2depictsthelevelof

federalandstateenforcementforeachoperatorintheanalysis.Thethreefederalmeasures

–FederalInspections06‐08,FederalCasesInitiated06‐08,andFederalProposedPenalties06‐08–

eachcaptureadifferentaspectofthespecificdeterrenceaparticularoperatorfacesfrom

federalsources,astheycapturethelevelofinspectionsandenforcementforthatspecific

operatorduringthe2006‐2008period.Incontrast,thethreestatemeasuresareallgeneral

deterrencemeasuresthatcapturethegenerallevelofenforcementforthestatesthrough

whichtheoperator’spipelinesrun.Thestatemeasuresaregeneralmeasuresratherthan

13

specificmeasuresbecausethestatedataisonlyavailableattheaggregatelevel.Foreach

stateIfirstnormalizetherelevantvariableX–StateInspections06‐08,StateActionsTaken06‐

08,andStateAssessedPenalties06‐08–bythetotalnumberofpipelinemilesinthestate.For

eachoperatori,Ithenusedataonthetotalnumberofpipelinemilestheoperatorhasin

eachstatejtoconstructeachmeasureXforthatoperatorasfollows:

Milesij *X j

Miles jj

ThestateinspectiondatawasobtainedthroughaFreedomofInformationActrequest,

whilethedataonstatecomplianceactionstakenandpenaltiesassessedwascollectedfrom

thePHMSAwebsite.

Thenextsetofexplanatoryvariablesmeasurespastreportedperformance(i.e.,

reportedperformanceduringthe2006‐2008period)and,duetolimiteddatacapturing

operatorcharacteristics,isusedinconjunctionwiththeanalogous2009‐2010variablesto

controlfordifferencesinunderlyingpropensitiestocomplywithpipelineregulations.

Additionally,SosaandAlvarez‐Ramirez[7]showthatthenumberofpreviousincidents

positivelycorrelateswithfutureincidents.OneoftheoperatorcharacteristicsthatIcan

controlforistheMilesofpipelinetheoperatorowns.BothMilesandMilesSquaredare

includedintheanalysistoaccountforthefactthatlongerpipelineshavemore

opportunitiesforfailure.IalsoincludethedummyvariableIntrastatethatindicates

whetherthepipelineisconfinedwithinasinglestate.WhileOPSconcentratesenforcement

effortsoninterstatepipelines,federalinspectorsdoinspectintrastatepipelineson

occasion.NumberofStatesmeasuresthenumberofstatesthroughwhichthepipeline

passes,whilethefourregionaldummiescapturetheCensusregion(s)inwhichthe

14

operatoroperates.Finallytherearefourdummyvariablesthatcapturethetypeof

pipelinesandthematerialstransportedinthepipelinesthateachoperatorowns:

GasGatheringlinescollectandmovenaturalgasfromwellsoroffshorevesselsto

storageorprocessingfacilities.

GasTransmissionlinestransportnaturalgasfromgatheringlinesorstorage

facilitiestodistributioncenters,storagefacilities,powerplants,andindustrial

customersandmunicipalities.Thesearegenerallythelongesttypeofgaslinesand

areusuallyunderground.

GasDistributionlinesmovenaturalgastoindustrialcustomersandresidencesand

areusuallylocatedinundergroundutilityeasementsalongstreets.

HazardousLiquidlinestransportpetroleumproductsandotherhazardousliquids,

usuallyoverlongdistancesandunderground.

5. ResultsandPolicyImplications

Table3presentstheresultsoftheordinaryleastsquaresregressionforeachofthe

2009to2010reportedperformancevariables.Inthefirstcolumn,thedependentvariable

isthenumberofincidentsreportedin2009and2010.Lookingfirstatthefederal

enforcementvariables,noticethatnoneofthecoefficientsarenegative.Moreover,the

positivecoefficientsforFederalCasesInitiated06‐08andFederalProposedPenalties06‐08are

bothsignificant–theoppositeofwhatonewouldexpectifpastenforcementactionsserved

toincreaseoverallenvironmentalperformance.Onepossibleexplanationcouldbethatit

takesalongperiodoftimeforoperatorstochangetheirperformance;thus,operatorswith

pastincidentsthatwarrantedsignificantenforcementmaybemorelikelytocontinueto

15

reportahighnumberofincidents.Tocontrolforthis,Ididincludepastincidents

(DependentVariable06‐08)intheregression,whichalsohasapositiveandsignificantsign,

butitmaynotbeaperfectcontrol.Theresultsforthestateenforcementvariablesaremore

consistentwithexpectations.BothStateInspections06‐08andStatePenaltiesAssessed06‐08

havenegativecoefficients,andtheformerissignificant.

Lookingacrosstheotherperformancevariables,itisinterestingtonotethatthe

resultsfortheenforcementmeasuresarequitemixed.FederalInspections06‐08andFederal

ProposedPenalties06‐08,alwayshavepositivecoefficients,andthosecoefficientare

significantinanumberoftheregressions.Ontheotherhand,FederalCasesInitiated06‐08

hasanegativecoefficientforallbuttheIncidentsregression,andthecoefficientis

significantforallbuttheFatalitiesregression.Thestateenforcementresultsarealso

mixed.Incontrasttothenegativerelationshipbetweenfederalcasesandperformance–or,

morecorrectly,non‐performance–allofthesignificantcoefficientsonStateActions

Taken06‐08arepositive.Similarly,whilefederalproposedpenaltiesarepositivelyrelatedto

non‐performanceinmostoftheregressions,allofthecoefficientsonStatePenalties

Assessed06‐08arenegative.StateInspections06‐08hasanegativeandsignificantcoefficient

onlyintheIncidentsregression,buthasapositiveandsignificantcoefficientintheNet

BarrelsLostregression.Whileonemightexpectthatsomeofthisinconsistencycouldbe

causedbymulticollinearityamongthefederalandstateenforcementvariables,the

variablesarenotashighlycorrelatedasonemightexpect.Onlythreepairsofvariables

haveacorrelationcoefficientabove0.6:FederalInspectionsandFederalCasesInitiated

haveacorrelationcoefficientof0.67;FederalInspectionsandStateInspectionshavea

16

correlationcoefficientof0.70,andStateInspectionsandStateActionsTakenhavea

correlationcoefficientof0.76.

Clearlytheseresultspaintaverymixedpictureoftheeffectivenessoffederaland

stateenforcementeffortsatdeterringpoorenvironmentalperformanceatpipelines.In

particular,intermsofpredictingthesuccessofthePipelineSafety,RegulatoryCertainty,

andJobCreationActinincreasingpipelinesafety,thereisnoevidencethatindicates

federalinspectionsorfinesincreaseenvironmentalperformance,althoughthereissome

evidencethatstateinspectionsandpenaltiescanhavesuchaneffect.5WhileFederalCases

Initiated06‐08doeshavearelativelyconsistentnegativeandsignificanteffectonnon‐

performance,thePipelineSafety,RegulatoryCertainty,andJobCreationActexplicitly

focusesonincreasedfederalinspectionsratherthanmorerigorousenforcement;thus,itis

notclearhowthenumberofcasesinitiatedwillchangewithincreasedenforcement

resources.

TheremainingresultsinTable3providesomeinsightintowhyfederalenforcement

maynotbeparticularlyeffectiveatdecreasingpoorenvironmentalperformance.First,

observethatthecoefficientonthelaggeddependentvariableineachregression(listedas

DependentVariable06‐08)ispositiveandsignificantforfouroftheregressions.Thus,for

overallincidents,injuries,grossbarrelsspilled,andnetbarrelslost,thereisconsiderable

persistenceacrosstime–particularlywhenonerecallsthedifferenceintimeframesacross

thetwovariables(threeyearstotwoyears).Thelesspredictablenatureoffatalitiesand

5Onemightbeconcernedthatduetotheself‐reportednatureofthedata,someincidentsaresystematicallygoingunreportedinwaythatbiasesthefindingsofthisstudy.AppendixApresentstheresultsofasensitivityanalysisthatsuggeststhatunder‐reportingcannotexplainthelackofsignificantnegativecoefficientsonfederalinspections.

17

propertydamagemakesintuitivesenseandisconsistentwiththeSosaandAlvarez‐

Ramirez[7]findingthatmoresevereincidentsareunpredictable.

Ihadexpectedthatthenon‐performancemeasureswouldallbepositivelyrelatedto

thelengthofthepipeline,butinterestinglyMileshastheexpectedpositiveandsignificant

coefficientonlyintheIncidentsregression.ForFatalities,Injuries,andPropertyDamage,

longerpipelineshavefewernegativeoutcomes,ceterisparibus.Also,acrossallofthe

regressionsthecoefficientonMilesSquaredisnegative(althoughsignificantinonlytwoof

thesixregressions).Ofcourse,thereareanumberofothervariablesthatindirectlycapture

thelengthofthepipeline,includingthestateenforcementvariables.However,theseresults

suggestsforatleastsomeoftheperformancevariables,theremaybeimportantnon‐

linearities.

Whileveryfewoftheremainingexplanatoryvariableshaveaconsistenteffecton

theperformancevariables,notethatGasGatheringhasasignificantandpositivecoefficient

intheFatalities,Injuries,andPropertyDamageregressions.Comparingthesizeofthethree

significantGasGatheringcoefficientstothemeanandstandarddeviationforthethree

performancemeasures,notethatoperatingagasgatheringpipelineisquantitativelyavery

importantdeterminantforfatalities,injuriesandpropertydamageandmayhelpexplain

whyfederalandstatelevelenforcementactionsarenotmoreimportantdeterrentsforat

leastthesetypesofnon‐performance.

Althoughlaggingthefederalenforcementvariablesandconditioningonpriorvalues

ofthedependentvariablesshouldhelptoidentifyandestimatecausaleffects,asdiscussed

insection4onemightstillbeconcernedthattheseresultscouldbeduetoendogeneity.In

similarsituationsotherresearchershaveemployedaninstrumentalvariablesapproachto

18

trytocontrolforpotentialendogeneity,butgiventhelimitedinformationavailableabout

pipelineinspectionsandenforcement,Ihavenotbeenabletofindvalidinstrumentsfor

suchanapproach.Asanalternative,IuseManski’spartialidentificationapproach[21,22]

toestimateplausibleboundsforthecausaleffectsoffederalenforcement.Asdiscussedin

moredetailinAppendixB,thispartialidentificationapproachsuggeststhatallofthe

positivecoefficientsonthefederalenforcementvariablesintheIncidents,Property

Damage,GrossBarrelsSpilled,andNetBarrelsLostregressionsareplausibleevenifone

assumesthatregulatorsdotargetoperatorswithhigherlevelsofnon‐performancefor

enforcementactions.However,partialidentificationsuggeststhatsignificantpositive

coefficientsonthefederalenforcementvariablesfortheFatalitiesandInjuriesregressions

wouldnotbeconsistentwiththatassumption.Interestingly,Idonotfindanypositiveand

significantcoefficientsforthefederalenforcementvariablesineithertheFatalitiesorthe

Injuriesregressions.

ToprovideadditionalinsightintothemixedresultspresentedinTable3,Ialso

analyzedfederalinspectionsandenforcementasafunctionofpastperformance.Table4

presentstheresultsofordinaryleastsquareregressionsofFederalInspections09‐10,Federal

CasesInitiated09‐10,andFederalProposedPenalties09‐10asafunctionofthelagged

performancemeasures,thelaggeddependentvariable,andtheexplanatoryvariablesused

intheperformanceregressions.LookingfirstattheresultsfortheFederalInspections

regression,notethatonlyFatalities06‐08hasapositiveandsignificantcoefficientamongthe

performancemeasures,indicatingthatfederalinspectorsdotargetoperatorsfor

inspectionsiftherehavebeenfatalitiesattheoperator’spipelinesintherecentpast.

Interestingly,thecoefficientonInjuries06‐08isnegativeandsignificantwhichisnot

19

consistentwiththeideaoftargetingbasedonpastperformance.Thispatternshowsupin

bothoftheotherregressions;thatis,thecoefficientonFatalities06‐08ispositiveand

significantforbothFederalCasesandFederalPenalties,whilethecoefficientonInjuries06‐08

isnegativeforbothandsignificantforFederalCases.BarrelsLost06‐08isalsoasignificant

determinantofFederalCasesInitiated09‐10,whileIncidents06‐08andPropertyDamage06‐08are

significantdeterminantsofFederalProposedPenalties09‐10.

ForbothFederalInspectionsandFederalCasesthereissomepersistenceacrossthe

twoperiodsgiventhepositiveandsignificantcoefficientsonthelaggeddependent

variable.Thereareanumberofpossibleexplanationsforthisresult.Recallthatpipelines

whichpassthrough“highconsequence”areasaresubjecttomorestringentregulationand

mayalsofacemoreinspections.Similarly,pipelinescarryingparticularlyhazardous

materialsmaybeinspectedmoreoften.Interestingly,thereisanegativerelationship

betweencurrentandlaggedFederalPenalties,sothatfacilitiesthatfacedhigherpenalties

inthepastfacelowerpenaltiescurrently,ceterisparibus.

Next,considerthelaggedstateenforcementvariables.StateInspections06‐08hasa

positivecoefficientinallthreeregressions,anditissignificantforFederalCasesand

FederalPenalties.Ifstateinspectionsuncoverbehaviorthathelpsfederalregulatorsinitiate

enforcementproceedings,onewouldexpecttoseeapositiverelationshipbetweenthese

variables.Interestingly,thenegativeandsignificantcoefficientonStateActionsTaken06‐08

suggeststhatfederalregulatorsmaytakeintoaccountstateactionsandholdoffontheir

ownenforcementactionsagainstoperatorsthathavebeensubjecttostateactionsinthe

recentpast.However,thepositiveandsignificantcoefficientonStatePenaltiesAssessed06‐08

intheFederalCasesregressionisinconsistentwithsuchaninterpretation.

20

Lookingnextattheoperatorcharacteristicsvariables,asexpectedlongerpipelines

facemoreinspectionsthanshorterpipelines,althoughtheyarenotsubjecttomorefederal

casesorhigherfederalpenalties.Thisfindingmakessense,asinspectionsshoulddepend

onthepotentialforharm,whileenforcementactionsshoulddependonthepresenceof

actualharmorviolations.Theinsignificantcoefficientsonalloftheregionaldummies

indicatetherearenotsignificantdifferencesinthenumberofinspectionsbasedonthe

regionsthroughwhichapipelineruns.However,therearesignificantdifferencesinthe

numberoffederalcasesandpenaltiesproposedbyregion,evenaftercontrollingfor

performance.Whiletherearemanypossibleexplanationsforthesefindings,theyare

consistentwithregulatorsindifferentregionshavingdifferentopinionsaboutwhencases

shouldbeinitiatedandhowpenaltiesshouldbeset.Interestingly,eventhoughthe

regressionsinTable3suggestthatpipelineperformancedependsonthetypeofpipeline–

GasGathering,GasDistribution,etc.–thereisnovariationinfederalenforcementacross

thedifferenttypesofpipelines.

6. Conclusion

Thegoalofthispaperistoprovideinsightintotherolethatfederalinspections,

enforcementactions,andfineshavehadonpipelineperformanceand,inparticular,to

examinewhethertheincreasedinspectionsfundingandcivilpenaltiesmandatedunderthe

PipelineSafety,RegulatoryCertainty,andJobCreationActarelikelytoincreasepipeline

safety.Theresultsoftheanalysisdonotprovidecompellingevidencethateitherfederal

inspectionsorcivilpenaltiesserveasparticularlyeffectivedeterrents.Infact,Ifindthat

laggedfederalinspectionsandpenaltiesarepositivelyassociatedwithenvironmentalnon‐

21

performance,althoughtheresultshavetobeinterpretedwithsomecareaslaggingthe

enforcementvariablesmaynotfullycorrectforomittedvariablesorendogeneitybetween

enforcementandperformance.

Interestingly,myanalysisdoesfindthatthenumberoffederalcasesinitiated

againstanoperatordoeshaveasignificantdeterrenteffectonmanyformsofnon‐

performance,althoughnotforincidentsingeneral.Thus,intheoryincreasingthenumber

offederalcaseswouldresultinbetterenvironmentalperformance.However,thePipeline

Safety,RegulatoryCertainty,andJobCreationActfocusesonincreasinginspectionsand

fines,notincreasingthenumberofcases,althoughadditionalcasescouldindirectlyresult

fromtheAct.

Theanalysisoffederalinspections,enforcementcases,andproposedpenalties

suggeststhatsometargetingoffederalenforcementresourcesisbasedonpast

performance,buttheresultssuggestthattheremayberoomforimprovement.Iffederal

enforcementresourceswerebettertargeted,thedeterrenteffectofsuchresourcesmight

increase.Theanalysisalsopointsoutsomevariationacrossregionsinenforcementthat

couldindicateinefficientresourcedeployment.Finally,theanalysisrevealsinteresting

patternsbetweenstateandfederalenforcementefforts.Additionalresearchtobetter

understandtherelationshipbetweensucheffortscouldhelpincreaseourunderstandingof

howsuchresourcesarecurrentlycoordinatedandwhetherbettercoordinationmight

increasedeterrence.

22

Table1:2010PerformanceMeasuresforOperatorswith100orMoreMilesofPipeline(N=344)

PerformanceMeasure

FacilitieswithNothingtoReport

ForFacilitiesthatReportMean Std.Dev. Minimum Maximum

NumberofIncidents 236(69%) 3.96 4.75 1 26NumberofFatalities 340(99%) 2.75 3.50 1 8NumberofInjuries 337(98%) 9.14 18.57 1 51PropertyDamage(inMillion$s)

235(68%) 10.40 67.50 0.003 601

GrossBarrelsSpilled(thousands)

285(83%) 2.91 10.21 0.002 70.19

NetBarrelsLost(thousands)

298(87%) 2.66 10.68 0.001 70.19

23

Table2:SummaryStatisticsfortheVariablesUsedintheAnalysis

Variable Description Mean Std.Dev.PerformanceMeasures(DependentVariables)Incidents09‐10 Numberofincidentsreportedduring

2009‐2010.2.51 6.33

Fatalities09‐10 Numberoffatalitiesreportedduring2009‐2010.

0.04 0.47

Injuries09‐10 Numberofinjuriesreportedduring2009‐2010.

0.23 2.79

PropertyDamage09‐10 Propertydamagereportedduring2009‐2010inmillion$s.

3.63 38.29

BarrelsSpilled09‐10 Barrelsreportedspilledduring2009‐2010inthousandsofbarrels.

0.65 4.44

BarrelsLost09‐10 Netbarrelsreportedlostduring2009‐2010inthousandsofbarrels.

0.45 4.04

EnforcementMeasuresFederalInspections06‐08

Numberoffederalinspectionsattheoperator’sfacilitiesduring2006‐2008(100’s).

0.33 0.73

FederalCasesInitiated06‐08

Numberoffederalenforcementcasesinitiatedagainstoperatorduring2006‐2008.

1.24 2.40

FederalProposedPenalties06‐08

ProposedPenaltiesontheoperatorduring2006‐2008(million$’s).

0.39 0.25

StateInspections06‐08 Weightedsumoftotalstateinspectionsduring2006‐2008(100’s).

0.79 1.58

StateActionsTaken06‐08

Weightedsumoftotalstateactionstakenduring2006‐2008.

8.87 18.90

StatePenaltiesAssessed06‐08

Weightedsumoftotalstatepenaltiesassessedduring2006‐2008($100,000’s).

0.07 0.28

PastPerformanceMeasuresIncidents06‐08 Numberofincidentsreportedduring

2006‐2008.3.93 9.75

Fatalities06‐08 Numberoffatalitiesreportedduring2006‐2008.

0.03 0.20

Injuries06‐08 Numberofinjuriesreportedduring2009‐2010.

0.07 0.46

PropertyDamage06‐08 Propertydamagereportedduring2006‐2008inmillion$s.

2.06 8.63

BarrelsSpilled06‐08 Barrelsspilledduring2006‐2008inthousandsofbarrels.

0.92 5.04

BarrelsLost06‐08 Netbarrelslostduring2006‐2008inthousandsofbarrels.

0.52 3.62

24

Variable Description Mean Std.Dev.OtherOperatorCharacteristicsMiles Milesofpipeline,inthousands 1.42 2.60Intrastate =1ifalloperationsinthesamestate 0.39 0.49NumberofStates Numberofstatesthroughwhichthe

operator’spipelinepasses.3.17 3.28

Region1 =1ifanypipelineislocatedintheNortheast.

0.11 0.32

Region2 =1ifanypipelineislocatedintheMidwest.

0.38 0.49

Region3 =1ifanypipelineislocatedintheSouth. 0.63 0.48Region4 =1ifanypipelineislocatedintheWest. 0.29 0.45GasGathering =1ifoperationsincludenaturalgas

gathering.0.24 0.43

GasTransmission =1ifoperationsincludenaturalgastransmission.

0.75 0.44

GasDistribution =1ifoperationsincludenaturalgasdistribution.

0.26 0.44

HazardousLiquid =1ifoperationsincludehazardousliquidtransmission.

0.44 0.50

25

Table3:OLSResultsforVariousMeasuresofEnvironmentalPerformance Incidents Fatalities InjuriesFederalInspections06‐08 0.33

(0.34)0.04(0.06)

0.43(0.32)

FederalCasesInitiated06‐08 0.17*(0.09)

‐0.02(0.01)

‐0.18**(0.08)

FederalProposedPenalties06‐08 1.32**(0.64)

0.02(0.11)

0.70(0.58)

StateInspections06‐08 ‐0.68**(0.34)

‐0.05(0.05)

0.003(0.31)

StateActionsTaken06‐08 0.01(0.02)

0.02**(0.002)

0.15**(0.01)

StatePenaltiesAssessed06‐08 ‐0.73(0.59)

‐0.12(0.10)

‐0.65(0.54)

DependentVariable†06‐08 0.47**(0.02)

0.13(0.13)

0.58**(0.30)

Miles 0.75**(0.30)

‐0.08*(0.05)

‐0.63**(0.27)

MilesSquared ‐0.02**(0.01)

‐0.002(0.002)

‐0.01(0.01)

Intrastate 0.37(0.36)

‐0.01(0.06)

‐0.15(0.32)

NumberofStates 0.31**(0.10)

0.02(0.02)

0.04(0.09)

Region1 0.20(0.58)

0.04(0.09)

‐0.23(0.53)

Region2 ‐0.51(0.41)

0.03(0.06)

0.22(0.37)

Region3 ‐0.52(0.42)

‐0.10(0.07)

‐0.58(0.38)

Region4 ‐1.12**(0.42)

‐0.002(0.07)

0.21(0.39)

GasGathering 0.04(0.36)

0.13**(0.06)

0.82**(0.33)

GasTransmission ‐0.71*(0.40)

‐0.08(0.07)

‐0.31(0.37)

GasDistribution 0.56(0.40)

0.07(0.06)

0.43(0.36)

HazardousLiquid 0.51(0.39)

‐0.07(0.06)

‐0.39(0.36)

Constant ‐0.18(0.61)

0.04(0.10)

0.26(0.55)

R‐squared 0.84 0.27 0.33Sig.atthe5%level;*Sig.atthe10%level.;†Equaltothedep.var.fortheperiod2006‐2008.

26

Table3,Continued

PropertyDamage

GrossBarrelsSpilled

NetBarrelsLost

FederalInspections06‐08

0.08*(0.04)

0.55(0.45)

0.42*(0.25)

FederalCasesInitiated06‐08 ‐2.04*(1.06)

‐0.39**(0.12)

‐0.51**(0.07)

FederalProposedPenalties06‐08 77.80**(7.76)

3.93**(0.82)

1.07**(0.46)

StateInspections06‐08 2.55(4.22)

0.41(0.45)

0.66**(0.25)

StateActionsTaken06‐08

1.37**(0.20)

0.004(0.02)

‐0.01(0.01)

StatePenaltiesAssessed06‐08 ‐14.17*(7.25)

‐0.33(0.76)

0.38(0.42)

DependentVariable†06‐08 0.22(0.25)

0.62**(0.04)

1.11**(0.03)

Miles ‐7.73*(3.69)

‐0.19(0.39)

‐0.08(0.21)

MilesSquared ‐0.19(0.13)

‐0.01(0.01)

‐0.02**(0.01)

Intrastate ‐3.88(4.40)

‐0.60(0.46)

‐0.31(0.26)

NumberofStates 0.75(1.22)

‐0.05(0.12)

‐0.12*(0.07)

Region1 6.71(7.14)

‐0.35(0.76)

‐0.56(0.42)

Region2 0.02(4.94)

‐0.59(0.52)

0.12(0.29)

Region3 ‐13.02**(5.11)

‐0.65(0.54)

‐0.08(0.30)

Region4 ‐8.49*(5.14)

‐0.60(0.55)

0.11(0.30)

GasGathering 8.58*(4.45)

‐0.13(0.47)

‐0.04(0.26)

GasTransmission ‐12.00**(4.98)

0.52(0.53)

0.31(0.29)

GasDistribution 3.05(4.85)

‐0.56(0.51)

‐0.76**(0.28)

HazardousLiquid ‐5.46(4.78)

0.51(0.51)

0.16(0.28)

Constant 17.31**(7.31)

0.97(0.77)

0.58(0.42)

R‐squared 0.36 0.46 0.80**Sig.atthe5%level;*Sig.atthe10%level.;†Equaltothedep.var.fortheperiod2006‐2008.

27

Table4:OLSResultsforVariousMeasuresofFederalEnforcement,2009‐2010

Federal

InspectionsFederalCasesInitiated

FederalProposedPenalties

Incidents06‐08 0.25(0.22)

0.003(0.008)

0.002**(0.001)

Fatalities06‐08 43.14**(8.20)

1.02*(0.31)

0.08*(0.05)

Injuries06‐08 ‐13.31**(3.50)

‐0.32**(0.13)

‐0.02(0.02)

PropertyDamage06‐08 0.10(0.21)

‐0.01(0.01)

0.008**(0.001)

BarrelsSpilled06‐08 0.05(0.47)

‐0.01(0.02)

0.001(0.002)

BarrelsLost06‐08 ‐0.07(0.65)

0.06**(0.02)

‐0.001(0.003)

DependentVariable†06‐08 0.18**(0.03)

0.21**(0.03)

‐0.06*(0.03)

StateInspections06‐08 0.02(0.03)

0.006**(0.001)

0.00002**(0.0001)

StateActionsTaken06‐08

‐0.26*(0.16)

‐0.12**(0.01)

‐0.0013*(0.0007)

StatePenaltiesAssessed06‐08 3.78(5.52)

0.37*(0.21)

0.02(0.02)

Miles 6.03**(2.84)

‐0.15(0.11)

0.016(0.013)

MilesSquared ‐0.18*(0.11)

‐0.0003(0.004)

‐0.0011**(0.0005)

Intrastate ‐3.28(3.37)

0.27**(0.13)

0.01(0.01)

NumberofStates 0.62(0.94)

0.08**(0.03)

0.004(0.005)

Region1 1.56(5.42)

0.44**(0.21)

0.10**(0.03)

Region2 ‐2.42(3.79)

0.28*(0.14)

0.01(0.02)

Region3 ‐1.90(3.91)

0.08(0.15)

0.02(0.02)

Region4 ‐0.04(3.92)

0.34**(0.15)

0.03*(0.02)

**Signif.atthe5%level;*Signif.atthe10%level;†Equaltovariableatthetopofthecolumnfor2006‐08

28

Table4,Con’t

Federal

InspectionsFederalCasesInitiated

FederalProposedPenalties

GasGathering ‐2.18(3.36)

0.03(0.13)

0.003(0.016)

GasTransmission ‐0.43(3.85)

0.05(0.15)

0.026(0.018)

GasDistribution ‐2.73(3.70)

‐0.09(0.14)

0.005(0.018)

HazardousLiquid 0.66(3.66)

0.17(0.14)

0.026(0.018)

Constant 3.43(5.67)

‐0.52**(0.21)

‐0.84**(0.27)

R‐squared 0.59 0.63 0.43**Signif.atthe5%level;*Signif.atthe10%level;†Equaltovariableatthetopofthecolumnfor2006‐08.

29

AppendixA:SensitivityAnalysisforSelf‐ReportedPerformanceData

Ifnon‐performanceissystematicallyunder‐reportedbyoperatorsthatarenot

subjecttofederalenforcement,theresultsinTable3wouldbebiasedupward.To

investigatewhetherunder‐reportingcouldprovideanexplanationforthepositiveand/or

insignificantcoefficientsonthefederalenforcementvariablesintheregressionspresented

inTable3,Iconductedthefollowingexperimenttoseehowbadlyunder‐reportedthe

performancedatawouldhavetobetoestimatenegativeandsignificantcoefficientsforthe

federalenforcementmeasures.

Theexperimentisbasedontheconjecturethatoperatorsaccuratelyself‐reportif

theyareorhaverecentlybeeninspectedbutmaychoosetounder‐reportiftheyarenot

inspectedregularly.SincetheperformancedatausedintheTable3regressionscoverthe

2009‐2010period,Iassumethatoperatorsthataresubjecttofederalinspectionsduring

2009and2010areaccuratelyreportingtheirperformancebutthatoperatorsthatarenot

subjecttofederalinspectionsduringthatperiodmaybeunder‐reporting.Thisassumption

isconsistentwiththemeanvaluesoftheperformancevariablesforthetwogroups.As

showninTableA1themeanvalueofalloftheperformancemeasuresexceptInjuries09‐10is

lessforthenon‐inspectedgroupthanfortheinspectedgroup,andallofthedifferencesare

statisticallysignificantatthe5%level.

ForthoseoperatorsthatwerenotinspectedbytheOPSduring2009‐2010,I

constructed“adjusted”performancevariableswheretheadjustedvariableisequaltothe

self‐reportedperformancevariableplusonestandarddeviation(calculatedovertheentire

sampleof344operators).ForoperatorsthatwereinspectedbyOPS,theadjustedvariable

isequaltotheself‐reportedperformancevariable.IthenrantheTable3regressionsusing

30

theadjustedperformancemeasures.TableA2showsthecoefficientsandstandarderrors

forthethreefederalenforcementvariablesfortheseadjustedregressions.Notethatthese

resultsdonotchangetheoverallconclusionsfromTable3.Asshownbytheinsignificant

coefficientonFederalInspections06‐08inallofthesixoftheregressions,thereisnoevidence

thatfederalinspectionsimproveoperatorperformance,evenwiththeseadjustmentsfor

potentialunder‐reporting.Withtheadjustments,thecoefficientonFederalCases

Initiated06‐08doesbecomesignificantintheFatalitiesregression,whichisconsistentwith

thestudy’sfindingthatinitiatingfederalcasescanincreaseenvironmentalperformance.

WithrespecttoFederalPenaltiesProposed06‐08,theonlyqualitativedifferenceforthe

adjustedregressionsisthatthecoefficientisnolongersignificantintheIncidents

regression.

EvenifIadjusttheperformancevariablesfortheoperatorsthatarenotinspected

bytheOPSduring2009‐2010byadding4timesthestandarddeviationtotheinitiallevelof

performance,Icannotoverturntheinspectionresult–thecoefficientsonFederal

Inspections06‐08remaininsignificantinallsixregressions.However,forthisextreme

adjustmentthecoefficientonFederalCasesInitiated06‐08doesbecomenegativeand

significantintheInjuriesregressionaswell.

31

TableA1:2009‐2010PerformanceMeasures

PerformanceMeasure

AllOperators(N=344)

OperatorsInspectedin2009‐2010(N=147)

OperatorsNotInspectedin2009‐2010(N=197)

Mean Std.Dev. Mean Std.Dev. Mean Std.Dev.Incidents09‐10 2.51 6.33 5.06 8.81 0.60 1.91Fatalities09‐10 0.04 0.42 0.05 0.30 0.04 0.57Injuries09‐10 0.23 2.79 0.16 0.76 0.27 3.63PropertyDamage09‐10 3.63 38.29 5.78 49.66 2.03 26.88GrossBarrelsSpilled09‐10 0.65 4.44 1.41 6.70 0.08 0.56NetBarrelsLost09‐10 0.45 4.04 0.96 6.12 0.07 0.56

TableA2:OLSRegressionResultsfortheFederalEnforcementVariableswhenAdjusted2009‐2010PerformanceMeasuresaretheDependentVariables

AdjustedPerformanceMeasure

FederalInspections06‐08

FederalCasesInitiated06‐08

FederalProposedPenalties06‐08

Incidents09‐10 ‐0.03(0.48)

‐0.18(0.12)

1.30(0.87)

Fatalities09‐10 0.02(0.06)

‐0.04**(0.02)

0.05(0.12)

Injuries09‐10 0.29(0.34)

‐0.32**(0.09)

0.72(0.62)

PropertyDamage09‐10 5.64(4.67)

‐4.02**(1.16)

78.20**(8.52)

GrossBarrelsSpilled09‐10 0.35(0.47)

‐0.65**(0.13)

4.01**(0.86)

NetBarrelsLost09‐10 0.21(0.30)

‐0.72**(0.08)

1.08**(0.05)

32

AppendixB:PartialIdentificationofCausalEffectsforFederalEnforcementVariables

CharlesManski[20]isoneoftheprimarycontributorstotherecentliteratureon

partialidentification.Thepartialidentificationapproachfocusesonestablishingplausible

valuesfortreatmenteffectparametersthatareconsistentwiththeobserveddataunder

relativelyweakassumptions.Usingtheseweakassumptions,researcherscandevelop

boundsforcausaleffectsratherthanpointestimates.InthisappendixIuseManski’s

approachtodevelopplausibleboundsforthecausaleffectsoffederalenforcementon

pipelineoperatorperformanceunderminimalassumptions.

BecausethebasicpartialidentificationapproachdevelopedbyManski[21]focuses

onbinarytreatments,Icreatedthreebinarytreatmentvariables,FederallyInspected06‐08,

FederalCaseInitiated06‐08,andFederalProposedPenalty06‐08.Theobjectiveofpartial

identificationisthentodevelopboundsfortheeffectsofeachofthesethreetreatmentson

thesixperformancevariables:MeanIncidents09‐10,MeanFatalities09‐10,MeanInjuries09‐10,

MeanPropertyDamage09‐10,MeanGrossBarrelsSpilled09‐10andMeanNetBarrelsLost09‐10.

LetTi=1ifoperatoriistreatedand=0ifoperatoriisuntreated.LetY1ibethe

performanceofoperatoriwhentreatedandY0ibetheperformancewhenuntreated.Then

theindividualimpactofthetreatmentonoperatoriis∆i=Y1i‐Y0i.Ofcourse,wecannot

observebothY1iandY0iastheoperatorcannotbebothtreatedanduntreated.Thusto

estimatethecausaleffectofthetreatment,weestimatetheaveragetreatmenteffectwhich

isequaltothedifferencebetweentheexpectedvaluesofY1andY0inthepopulation(i.e.,

ATE=E[Y1]‐E[Y0])usingobservableinformationfromdifferentoperators.Iftheexpected

averagetreatmenteffectdiffersacrossthoseoperatorsthataretreatedandthosethatare

untreated,thentheATEistheweightedaverageoftheaveragetreatmenteffectonthe

33

treatedandtheaveragetreatmenteffectontheuntreated.Morespecifically,letw1equal

theproportionofoperatorsthataretreated.Then

ATE=w1*E[Y1‐Y0|Ti=1]+(1‐w1)*E[Y1‐Y0|Ti=0].

Asafirststep,inestimatingtheATE,Iassumethatthepotentialperformancerates

forbothtreatedanduntreatedoperatorsliewithinthesupportobservedinthedataon

performancefrom2006to2010.Underthisassumptionthemaximumpossiblenegative

treatmenteffect(i.e.,improvementinperformance)wouldoccurwhenanoperator’s

untreatedperformancewouldbeequaltothemaximumvalueandthatsameoperator’s

treatedperformancewouldbeequalto0.6Themaximumpossiblepositivetreatmenteffect

wouldoccurwhenanoperator’suntreatedperformancewouldbeequalto0andthatsame

operator’streatedperformancewouldbeequaltothemaximumvalue.LetMbethe

maximumobservedperformancelevel.Then

0‐M≤E[Y1]‐E[Y0])≤M‐0.

AsshowninTableB1,usingthehistoricalmaximumvaluesforeachofthesixperformance

variablesprovidesinitialboundsonthepossibletreatmenteffect,althoughbyconstruction

theseboundsarerelativelylargeaswellassymmetric.

Thenextstepinthepartialidentificationapproachistofurtherrefinethebounds

usingtherelativelyweakassumptionthatthemeanobservedperformancedataforthe

treatedgroup(D1)givesusanunbiasedestimatoroftheexpectedperformanceofthe

treatedgroupwhentreated(E[Y1|Ti=1])andthatthemeanobservedperformancedata

fortheuntreatedgroup(D0)givesusanunbiasedestimatoroftheexpectedperformance

fortheuntreatedgroupwhenuntreated(E[Y0|Ti=0]).RecallingthattheATEisaweighted6Thenatureoftheperformancevariablesruleoutnegativevalues,so0isthesmallestpossiblevalue.

34

averageoftheexpectedcausaleffectsforthetreatedanduntreatedgroups,thebounds

become:

w1*(D1‐M)+(1‐w1)*(0–D0)≤E[Y1]‐E[Y0])≤w1*(D1–0)+(1‐w1)*(M–D0).

AsshowninTableB1,thisrelativelyweakassumptionsubstantiallyshrinkstheboundsof

theATE.7

Giventheconcernthattreatmentselectionmaybeendogenous,wecanfurther

shrinktheboundsbymakinganotherweakassumption:thattheaverageperformancefor

treatedfirmsisweaklyhigherthantheaverageperformanceforuntreatedfirmsbothwith

andwithouttreatment,or

E[Y1|Ti=1]≥E[Y1|Ti=0]andE[Y0|Ti=1]≥E[Y0|Ti=0].8

Underthisassumptiontheupperboundbecomes

w1*(D1–D0)+(1‐w1)*(D1–D0)=D1–D0,

whilethelowerboundisunchanged.Thusifoneassumespositiveselection,asshownin

TableB1theboundsontheATEforMeanFatalities09‐10andMeanInjuries09‐10ruleoutany

quantitativelysignificantpositiveeffectswhiletheboundsontheATEfortheremaining

performancevariablesdoallowforquantitativelysignificantpositiveeffects.Comparing

theseresultstothesignandsignificanceofthecoefficientsinTable3,partialidentification

suggeststhatallofthepositivecoefficientsonthefederalenforcementvariablesinthe

Incidents,PropertyDamage,GrossBarrelsSpilled,andNetBarrelsLostregressionsare

plausibleevenifoneassumesthatregulatorsdotargetoperatorswithhigherlevelsofnon‐

7Manskicallsthisthe“NoAssumptions”bound.8Manskitermsthisthe“MonotoneTreatmentSelection”assumption.

35

performanceforenforcementactions.9Howeversignificantpositivecoefficientsonthe

federalenforcementvariablesfortheFatalitiesandInjuriesregressionswouldnotbe

consistentwiththatassumption.Interestingly,Idonotfindanypositiveandsignificant

coefficientsforthefederalenforcementvariablesineithertheFatalitiesortheInjuries

regressions.

9BecausetheregressionspresentedinTable3usecontinuousratherthanbinaryfederalenforcementvariables,thesizeofthecoefficientscannotbedirectlycomparedtothebounds.

36

TableB1:PartialIdentificationBoundsonAverageTreatmentEffects

Mean

Incidents09‐10Mean

Fatalities09‐10Mean

Injuries09‐10MeanPropertyDamage09‐10

MeanGrossBarrels

Spilled09‐10

MeanNetBarrelsLost09‐10

HistoricalMaximum [‐28,28] [‐8,8] [‐51,51] [‐601,601] [‐70.19,70.19] [‐70.19,70.19]Inspected06‐08ObservedData [‐12.16,15.84] [‐3.75,4.25] [‐23.92,27.08] [‐281,320] [‐32.59,37.60] [‐32.67,37.52]PositiveSelection [‐12.16,2.06] [‐3.75,0.00] [‐23.92‐0.08] [‐281,1.47] [‐32.59,0.57] [‐32.67,0.38]CaseInitiated06‐08ObservedData [‐9.75,18.25] [‐3,5] [‐19.17,31.83] [‐225,376] [‐26.06,44.13] [‐26.15,44.05]PositiveSelection [‐9.75,2.26] [‐3,0.01] [‐19.17,‐0.04] [‐225,2.25] [‐26.06,0.73] [‐26.15,0.50]PenaltyProposed06‐08ObservedData [‐3.96,24.04] [‐1.17,6.83] [‐7.47,43.53] [‐87,514] [‐10.29,59.90] [‐10.32,59.87]PositiveSelection [‐3.96,4.02] [‐1.17,0.03] [‐7.47,0.08] [‐87,7.02] [‐10.29,0.58] [‐10.32,0.18]

37

References

[1]Terkel,Amada.“112thCongressSetToBecomeMostUnproductiveSince1940s,”TheHuffingtonPost,December28,2012.

[2]Frosch,Dan.“CongressPassesBillBolsteringPipelineSafetyRules,”TheNewYork

Times,December17,2011.[3]May,PeterJ.1991.“ReconsideringPolicyDesign:PoliciesandPublics,”JournalofPublic

PolicyVol.11,No.2,pp.187‐206.

[4]Parker,CarolM.2004.“ThePipelineIndustryMeetsGriefUnimaginable:CongressReactswiththePipelineSafetyImprovementActof2002”NaturalResourcesJournal44:243‐282.

[5]Frosch,DanandJanetRoberts.“PipelineSpillsPutSafeguardsUnderScrutiny,”,The

NewYorkTimes,September9,2011.[6]Quarterman,CynthiaL.(Administrator,PipelineandHazardousMaterialsSafety

Administration,U.S.DepartmentofTransportation).TestimonyofbeforetheHouseEnergyandCommerceCommittee,June16,2011.

[7]E.SosaandJ.Alvarez‐Ramirez.2009.“Time‐CorrelationsintheDynamicsofHazardous

MaterialsPipelineIncidents,”JournalofHazardousMaterials165:1204‐1209.

[8]Becker,GaryS.1968.“CrimeandPunishment:AnEconomicApproach.”JournalofPoliticalEconomy,76:169‐172.

[9]Russell,CliffordS.,WinstonHarringtonandWilliamJ.Vaughan.1986.EconomicModels

ofMonitoringandEnforcement:EnforcingPollutionControlLaws,Washington,D.C:ResourcesfortheFuture.

[10]Gray,WayneB.andJayP.Shimshack.2011.“TheEffectivenessofEnvironmental

MonitoringandEnforcement:AReviewoftheEmpiricalEvidence,”ReviewofEnvironmentalEconomicsandPolicy5:3–24.

[11]Gray,WayneB.andMaryDeily.1996.“ComplianceandEnforcement:AirPollution

RegulationintheU.S.SteelIndustry,”JournalofEnvironmentalEconomicsandManagement31:96–111.

[12]Gray,WayneandRonShadbegian.2005.“WhenandWhyDoPlantsComply?Paper

Millsinthe1980s,”LawandPolicy27:238–61.[13]Earnhart,Dietrich.2004.“RegulatoryFactorsShapingEnvironmentalPerformanceat

PubliclyOwnedTreatmentPlants,”JournalofEnvironmentalEconomicsandManagement48:655–81.

38

[14]Glicksman,R.,andDietrichEarnhart.2007.“TheComparativeEffectivenessof

GovernmentInterventionsonEnvironmentalPerformanceintheChemicalIndustry,”StanfordEnvironmentalLawJournal26:317–71.

[15]Stafford,Sarah.2002.“TheEffectofPunishmentonFirmCompliancewithHazardous

WasteRegulations,”JournalofEnvironmentalEconomicsandManagement44:290–308.

[16]Gray,WayneB.,andJohnT.Scholz.1993.“DoesRegulatoryEnforcementWork?A

LongitudinalStudyofOSHAEnforcement,”LawandSocietyReview27:177–213.[17]Gray,WayneB.andJohnMendeloff.2005.“TheDecliningEffectsofOSHAInspections

onManufacturingInjuries,1979‐1998,”IndustrialandLaborRelationsReview58:571‐587.

[18]Haviland,Amelaietal.2010.“WhatkindsofinjuriesdoOSHAinspectionsprevent?”

JournalofSafetyResearch41:339‐345.[19]Khanna,MadhuandWilliamRoseQ.Anton.2002.“CorporateEnvironmental

Management:RegulatoryandMarket‐BasedIncentives,”LandEconomics78:539‐558.

[20]Thornton,Dorothy,NeilGunningham,RobertKagan.2005."GeneralDeterrenceand

CorporateEnvironmentalBehavior,"Law&Policy27:262‐288.[21]Manski,CharlesF.2007.IdentificationforPredictionandDecision.Cambridge,MA:

HarvardUniversityPress.[22]Morgan,StephenL.andChristopherWinship.2007.CounterfactualsandCausal

Inference:MethodsandPrinciplesforSocialResearch.Cambridge,UK:CambridgeUniversityPress.