You are on page 1of 13

Bus Stop-Environmental Connection: Do Characteristics of the Built Environment Correlate with Bus Stop Crime?

Robin Liggett Anastasia Loukaltou-Slderis H1royukiIsekl

UCTC No 613

The University Transporta[ion

of California Center

Th~ Umverszty C.~lifcrma of Transportation Center(UCTC) is oneoft~nregional urnrs mandated Congress by and establishedFall1988to in support research, education, andtraining surface in transportataon. The UC Center servesfederal R~on IX mad is supported matching by granU fromthe U.S.Departmeatof Transportataon, the Ca/dfornm Department of Transpormuon (Caltrans), and the Ur6versity Based on the Berkeley Campus, UCTCdraws upon erdsdng eapabllitzes and resources the Institutes of of Transponataon Srcdmsat Berkedey, Davis, Irvme, and Los Angeles, the Instzmte of Urbean and Rc=~onal Development a: Berkeley, and several academic departments at the Berkeley, Davls, irvine, and Los .Angeles campuses FaeulW. and students on other UmvemWof Cahfornia campuses participate may m

Ccnmr activiues Researchers nt other universities the wi~in re,nonalso have oppormmdes to collaborate with UC f~culry on seiecmd smdaes. UCTCseduc~onaland research programs focused are on strategic planning for :reproving metropolitan acce~s~biL{ty, with emphasis on the specSal con&Uons m ReBaon P~t~cul~r IX a~enfion is directed to strateDes for udngtransportation as an msuument economic of development, a!soacwhile commodatJng to the regmns perszstent expansm~ and while mmntmmng and enhmacmgthe quality of life there The Center d, st~butes reports on im research m working papers, monogTaphs, and m reprin~ of pu~hshedz.rucles. It also pubhshes Acce.cs, a magazine presennng summanes of selected studies For a listof pubheafons pnnt, in write to me a~aress below

Umvernty CaRternis of Transportation Center l O~Nava/Axchiteemre Building Bef~celey,Califorma94720 Tel 510t643-7378 FAX:510/6435~56

D|SCLAIMER The contents report ofthis reflect viewsthe the of authom, wh~ are responsible fa~ theaccuracy information forthe and ofthe presented herein. document This is disseminated sponsomhip under the of the Department ofTransportation, TransportationProgram, University Centers in the interest information The Government no of exchange. U.S. assumes Liability thecontents thereof. for oruse

Thecontentsof this rcpor~r~flect the vzews the authorwho responsible of = for the fa~ and aceur~,,cy of the daUt presented h~mm costents,do The not ~e.ecssanlyreflect the offiCm~ or pohmes th~ Sta~of Callforni~or the mews of U.S Deparunen~ ofTranspommor~ ~pon does not consdtnt~ a standard, Th:s spccificaUoa, regulation. or

Bus Stop-EnvironmentalConnection: Do Characteristics of the Built EnvironmentCorrelate with Bus Stop Crime?

Robin Liggett Anastasia Loukaitou-Sideris Hiroyuki Iseki

Replanted from Transportatton Research Record 1760 Paper No 01-0441, pp 20-27 (2001)

UCTCNo. 613 The University of California Transportation Center University of California at Berkeley

20

Jl Trensportatlon Research Record1760 Paper No 01-0441

Bus Stop-Environment

Connection

Do Characteristics of the Built Environment Correlate with Bus Stop Crime?


Robin Liggett, Anestasia Loukaitou-Sideris, Hiroyukitseki and
Can urtderstand somebus stops are safe andothers are crimewe why ridden? wepredict wluch C.m features of the bus stop enwronment are hkelyto eucourage discourage or crime? wedesign safer bus stops? Can Thesequestionsare addressed exploringthe relationship between by environmental variables and bus stop crime. Anearlier study used crime datv, alongwithenvironmental inchcators,for a sample 60 bus of stops in downtown Angeles.Crime Los rates werehigherfor bus stops nearalley~, multifsmtly housing,Hquor stores andcheck-cashing estab~ lishments, vacant buildings, graffiti andlitter. In contrast,good and vie. lbliity of t~e bus stop fromits surroundings the existence of bus and shelters contributed lowercrime to rates. Thisearlier studywasindicative butnot predictive the elements contribute busstopcrime. of that to Wtththe geographicand temporal expansionof the dam (covering larger city part overa longertimespan),a series of regression models wasgenerated identify environmental that predictors busstop crime. of Thesemodels showthat the mostimportant predictorof crimeIs location. If the envtronmentcontrolled,undesirable ~s facilliies andlitter result in h~gher crimerates, whereas v~ibitity and many pedestrians lead to lower crimerates. Thepresenceor absence certain charac. of teristics mthe bus stop mlcroenvironment affect crnne. Also~the can approprialte design andlayoutof the physicalenvironment reduce can opportunities crlmLnal for actions. pie size, a series of regesslon modelswere generated that Identify environmental predictors of bus stop crime Because of the spaUal nature of the problem, issues of spaua! autocorrelaUon have been explored and accounted fox in building the regression models The expanded sample and scope of the research enabled some of tbe rnethodological flaws of the prewousstudy to be corrected Given the results of tins expandedstudy, drawingconclusions about whleh environmental factors create opportumties for or hinder bus stop crime can be done vc~th more confidence

THEORETICAL FRAMEWORKOF THE RESEARCH

Increasingly, cnminolog~.shave becomeaware of the importance of places of crLme A place is a vmy (2) small area, a si~eet comer,a bus stop, or an mtersectiono Researchers have noUced that crime often tends to concentrate heavily and chsproporUonately a few places, or m "hot spots" (3-5) Suchobservationshave led to argumentsfor a reonentaUonof cnmeprevention efforts and a focus on the environmental context of crime instead of the soc~o-demograph~c charactensUcsof the offenders (6) Th~s debate underlines two distract approaches mcrime research stu&es (see Table 1) So-called composmonal nonecolog~calstudor ~es stress the ~mport~ce the offenders soc~o-demograptnc of charCan we makebus stop enwronmentssafer for passengers by givmg actenst:cs Therefore, these studies seek to idenufy relauonsh~ps guldance pohcy io makers thelocation deslgn busstops~ on and of betweennexghborhood level thecharactensUcs a s crime and ofrace Following work crnnmologzsts, the of environmental psychologists, and ethnicuty,and age gender, poverty levels, social and mobi.hty of and planners, argued the ItIs that effectsthe of built environment on mhabztants Somestuchcshaveaiso ~ttemptedto correlate crLmewith busstop crlrne bemeasured used prechct busstop can and to wluch measures of family &eruption(e g, percent of chvorcedhouseholds locatmns to inwte higher tend a propomon criminal This of acts or female-headedfamflzes w~children) analysis, m turn, lead generalrecommendattons and can to onland use In contrast, ecological stuches focus on physical atU~butes as other btult envLronment features would that promote bus safer stops covanatesof crmae(7) For such stuches, zt ~s the location and physzo This analysls expands a prewous on study that attemptedmeato cal context of crime--not the socio-demo~aph~c characteristics of sure effect certain the of enwroranental onbus variables stop crmle (I) the offenders--that acqmre s~gn~cance Of pamcutar interest are Theimttt2 study used exlsttug andndersbap along crime data, wlth place charactensttcs (land use, built-form condinon,vzs~b~tylevels), environmental landusemchcators, weredoctunented and that as weR a s~tes access characteristics as through extensive fieldwork 60 intersectlonsbusstops for w~th m Clearly, the two approaches of crtme research lead to different downtown Angeles tentative Los The concluslons first inthe study types of crime preventaon strategies Composlt~onal studies target were mostly based t-tests on comparing average stop bus crime rates the potenual offenders They advocate socral and educational serbetween mtersectlons categorizedtheexistence particular by of vices to tackle teenage delinquencyand recid:v~sm. Theyargue for environmental characteristics (eg,presenceabsence analley) or of changes m the system of pohcmg (e g, commumty pohcing) and ThemethDdological approach hn~ted was because therather of reforrnulaUons of the cnnunaljusUceand penal systems to address small sample In th~newphase research, study s~ of the area and crime thetime frame have been expanded theexpauslon thesamWith of ImpIicit mthe ecological studies ~s the behef that the redesign or transformationof certain pIace characteristics can lead to lower Department PIanmng, ofPubhc ofUrban Schoo[ Pohcy Soc=el end Research, Umvers~ty clCehfomle-Los 3250 Angeles, Public BuildEng, Pchcy Box 951656, levels of crime Theseefforts are called "s~tuat~onal" because they Los Angeles, CA900S5 hnkcnrmnal acuwttes thespecific to physical attributeshot of

IJggett et al TABLEI CrimeStudies Comnosiflona! Importance ofphysmaland cologie~1 atmbutes Studyof the n~ronmental con~gt Importanceofsoclodemogmpbae attributes (age, cthniezty,gender, class, social moblhty) Studyof offenders

PaperNo

0%0447 2t

CrimePrevention Strate~tes Target enwronmeataI context ("demgrang came") out Situational erlms prevemion Crime Prm/en~onThrough EnvtronmentaI Demgn (CPTED) Targetpotsatta! offenders Socml/educattonal sorvtces Pohcmg

Criminal ]usac~

s~ts Ecological studies lead to crmae prevention efforts that use envaronment2d demgnas a tool for "designing out" crime. Environmental dedgnis used to reduce the attributes that are behoved tc enhance crime increase the "defenmble space" dementsthat and mebehovedto block opportunitaes for crmae (8, 9). Before such design efforts and preventmn pohcmsare amplemeated, however, tte dLfferent physical attributes that can encourageor discourage crime must be clearly understood TILtsstudy falls mthe ecologicalcategoryand intendedto idenis tify and ob3ectavdy measurethe spatial parameters of ermae The selecuon of the ecological approach has more to do with gained kaowledge than with the belief that ecological studies are inherently b~tter than compomttonal studies These two perspectives are not vtewed antlthetic.al, becausethey can both contribute to the better as understanding of enmeand its prevention

ELLIS STOP CRIME |N LOS ANGELES

Ymm prevlous studies, bus stop crane m Los Angeles was found to be pRmanly concentratedin certain hot spots (/, 5) Theseareas tend to be mostly mdowntown its outlying areas. Some spots arc and hot s ?aUMly cluslered, while others are isolated In the previousresearch, d~ewnportancetheenvironmental onbusstop of setting came was esmbhshcd stops proranmy, thesame routes, Bus m along bus and ~resumably passengers with having samesoclo-demograpbac the characteristics were markedby different crime rates (10) Ewdence wasalso found that certain urban form madbus stop charaetenstics ENVIRONMENTAL INVENTORY DATA ~ave an effect on cameCame rates were bagher at bus stops near alleys, multtarmly housing, undesirable estabhshments such as liquor As in the earlier study, detmled pnmary data of enwronmental stores and cheek oasl~ng estabhshrnents, and vacant buildings, and indicators were collected at each of the mtersecttons added to wheregraf~t~ and httsr we.re present. Posmve environmental factors the study [see Loukmtou-Sidens al (1) for details on the data et (that often traustated rote lower camerotes) included goodvmb~ty collecbon process] Data were collected for three groups of charof the bus stcp from surroundingestabhshmentsand the existence of actenstms" (a) urban form charactenstxcs around intersecUons, bus shdters. wluch included mformatmn the land use and eondmonof the on Prevmusstudies have been mdaeatave but not prechetave of the surroundingarea; (b) bus stop eharactensUes,such as the exxstenee physical dements that contribute to bus stop enme(1, 5, 11) The goal of bus shelters, wslbflity, andhghtmg; (c) street characteristics, and of tim study wasthe ability to preclxct, watha mgmficant level of eersuch as street and sidewalk wxdth, on-street parting, and traffic t~uaty, the phymcal elementsat bus stops that wouldaffect bus stop levds

cameMeasunng effect of each attribute wasalso an objeetlve and the reztmred a mgmficant xpanmon the extstmg dare, of The data were expandedgeographaeally(covered a larger part of the city) and temporally (crime data for a larger tmae span) expanded sample included 100 mtersccUonswith bus stops m both downtown Angeles and in the adjoining neighborhoods of Pico Los Umon and Wesflake. These areas were selected because of their disproportionate eoncentr~Uonof bus stop came me, dents m Los Angeles (12). The map &splayed m~gure 1 shows came and nder* ship levels for the sdected bus stops The ingh-enmebus stops are concentrated along oertmn maul streets m the downtown area, m what is considered the old historic core and along one of the ma2or arteries through the Pieo Umon Wesflakeareas and Crime data collected by the Los Angeles Metropohtan Transo portalaon Authority (IWIA) Tranmt Pohee allreported inciof crime dents busstops at from1996 through were 1998 merged with bus stopcamedataeollected theLosAngeles by Pohce Department from 1994 1998. to Because crane datawere normahzcd busndby ershtp to ex~mme crimes per rider, data were also obtained from the MTA average numbers of passenger boardmgs and ahghtmgs for per bus stop. Theumt of analysis for whachdata was eotieeted and analyzed wasthe intersection wherebus stops were located All bus stop crime and ridembap data for bus stops ma 46-m (150 f-t) radius of an mterseetaon(about one-tl~d of a city block) were aggregated form a single observation mthe data set Tias was accomptmhed using Environmental Systems Research In~tutes ArcVlewto geocode data to the closest ~tersectton The database conmatsdof 2,805 bus stop crimes (crmaes against people who were wmtmg a bus or whohad just comeoff a bus) for rec~rcled at the 100 study intersections Crimeswere eategormed rote one of two types, ranging fromserious crime (Type1-) such as rape, robbery, and assault, to less serious chines against people(TypeH), such pink-pocket .lewelry as and snatching pubhc and nmsanee or pubhc offense aspubhe such drmlang, ordisorderly lewd eonduct, anddrug dealing About three-fourths crime of the mcldents were Type1I crimes (2,228), 577 mcadentswere Type I cranes SLxteenof the mterseeUonsin the sample had more than 50 crones, only 4 of whachwere outside the downtown area Pddersh~p, which is measuredas the average dmiy boardmgsand ahghtmgs an intersectton, rangedfrom s rnmlmum 15g lIders per at of day to a maximum 12,685Eaghty-threepercent of the intersections of had less than 5,000 riders per day, with ordy two mterseeuenshaving more than 10,000 daily riders When crnne per intersection was normahzedby the number of nders about one-half of the ratersections (45) had morethan one crime per I00 riders. Of the top intersections mtotal crmaeper 100 riders, 8 mtersectaons were m the downtown area

~eper No 0%0447

Transportst~onResearch Record1760

DOWNTOWN AREA

FIOURECrimeandrldersh=p in the study area I

Table 2 hats env~onmentalvariables measured and assoc;ated correlations v~th crime counts (numberof crimes) af~d crime rates (crimes pe~ I00 ~de~s)Th~su-npl~ correlation aualys~sled to a number of tentative conclusionsabout the relationship d certain physical at~butes and bus stop cringe These conclusions are summm~ed m Table 3

A 0-1 variable wasused to measure absence or presence of a parthe tacular 5aegatzve"land use (e g. hquorstores, adult movietheaters and bookstores) Restdentm! Single fam~yhousing was present at only four ratersect.tons, however,multffamdyhousing crusted at 30 mtersecttous Contrary to studaes that found res~dentxal crime correlated w~th multffamtly housing, thts correlatmn was not found between such housing and any of the cr~maemeasures (13-t5) This result also contradzets findings ~u the prekm~arystudy that conmdered only bus stops mdowntown Angeluswhere there washttle reszdenfial Los housing (1) CommePcial Three eommerctal land use types were measured small or open-art estabhshments (concessions, ~osks), small closed-front establishments(retml stores), and large or closed-front estabhshraents (e g banks, department stores). Only 20 percent the intersections had sinai1 or open-air commercial properttes, wlMe almost all mterseettons (86 percent) had small or closed-front commercialestabhshments About80 percent ofthe mtersectaons had at least one large or dosed-front commcrcaal use A moderate land correlataon was found between the number of small commercial estabhshments (either open atr or closed front) and cnmecounts. Small commerczal estabhshmentshad a stgrafieant correIation with nderstap (busy intersections tended to have small commercial estabhshments) A weak but szgraficant negauve correlation was found betweencrime counts and crime rates and the numberor percentage

Urban Form Characteristics

AfleyandM~dblock Connections
S~xty-two the 100 intetsectaons had e~ther aa alley or midblock of cormecuon Although a moderate but sigmficaut corretauon was found betweenthe exzstence of an escape route such as an alley or midblockconnection and crime counts, no sigmficant correlatton was found whenlooking at crime rates

Land Use~
Landuse wasmeasuredm the field by counting the number estabof hshmentsm a pamcuJarclassification (e g, numberof single farofly homes,number surface parking lots) in the area defining the of mtersectmnIn ad&taon, as a surrogate measurefor the length of street front dechcated a partacular land use at each intersecuon,the to percentage of lots dedacatedto a pameularla~d use was calculated

IJgge~et al TABLE2 Conelstlon Between Environmental ladmatora and Bus Stop Crime Total CrhneCounts CrimeRates U~rban Form Characteristics heron F=cllltatmg .~cspe Alley/Mzd-bloek Cormectmn LandUse Single-family Restdontial Mu~t~-fcmdy Res~dentml Small/Op~-/Ur Commormal Small/Closed Front Commercml LargclClosod Front Commmm| LiquorStores CheckCashlng Esmbhshmcut Adult Movlc Theatres Adult Stores Book (en~y one) SutfaoeParkingLot Fc~c=d Unfenced W~th Att=ndaat Parhng 8t~otum Conddmn VsuantLots Vaoiat Buildings Run-down F~tabhshments Oraffit~ L~.=r Bus $ton Characteristics V1mb:hty Standard Street Light P~le~e.nStreet hght Otherl.zghts PubhoPhones BusShelters Vmblo Caretaker/Guard PohoeSubstaUon Pedestrmn P~en~ (WmtT~me) Street Chara~teristic~ Street Vehtole Traffic On-slrc~t Parhng

Paper NQ Q10,~41

23

TYP~ [ Crime Coun~ Cr]ume Rates

0 529** -0081 0 003 0 462 ** 0 493** -0 025 0 191 0 198 * ,-0 014 -0 145 -0 048 -0 126 -0 159 0 057 -0 131 0 130 -0 042 0207 ** 0 419 ** 0 148 -0 016 0 293 ** 0 037 0,238* -0 123 0 239 * 0 404** 0 493** -0 070 0 I00

0 113 -0105 0 078 0 041 0 008 -0 12l 0 287 * -0 017 0 165 0 045 0.305* ,-0 206 *0 096 0 032 0 023 0256 * 0249 0 036 0.279 * -0 308 * 0 049 0 021 .4) 078 0 034 .-0.250 -0 034 0 189 -0 252 -02,14 0 156

0.2.48 * 0 071 0 047 0 253* 0263 * -0.294 * 0 163 0 164 0 075 -0 007 ~0 136 0 1II -0159 o0053 0 059 -0 102 ~0 157 0.202 * 0 402* 0.266 * 0 166 0 180 -0 084 0262* 0 043 0 159 0 059 0 466 * 0 750 0 016

-0 091 0 033 0 037 -0 085 .4) 094 -0 235 0 123 -0 046 -0 117 o 105 0 085 0047 4)096 -0 084 0241 0 126 0247 -0 058 0 129 -0 240 0 t36 -0 009 -0 087 0 082 .-0117 =0 080 -0 039 -0/.74 o0026 -0 051

td largeor closed-front conun~mal estabhshments The reason for tlus finding nmy be that most of these estabhshn~nts were banks and ch~partrnent stores, wlmch typloal/y an added haws layer security, of with roving secm~ty guards m the=tmmechate v]cm.tty. Undestrable Establishments The prevmus study mchcatedthat c,ezt~Jn establishments close a busstop increase crnne to can rates

In this study, 39 mtersectmns had at least one hquor store (five intersections had the maximum of three hquor stores) and 17 mtcrscctmns at leastone check-cashing estabhshment A had moderate pOSl~Ve correlation was measured between the emstence of a hquor store andtotal crime rate This relations~p dldnot hold, however, when loohng only at Type I (serious) crune rates. Only slx interseetmns had adult mowetheaters (there was no difference

TABLE Environmsntel Variables Related to Bus Stop Creme 3 Variables Associated with Nigher Crime Rates Total Crime Rates Liquor Stores O~her and Undemrable Estabhshments Vacant Btuldmgs and Lots RundownBuildings Level of L~tter Type I Crime Rates * Vacant Lots Rundown Bmldmgs Variables Associated wsth Lower Crime Rates * Large/Closed Front Commercml , V:mbfllty * Bus Shelters * Street Traffic * Pedestrian Presence mt T~me * * * LwgdCIosed Front Commercml V~s*b~Itty Pedestrian Presence (Wait T~me)

24 PeperNo

01-0447

Trensport~laonResearch Record1760 hshmentat an intersection and crime rates Only three intersections had police substataons and these tended to have h~gher levels of comeHowever,it rmght be expected that station locations were selected because of enme levels and are an effect rather than a cause A measureof pedestrian presence m the vmanttyof bus stops was ealculated using the average time that lapses betweentwo consecutive buses weighted by the numberof hoardings (watt tame) This measure m negatavely correlated with all crime rates, mdlcatmg lower levels of comewherethere are more"eyes on the street" (16)

in avengec~e counts or rates between mtersechonsw~th and without adult movie theaters), and ordy one mtersectaon had an aduh bookstore Sumramg types of undesrrable estabhshments all per intersection, a posltave correlation wasfound between exasthe tenet of an undesirable estabhslunent and come rates (r = 0 322 for total cronerate), Perking Surface parking lots were located at 76 mtersectaons. Lots were classified as fenced and unfenced, assumingthat fenced lots provideda safer environment because cnnnnals could not escape by running through the lots Unfenced purling lots were located at 27 intersecUoas The relauomtup between the numberofparlonglots and crmaerams was the opposite of that expected Total crime per 100 nders was poslUvelycorrelated with the numberof fenced parking Iota and negatively correlated with the number unfencedparkof LugIol:s Asa result, no sagmfioaut correlatmnwasfoundbetween the total surface parking lots (fenced and unfenced) and crime rates Data were also coUectedon whether or not there was an attendant at a purling lot Thiny-oneof the mtersectaons had parking attendants The presence of an attendant and comerotes had no correlation, however,mformauon not cotlected on the hours an attenwas dant was present Therefore, it was not known whether an attendant was present when a crane took place Fifteen mtersecUons had paring structures Crime and the erastence of these structures at an mtersectaon had no relationship Vacant Lots andBuildings Vacant lots cxastedat 18 intersections Morethan one-h~Ifthe of mtersectaons had from to 20vacant I buddrags posltzve A correlauon found was between number vacant the of lots and TypeI crime rates, as well as betweenthe number vacant of buildings and total comerates Summing numberof vacant lots the and buildings per intersection, a correlauon of 0 361 with the total come rate was found Bu,[t-Form Condemn The emsmnce rundown of estabhshments at an intersection and the total ermaerate and Typet crime rate had a weakbut significant eorrelatmn OMy mtersectaons had build13 ings chtssuSed as rundownA s~gmficant positzve correlation was found betweentotal crime count and crime rate and the leveI ofhto ter at art mtersectaon(the level of hirer ~asmeasured a 24-point on seals) Al~ough atonal study found that crane rates were higher the at mtersectaons where graf-fitx were present (also measured on a 24-pcmt scale), tins relataons~p dad not hold m the expanded data set

Street Characteristics Street sldewalk or wldth come and rat~hadno correlaUon However, negators a correlataonfound was between level street the of traffic total and crime Higher rate levels traffic associated of were with lower rates exastenceon-street come The of parlong crime and had norelaUonstnp Table 3 is a prehminary summary, environmental of variables associated with bus stop crime rates Photographsin Figure 2 showenwromnents typically associated vnth l:agh- and low-come stops bus The enviromnentalvariables were seleeted as potential independent variables for a mulUple reg~essaonanalyszs

(a)

Bus Stop Characteristms Vlsiblh~/ was important for bus stop safety AII crime rates tended to be lower at intersectaons where bus stops were vlsible from estabhshments(there was a direct line of sight from ~ nearby estabhshment to a bus stop) No relataonsh~p was found between cnmerates andthe presence of streethghts atbus stops However, the fieldwork was done in the daytime and dlun~naUonlevels at each intersection were unknownA posture relationship was found between the existence of pubhc telephones at bus stops and crmae counts, but not enme rates The police stated that pubhc telephones are often used as hangouts for drug dealers Bus stops at 32 mtersecuonshad shelters Bus shelters and total crune rate had a weak but sIgmficant negative correlataon No correlation wasfound betweenthe existence of a visible caretaker at an estab-

Co) FIGURETypical htgh- endlow-crimebus stops (a) hzgh-orlms 2 busstop, (b) low-crime stop bus

BUILDING A REGRESSION MGDEL A set of muluple regression models was explored to measure the e, ffec~s of enwronmental factors on enmeratesPlots of residuals from the lmtlat regression modelsmchcated probtemsof heteroskedasttelty (tmequalvataance) Tins variance was corrected by a log transformanon of the dependentvariable Thefour best envzronmentrd predictors of the n,~mallog (LN)of total crimesper 100 riders are shown Model1 mTable 4 These predictors include the presence of liquor stores in the near v~canty,the amount hirer at the mtersectaon,~sof zblhty thebusstop of from adjacent estabhsbments, tame andwmt (a measureof pedestrmnpresence). Crmae rates were hgher (posxtlve regression eoefficmnts) at mtersectmns wherektter and hquorstores exasted C~nerates were lower (negative coefficients) whenv~ibiiity and pedestrians were present at the bus stop Theadjusted R-square for this mo&~l weak 249), and coefficients for two independent was ((3 vanables--hquor store and wmttime--were significant only at the 0.08 level Not~agthe hgh concentrations of crime m localmed comdorsin the study area, dummy variables measva~g locaUonwere tested in the regression equation. Only one location vanabte contnbuted sign~cautty to the model-- locatmnm the bastonc core of downtown Los Angeles The adjusted R-square for tlns model (shownas Model2 Table 4) increased to 0 389, wlth sagraficant coefficients on all vanabtes Although the magmtude signs of the coefficaems on the and onganal fouz independent variables remained s~niIar to Model1, the standar&zedcoefficmnts (beta weights) mdacate that location mthe kistorm core had the largest effect on crtme rates, followed by levels of litter and pedestrian presence (wmttime).

Spatial

Autocorrelatmn

Because of the spataal clnstenng of the data, there was concern about the posslbihty of spatial autoeorrelatmn Problemswith the regressaon analysis because of the spatial nature of the data ear 1end to incorrect conclusions about the effects of the independent variables on the dependentvanable For example,there is the problem of modelnnsspeexficatmnbecause of poss~ie spLI1overeffects The number of crime incidents m one area can be expected to affect the numberof crimes in nelghbonngareas If the model does not account for splllover effects, these effects maybe incorrectly attributed to the independent variables m the analysis AIso, the corretatmn between error terms of the regression model for spatlal umts that are near each other (spatml autocorrelataon) vmiates one of the assumptionsof the regression analysis. Roneekand Montgomery prowde a more detailed chscussmnof the statasticai consequences of crane m one area being affected by crmae mits surroundings (17). Asa first step mtesting for spataal autocorrelation, the Morans I index was ealcuIated using a program recently madeavmlabIe by the Crime MappingResearch Center at the National Instatute of Justice (18). Slgmficancetests for the MoransI index re&eared spatial autocorrelafion m the dependentvariable (LNof crime per 100 riders) and m the residuals of Model I However,the value calculated for the Moran I indexfor the remduals Model which s of 2, included the dummy variable for tnstone core, chd not inchoate any sigmficant spatial autocorrelation Tlns finding led to the behef that the spatial autocorrelatton mthe data set mainlyresulted from the concentratton of h~gh-cnme intersections m the historic core

TABLE Regression 4 Models MODEL 1 MODEL b BETA sig. b BETA sig. 4) 736 0 002 -0.853 0 000 0 449 0 163 0 080 0 452 0 164 0 051 0 096 0 111 0.377 0 000 0 324 0 000 0 755 o0 280 0 003 -0 593 -0 220 0 011 -3 53E-05 .4) 170 0 080 -5 98E--05 -0 288 0 002 1 269 0 398 0,000 0 249 4 850 320013 333039 0 389 1 892 300 252 315 883

Constant LiquorStore(0,1) L1Rer Vis~bihty (0,1) Wa~tTune MstoneCore R-square(Adjusted) Morans (Z-score) "I" Akatl~Criterion Schwartz Criterion

iConstant Lxquor Store (0,1) Litter Vmb~aty (0,I) WgatTime ~stonc Corn SpatmlLag LL~mae Potentud R-square(Adjusted) Moms (Z-score) "I" Akaike Cmenon Schwartz Criterion

MODEL 3 ~ODEL 4 BETA sig. b b BETA sig. 4) 513 0 010 -0 652 0 001 0 453 0 164 0 024 0 481 0 175 0 028 0 063 0 213 0 005 0 060 0 202 0 020 -0 662 -0 246 0 028 -0 492 .4) 182 0 014 -5 2"/E-05 -0 254 0 00] -8 07E-05 -0 389 0 000 1 088 0.341 0 000 0 654 0 261 0 000 1165 747 0 547 0 000 0 471 .4) 091 286 989 305 225 0 460 1 560 287 959 303 590

NOT~. Dependent variable = JaN (crime I00 riders) per

2~ Paper No 0%0441 Altaough slmpteindexsuchas Morans canbe usedto ldenttf a I 7 whether spatial autocorrelatiou ex.~[s,more complex techniques are available estimating effectsof spatialautocorrelation a mulfor the on uple regressionmodelTwo approaches wereexploredFxrst, a spaual staasUcs package, SpaceSta~ was used control the to for degree of sparta1autocorrelataon includinga spatially laggeddependent by variableas an independent variablein the multiple variableregression mode1 (19) second ap proach wa adapted fr om wo byRonek s rk e and Montgomery includedan independent that variable in the multiple regression that measured potentaal (17) Although came Roneek and Montgomery suggestthis approach a wayof handbag as spatial autocorrelation largesamples, approach adapted an alterin the was as naUve means including came of spdIovereffects mthe regression becauseof how samplewasstructured the

Transportation Record Research 1760 Thepresent data are somewhat dLfferentBecause resources the werenot avmlable collect envaroranental to inventory data for each intersection the area, not all mterseetious, thus notall busstop m and came incidents, are includedmthe databaseAlso, a random sampIe of intersections was unav~able. mtersectlouswlth stgmficant All levels of bus stop enme wereincludedmthe study, with a random sampleof low-came intersections to completethe 100 cases This method sample of selection has twoeffects on the analysis. Asnoted before, many the high-came of intersectionsare concentrated two on streets mthe lnstodc cure of downtown Angeles Los This area not onIyhas common values for the dependent variable (crune levels), but also has s~malar env-~ronmental characteristics Therefore,the effects of the independent variablesmthe regreasmn overstated are becauseof fewerindependent observationsIf a correcUon made, is for example, usinga spatiallylaggeddependent variablemthe regresslon, spfllovereffects fromall cronesmthe adjacentareas are not actually accounted because for, these erknedata are not necessanIy mthe database Roneek Montgomery and introduceda elaine potentaal variable rote then: regression analysisto account spitlover effects (17) for Thecame potential as basedon the generahzed potential modelin which crimepotential at each observation tins case, mterthe (m sectiou) Is c, aieulatedas the sum the enme of levels at everyother observation chwded the dastance these observations by to (calculated mthe samewayas the lagged dependentvariable mthe spatial regressionanalysls) Thiscrmae potenuatvaaablecapturespossible diffusion all locat/onsof the studyarea so that mlsspectfication from of the exactrangeof &ffusion effects is not a concern Toapprox.unate came a potential variable giventhe lmntations of the data set, twoadditionalcnme variableswereaggregated for eachmtersectiow buscame a rachusof 153m(500ft) of (a) in intersection,but excluchng 46-m the (152-ft) radius that definesthe dependent variable and (b) bus came a radmsof 229m(750 m but not within 153m. Acame potentaal varutble wascreated that weightedthese two aggregates of crane basedon distance from the intersection (againselecting the squareof &stance the best as function) and includedthis potential variable mthe OLS regressmn equationModel mTable4 showsthat tbas crmaepotential van4 abIe cuntnbuted slgmficanfly the explanation variation mthe to of natural log of crimerates and increasedthe R-square 0 460Coto efficients on the other independent variables wereaUmgraficant, withshght changes value Thebeta weights m show that enme potential hasthe greatest effect, followed wattttrne andvts~bfllt3, by (eyes on the street), as mModel Litter hada smallereffect than mMud3 ds 1 and 2 Thehistoric core variable and came potential couldnot be mcluded the regression modelbecause of multi-colmeanty m problems(they have a sxmpIe correIat~on ooefflcmutof 0 830) Clearly,areas withthe tnghestenme potentml mthe hastonc are core Although calctflanonof the Morans indexfor the reatduals the I of Modeis 3, and4 indicated significantsparta1autocorrelation, 2, no moreconfidencerests mthe results of Model Although sam3 the ple composition problen~,the smlilarity of results among had the models mchcates reasonablespecification for the model a Clearly, an nnportantpredmtor crimeis the location~physmalIy of related to factors associated the b.~stonc Aftercontrolling the with core for locatmn, undesirable fac~htiesandhtter result mhighercame rates, whereas visibflaty and many pedestnam to lowercrimerates lead UatngModel as a starting point, specific designvariables-3 presenceof busshelters, pubhc teIephones,and a caretaker--were tested for their effect on came rates None these factors added of szgnfficantly the equauon, to after the basic factors were controlled

lncluchng a Spatially LaggedDependent Variable TheSpaceStat program used to estunatea regressionmodel was that includeda spatially laggeddependent variable as one of the rodependentvanabIesA spatially laggeddependent variable is calculated as a weaghted average the values of the dependent of variable at ali otherobservauons (1.e., lntersectaons) this study,the value In of the ]agged dependent variablefor eachintersectionzs the sumof the naturallog of came (dependent rate variable)at eachother mtersecUor,weighted the squareof the reversechstaneebetween by the two mtersecUons resulting regression equaUon The enabled an assessment the significanceof the other independent of variables after the spaUaldependence controlled was Thebest regressmu model generatedusing the spatial regression package prechctmg dependent for the variable, LN (came 100ndper ere), is shown lVlodel3 mTable4 Thebeta weights as mdacate that the ~slonccore dummy variable contributesthe mostto the predicnon, followed the spaual lag variable Tinsfinchngcorroborates by earherresults that spataallocaUon the most is mzportant determinant of the cxLme at busstopsAfterthe effect of the spauallag vanrate abie ts J emoved the dependent from vanable,the same vauables m as earher models slgmficanflycontribute to the expIanation came of rates Theszgm magmtude the coefficients remainconstant and of Intersections withhquor stores andhtter tend to havemore craneper rider, whereas wslbflaty andpedestrianpresence(watt tgne) have lowerermae rates Although Table4 showsa higher R-squarefor the spalaal lag model (Model zt is not appropriate compare R-square 3), to th~s onege~crated by ordinaryleast squares(OLS) alternative meaTwo sures el fit reportedby SpaceStat directly comparable are between medeIsgeneratedby OLS spatialregressmnmodels~the and Akaake Criterion and the Schwartz Criterion Thebest modelxs one with the lowestvaluefor these cntena,mtl~s case, the spataallag model has slightly lowervalues than the OLS model

Crime Potential Most [ the hterature addressing o xssuesof spatial autocorrelauon assumes that the observations are spatmlly contiguousAlthough observanons be points on a grid suchas lntersectmns, zones may or suchas city blocksor censustracts, ali spatial umts assumed are to be included mthe anatysls A mum aspect captured ma regressmn model~th a spatialIy laggeddependent variable is the spxllover effects (ff came adjacentareas from

bggettst aI Type ! Crime A smular anatysls was conducted consdermgoniy Type I crane The interest wasmwhetherdifferent envtronmentalfactors were related to moreserious crimes againstpersons, such as murderorrape A sanllar patterr wasfound as that for total came----the most important envaronmeatal factors were pedestrum presence (measured as wait tame) and vlslb~hty, wluchwere both negaUveIy correlated with crune rams, followed by the existence of htter, whichwas posmvely correlated wlth crime rates The existence of undesirable estabhshments such as hquor stores d~d not enter rote the regressmnmodelLocation mthe bastonc core had no relaUonshtp TypeI came,rather, the slgto mficant location dummy variables were location m the shd row area mdowntown the Westlake region of the expandedstudy area. and ACKNOWLEDGMENTS

Paper No 01-D441 27

Thxs study was supported by grants from the ;ohn Randolph and Dora Haynes Foundation and the Cakforzia Department of Transportataon (through the Umverslty of Cahfomm Transportation Center).

REFERENCES A, Measunng 1 Louka~tou-S1deris, It. l.zgge~ tL Inch, and w.rSurlow the Effects ofButlt Envtronment Bus Stop Crane Enwronment on and Planmng 2001(m preparatton) B. 2 Eck, L, and D Welsburd,ads Croneand Place WillowTree Press, Monsey,New York, 1995 L Activmes the and 3 Sherman, Hot Spots of PredatoryCrane Roulane Criminology Place Cronmology, 27, 1989, pp 27-55 of Vd 4 MaJtz, M, A. Gordon,and WFriedman.Mapping Crime m Its CammumtySemng Spnnger-Verlag, NewYork, 1991 5 Loukaitou-Sldens, Hot Spots of BusStop Came Importanceof A The Environmental Attributes Journal the Amencan of Planmng Associatzon, VoL No4,1999, pp 395-411. 64, 6 Welsburd, D Reanermng Cr~nePrevent~on Research andPohcy Tram the Causes Cnmmahty the Contextof Came of to Na~onal Ir~tltute of Justice, Washington, C, 1997 D 7 Taylor, R B, D Gotl:fredson, and S Brewer Block Csune and Fear Defensble Local Space, Sacral Ties, Terntonsl and FuncUonmgJournal of Research m Cameand Dehnquency,Vol 21, 1984, pp 303--331 8 Newman, Defens~ble Space Crone Preventtan Through Urban 0 DesigrL MacM.lllanCompany, York, 1972 New O for Creating Defensible Space, 9 Newman, Design Gmdehnes NallonaIInshtute {)fLaw C~rrunal and Justice, Washington,C, 1976 D I0 Loulmttou-Sxdens,A, and R Laggett OnBusoStopCrane ACCESS, Vol 16, 2000, pp 27-33 at 11. Levme,N, MWachs,and E Shxrazl Came Bus Stops A Study of Eav-xronm_~ntal FactorsJournal Archttecture Planning of and Research, Vol 3, No4, 1986, pp 339-36I. 12 Lopez,R.Sttrwvmg the RTD, AngelesTimes, Feb21,1993,p22 on Lo# RobeNeighbozhoodDeslg-a Crmae and Journal 13 Greenberg, S, andW ofAmencanPlannmgAssactauon, 50, No I 1994, pp 48-61 Vd and 14. BursLk, R, and H Grasnnek. Nezghborhoods CrzmeLexington Books, New York, 1993 15 Coleman,A Utapm Trial V~swn Reahty on a PlannedHansmg on and Estate HdarySlupms~n, London, 1985 of C~nes.VintageBooks, 16, Jacobs, I TheLtfe andDeath GreatAmencan NewYork, 1961 17. Roncek, andA Montgomery. D., Spatial Autoco~relauon Rev~slmd Conceptual Underpmmngs and Prac~acal Gmdehncs the ofthe for Use Genemhzed Potenuala Remedy Spatial as for Autocorrelatmn m Large Ssmples In CameAnaly~s ThroughComputerMapping(C Block, MDabdoub, S Fregly, ads ), Pohcs BxecutiveResearchForum, and Waslungton,D C, 1995 18 Levme, CnmeStat, Spattal Stan~zcsProgram the Analys~sof N A for Came Incident Locatmns National Institut~ of Justice, Waskmgtor~ D C, 1999 I9 Ansehn,L SpaceStatA Program the Stat~#ncalAnalys~s Spa. for of ttal Data~ Nauonal Centerfor Geograpbac Infonnataonand Analyns, Umverslty Calfforma,Santa Barbara,1992 of Pubhcat~on of thts paper sponsored Cornn~eeBus by on Transtt Systems

CONCLUSION: WHAT CAN BE DONE ABOUT 8US STOP CRIME? Tl~s analys:s inchcates that ecological and compositlonalperspectwos can be complementaryin explammg enmeincidence Most bus stop crimes tend to occur m dangerous places. Whythese places have a h~g~er crime potentm]than others can be partly explainedby their soclal composmonal and charactenst~cs But witkmthese dangerous locales that concentrate manyhot spots of came, some sp~es are muchmore dangerous area others The design and layout of the phys cal envzronment be conduciveto crane or can reduce can opportumtiesfor ermanal actions. Thanstudy found many instances of bus stops mthe tnstunc core (an area with bagh camepotential) that were crime ridden, whereas others mthe same area and along the samebus route were mostly unscathed Onthe basis of the findrags, It cart be concludedthat the presence or absence of eertzan envzronmentalcharactens~cs m the mieroenvrronment a bus stop of can affec: the incidence of crime Transportation and mumelpalagencies can draw certain lessons from tins stud) Because came tends to be concentrated dlsproport~onately mspecific dangerouslocales, intense bxcycle and foot patrolImg by pohce in these areas should reduce opportunities for came. At the same tmae, an array of pohcy and demgn opt~ons~ some quxte smaple--can complementpohcmgGoodvxmbfl~ty from the surrounding buildings and pedestrian presence are important variables m reducangcame, Every effort should be madeto site bus stops away from desolate spaces, emptylots, and vacant braidings and mfront of cztabhshments that offer opportunities for natural survdllance The placement of bus stops near undesLrable establishments (hquor stores, bars, adult bookstores and marnetheaters) and near faclhtaes that favor manycash transactions (pawnshops, checkcasbang eslabhshments) should be avoided SomeUmes, tbas may smaplymeanmovinga bus stop a few yards up or downa street or at the opposite comerBus shelter design should not create an opucal bamer to wowinga bus stop from surrounding estabhshments. FmalIycwagone:as should strive to keep the bus stop envtroranent free of graffiti and litter, thus sending the messagethat someone other than the criminal is in eontroI of the bus stop environment.

TRANSPORTATION

RESEARCH

RECORD
JOURNAL OF THE TRANSPORTATION RESEARCH BOARD
NO,

I760

Transit
Bus Transit and Maintemance, Paratramsit, and NewTechnology

PublicTransit

A PEEI~REVIEWED PUBIJCATtON TRANSPORTATIONBOARD OF THE RESEARCH TRANSPORTATIONRESEARCHBOARD- NATtONALRESEARCH COUNCIL

NATIONAL ACADEMY PRESS Washington, C D 2001

Transportation

Research Record 1760


!

Contents
Foreword PART 1--BUS Bus Timetables with Even Passenger Loads as Opposed to Even Headways Avlsha= Ceder Trip TimeAnalyzers: Key to Transit Service Quality TheeH. J. Muller and Peter G. Furth Bus Stop-Environment Connection: Do Characteristics of the Built EnvironmentCorrelate with Bus Stop Crime? RobinL=ggett, AnastasiaLouka~ou-S=der=s, H~royukiIsek= and Designing Synchronization BusT=metebles into AvishszCederand Ofer Tel BusRapidTransit Institutional Issues: The Route from Research to Experience Mark A. ME[ler and StephenM. Buckley Bus Transit in the 21 st Century: Some Perspectives and Prospects HBrbert S LevJnson Evaluation of FeederBus Systemswith Probabilistic Time-Varying Demands and Nonadditive Time Costs StevenI Chien, Lszar N Spesovic,SpyridonS. Etefs~nJotls, and Renu S. Chhonkar 3 vii

10

2O

28

34

42

47