|USFDC Home||| RSS|
This item is only available as the following downloads:
xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam 2200433Ia 4500
controlfield tag 001 029197860
006 m o d
007 cr bn|||||||||
008 110928s2002 flue obt 000 0 eng d
datafield ind1 8 ind2 024
subfield code a C01-00122
Why people cross where they do
h [electronic resource] :
b the role of the street environment /
Xuehao Chu, Martin Guttenplan, Mike Baltes.
[Tampa, Fla. :
National Center for Transit Research, Center for Urban Transportation Research, University of South Florida ;
Springfield, VA :
Available through the National Technical Information Service,
1 online resource (14 leaves) :
Performed for the U.S. Dept. of Transportation and Florida Dept. of Transportation under contract no.
Includes bibliographical references (leaves 12-14).
Description based on print version record.
Pedestrian facilities design
Baltes, Michael R.
National Center for Transit Research (U.S.)
University of South Florida.
Center for Urban Transportation Research.
Dept. of Transportation.
Dept. of Transportation.
t Why people cross where they do.
d [Tampa, Fla. : National Center for Transit Research, Center for Urban Transportation Research, University of South Florida ; Springfield, VA : Available through the National Technical Information Service, 2002]
Center for Urban Transportation Research Publications [USF].
y USF ONLINE ACCESS
Why People Cross Where They Do: The Rol e of the Street Environment Final Report
\ RfPOI'I NC TR-473-06 4. Tll:lt end s..colle: s. Repo.t ();"' Why People Cross Where T hey Do : September 2002 The Role o f the Street Env i ronm ent i ,..rf
Wby People Cross Where They Do The Role of the Street Environment TRB Paper No 03-3078 XuehaoChu Center for Urban Transportation Research Un iversity of Sou th Florida 4202 East Fowler Avenue, CUT tOO Tampa, Florida 33620 email@example.com Tel : (813) 974 -9831 F ax : (813) 974 -5168 Martin Guttenp lan Florida Department of Transportation Systems Planning Office 60S Suwannee St. MS-19 Tallahassee, FL 32399 manin firstname.lastname@example.org .fl.us Tel.: (850) 414-4906 F ax : (850) 92 1-6361 Mike Baltes Center for Urban Transportation Resear<:h University of South Florida 4202 East Fowl er Avenue, CUT 100 Tampa, Florida 33620 email@example.com T el.: (813) 974-9843 Fax: (813) 974-5168 Revised September 2002 This paper model s the role of the street envirorunent in h ow p eople cross roads in urban $ettings. Respondents were placed in real traffic conditions at the curbside of street blocks in the Tampa Bay area for a three-minute observation of the street environment. \Vithout crossing the-blocks, each responde-nt stated his crossing preferenc e at each of six blocks. The origin and destination of each crossing were hypothetically set and varied across the blocks. So were the options avail ab le: two options for crossing at an intersection and up to four options for crossing a t mid-bl ock locations. Within the framework of discrete-ch oice models, the s tated p r e ferences are explained with the street environment, includin g traffic conditions. roadway characteristics, and sign.al-contro1 characteristics. All three components of thestreet environment arc considered: mid-b1oc. k loc-ations, intersections. and the roadside environment. The paper destribes survey design and data col1e<:tion efforts; estimates a nested logi t model of pedestrian stre<:t-crossi.ng behavior; and discusses its implications to researchers and practitioners.
INTRODUCTION Street crossing is a critical clement of the urban transportation environment for pedestr ians. A large body of work already exists on street crossing by pedestrian<, including the following by subject area: Crossing delays (1), Crossing opportunities (Z J), ltedestrians' behavioral parameters such as walking speed, start up time, and gap-acceptance ( 4 -6)., Pedestrian compliance (7), Pedestrian perceptions to,v.ard specific treatments (8 9), Determination of l eve l of serv ice ( 1013), Engineering parameters such as pedestrian clearance intervals (/4). Evaluation of treatments(/ 5), Drivers perspective, including pedestrian visibility effect of crosswa lk markings non compliance with signals (19-21), Safety (22 -23), and Empirical modeling (24 26). However,littlc research exists that can help answer questions related to pedestrjan pJarming, engineering solutions to pedestrian c rossing safety. and researc h methods for modeling behavior. Below are a few examples of lhese questions: Planning Questions- How can existing planning tools for detennining pedestri a n level of service for street crossing at m i d block locations and intersections be integrated to de termine pedestrian level of service at t he block level? Engineering Questions How and when migh t a pedestrian go to a marked crosswalk in mid block locations? How and whe. n might a pedestrian go to an inte rsection for street crossing? Where should transit bus stops be located so that transit users ar e likely to choose safe crossiog options to access them? Research Methodology Questions What sta tistical models are most appropriate for modeling the street-crossing behavior of ped estrians so that these planning an d engineeri .ng questions c-an be answered? What and h ow should data be collected in order to estimate such statist ical models? This paper models the role -of the street environment in how people cross roads in urban settings Specifically, 86 participants placed in real traffic conditions at the curbside of 48 street blocks in the Tampa Bay area observed the stree t environment for three minutes. \Vithout crossing the street bloc ks. each participant s t ated his crossing preference at each of six blocks. The origin and destination for each crossing were hypothetically set and varied across the blocks. So were the options avai lable: two options for crossing a t an intersection and up t o four options for crossing at a mid-block l ocation. Within the framework of discrete choice models, the sta ted preferences are expla ined by traffic conditions, roadway characteristics. and characteristics. The paper focuses on the street envirorunent so thitt all variables can be rea d ily mea sured for model applications. As an alternative, one could model the role oftbe direct attributes, such as safety and time, that pedestrians may tradeoff in c hoosing a crosslng option. By focusing: on the street environment, the paper as.sumes that the indirect attributes that c harac terize the sttcet environment detenni ne the di rec t attributes a n d that the street crossing behavior can be modele d with these indirect attribu tes equally well. As another alternative, one could include the street envirorunem as well as pedestrians' pe.f'Sonal characteristics. lc is recognized here that these characteristics are potentially importan t in how people cr oss roads. They are excluded solely because data on them arc not readily available for model app l icat ions. The impacts of these two a ltern ativ e spec ifications on model resu lts are reported elsewhere (27) and arc briefly described in this paper when its research implications are discussed. The rest ofthe paper has four sections. They describe: I) the design of the stated-preference survey, 2) data collection efforts, 3) model estimation resulrs, and 4) shortcomings of the study and its implications to pedestrian planning, engineering solutions to pedestrian sa.fecy for street crossing and research, respect ive ly I
SURVEY DESlCN The stated prefe -rence approach was chosen for several r easons It resulted in wide ranges of variation in the street environment It allowed so1icitation of crossing preferences in rea l traffic c onditions It also resulted in a manageable number of crossing options for modeling. The design process for thJs re.search involved four steps: 1 Identify potential determinants of pedestrian street-crossing behavior 2. Detennine levels of key detenninants through the selection of street blocks; 3. Formulate crossing scenarios by defining crossing origins and destinarions, crossing options, and temporary mid b lock crosswalks; and 4. Develop i nstruments for individual crossing scenarios. These reasons and design steps differ from those for a standard stated .. prcference survey (28). Potential Determinants Two steps we.re used to select potential determinants that describe t he street environment The flrst step identified the direct attribu tes that pedestrians may tradeoff in making a choice: c omfort safety, time, and predictability. Predictability refers to the uncertainty in the amount of time an opt.ion may take a pedestrian to cr oss. The second step identified the indirect factors that may determine the dire ct attributes. Comfort and Predictability Differences in comfort result largely from differences in expos we to unpleasantness (such as hot weather) and pe r sonal traits that influence comfort sensitivity (such as poor health) Such differepces are captured with roadside walking and crossing distance Roadside v.-alking could vary significantly across options. Cros.sing distance \aries when j aywalking js involved or when the choice involves intersections and mid-block locations that ha v e different width. Vari ation in predictability res ults from lbe presence or absence as well as the spacillg of traffic signals. Safety and Time The amount of time spent walking along a street is dctennined by the d istance involved and speed of walking Distance is already identified as a potential factor in the paragraph above. The potential factors for safety, crossing time. and waiting time are di s cussed below for crossing at midblock locations cro ssing at interse<::tions, and roadsi de wal king separately. Mid-block Locations. Chu and Baltes (29) identify potential determinanrs for pedestrian crossing behav i or at mid block locations based on supply of gaps, cr ossing time. and safety margin, which fonn the three components of the gapacce ptance behavior of pedestrians (24). Safety margin is the diffe r ence between the time a pedestrian t>kes to cross the traffic and the time the next vehicle arrives at the crossing point. Intersections. Crider ct al. identify potential deremlinants for pedestrian crossing behavior at intersections(/ 1) These are done separately for safety and delays. Safe .ty consists of confl icts with motor vehicles and pedestrian's exposure to these conflicts. Vehi cle movements ac an i ntersec tion that cross the crosswalk represent conflict volumes. Exposure consists of crossing distance, presence of crosswalks and presen ce -of curb or sidewalk, and median type. For pedestrian delays, the po tentia l detenninants differ between signalized and unSignalized in t ersections. At signalized intersections, pedestrian crossing delay depends on cycle length for crossing with a pedestrian signal an d on the faciJitys green ratio for crossing without a pedestrian signaJ. At un...signalized intersections, pedestrian crossing delay is a func-tion of the conflict volumes described above. Roadside. landis et al. identify a se t of potencial detenninants for pedestrians walking along roadsides (30). Through a step wise regression process. the authors identify a number of factors describing the roadside environment including the various components of l ateral separation between sidewalks and traffic lanes. Site Slection The selection of blocks for the field survey determined the values for most aspects of the street environment and the combinations of these values The following criteria were used: I All blocks had two roads at lhe two ends wilh through movement 2. All blocks were on roads that are functionally classified as eoJiecro r or above in urban settings. 2
3. The blocks were from different regions of the Tampa Bay area. In order to facilitate survey logistics, the selection was further limited to a circle radius within each of four subarcas: northeast Tampa, South Tampa, Clearwater, and St. Petersburg. 4. A number of potential detenninants were considered, including number of lanes, presence and type of medians, s i gna l ization and crosswalk marking. at intersections, pedestrian signal heads at intersections, side,valks,la ter al separation between sidewalks and traffic Janes, and block length. 5. A wide range of combinations of the values of the considered determinants was included. For example, it is desirable to have blocks on a 6-lane road with medians and b locks on a 6ane road without medians. 6. A total of 48 b l ocks were selected with 12 from each area. The number 4& was cho sen because it resulted in 12 blocks in each area. Field surv eys were done on different days in lhe diffe -rent areas. Furthermore, the 12 blocks in each area were divided into two groups of 6 each. These two 6 block groups were visited by two different groups of Sut\ey participants with each group taken by a bus. Based on the survey experience reported by Baltes and Chu (10), a single bus was able to visit six sites in a single day. "The acrual select ion was a manual process with hundreds of miles of driving and several steps: Produce GIS maps that show roads classified as collector or above within each circ:le. Identify blocks in u,e field that meet criterion 1 and record information on the determinants in criterion 4. Based on the infonnation from the field, select 12 candidate blocks within each area that meet criterion 5. Cheek selected blocks in the field and adjust when needed. Crossing Scenarios A crossing scenario is what was presented to a survey participant for soliciting his state-d crossing preferen ces. a crossing scenario for a block consisted of the street environment the origin and destination of the crossing, and the crossing options availab le to the pedestrian for the particular origin and destination. Much of the s tree t environment for any block was determined once it was included in the sample of blocks. The only exception was crosswatk markings. particularly at mid-block locations In addition to defining individual crossing scenarios. the design process detennined what set of crossing scenarios eac-h survey participant was pre.sented with. Start and End Points The origin and destination for any crossing scenario were cal led the start and end points (Figure 1). Five potential l ocations for either the start or end point were considered with equal distance between them For either the stan or end two potential locations were at the intersections. These potential locations allowed a total of25 different combinat ions. Two combinations of start and end points w ere randomly selected for each block. For ease of reference, the side of a block with the start point was called the nearside Qnd the other side the farside. 3
Figure 1. Sample Survey instrument for Stated Prefe.rences P le ase enter your PIN here: -------The diagram below shows your start point, your end poinl and your localion options for c r ossing the street within thi s block. stand at your start point and observe the b toek characteristics and traffic conditions for 3 minutes. Based on your observation of the block and evaluation o f the options during these 3 minutes, please tell us your choice for crossing this street by selecting one from below: A F D c B E 4
Crosswalk Marking Mid block crosswalks rarely e xist in the study area. In fact, none of the 48 street blocks had a mid-block crosswalk. Temporary marking was inslead used 10 define mid -block crosswalks. Aboul half of the sample blocks had a temporary mid-block c rosswalk wit h three. in each sixblock group. A manual process was used to detennine which three blocks in a six b l ock group got a mid block crosswalk or where a mid block crosswalk was placed on a gi\en block. This delermination was made visually with simullaneous consideralion of all blocks in !he same si-block group and with factors considered shown g.raphica11y. Facrors considered include roadway widtb, b l ock length (s h ort. mediul'l\. long). pre.sence and type of medians. crosswalk marking at inter sections, traffic signals, pedestrian signals at intersect i ons. and the two chosen start .. end combinations. Three materials for marking crosswalks were: tested on two clea r days. on two blocks, on a six-lane road, with 12 participants : pavement tapes, chalk powder, and four orange traffic cones with two on each side of the road. The ques t ion for the test participants was: Did the marking adcq\ l ately represent a marked crosswalk t o you during t he test ? The answers were on a 1-5 scale with 5 being adequate and 1 inadequate. Chalk powder was easily blown off by passing motor vehic1e.s. Both orange cones and pavement tapes were perceived to be adequate to represent rea1 crosswalk marking and orange traffic cones were as effec tive as pavement tapes Orange traffic cones were chosen over pavement tapes for Jogistiea1, material cost, and safety reasons. Crossing Options For a given scartend combination. a set of\lp to six discrete opt i ons was defmed tha t can approximate m o st of the polenlially infmile number of eross i ng options These options are labeled as A thtough F for case of reference and defined as follows (left and right are re l ative 10 !he nearside) : A= Crossing at the left intersection (left intersection) B = Crossing at a midblock Slart point a1 a right angle (cross firs! and walk laler) C Crossing wilh a jaywalk berween the slart and end poiniS (jayw a lk) D ... \Va\king to the opposite of a mid block end point and crossing there at a right angle (walk fust and cross la1er) E Crossing at the right intersection (righ t intersection) F Crossing at a mid-b1ock crosswalk that is away from a start or end point (midblock crosswalk) The phrases in the parentheses may be used co refer to these options. The e.acl options vary The availabil ity of oplions A through E depends on !he particu l ar Slart-end combination. If both the start and end points a re located at mid .. block locations but not across each other. for example, opt ions A through E would all be available If the start point is at t he left intersection instead. option B would disappear and option A would no longer involve walking along the nearside. If the start and end points are localed al the same inlerseclion, only A and E would be available. In general, !here are a lola! of five possible seiS of options from !he 25 possible Slart-end combina l ions discussed earlier. These are: A-E, A-C-E, A-B-E, A -C-0-E, A -\lC-E and A B C-0-E. On the olher hand, Oplion F is available only when a mid-block crosswalk is prcsenl and located away from a start or e nd point. All options are available in the diagram in Figure: 1. Group Crossing Scenarios Each survey participant provided 1 2 stated-ptefereoce responses with 2 responses on each of six blocks in the same circle. As discussed earlier. two comb i nations of start and end points were selected for each block. The two responses for a given block from the same participant were for these rwo different start-end combinations. More is discussed on how these rwo responses were obtained in the sec.tion on field surveys. The particular six blocks with i n the same area were detennined with two considerations. The six blocks would result in a route that is similar in length with the other six blocks in the same circle. Each six-b lock group would have as much variation as possible in key delerminaniS. Inscruments There were a lola! of 96 instrumenls with each block h8\ing IWo of !hem, corresponding to !he IWo combinalions of start and end points. Each instrument showed a scaled diagram of the actual block in color. The crossing options, 5
including both the path and the letter label, were coded in colors that were consistent across all instruments. For a given block, the start and end points for one of the two combinations were coded in red and the others in blue. Figure 1 shows an example of the instrument wit h a start-end combination in blue. Note that the duration oftluee minutes was chosen so that the participants can observe the street environment for a full signal cycle in most cases. Also the exact order of the options in an instrument depends on where the start and e.nd points are lo<::atcd. DATA COLLECfJON Several aspects of the data col1ection logistics were discussed earlier. This section focuses on collectlon of static data and field surveys. Stade Data Data describing the sta tic aspects of the stree t environment were -colle<:tcd whilethe survey instruments we.re bei.ng developed. A fonn was developed for field collection. It had a section for data re l ated to crossing at each of the five potential start points and a section for data related to the roadside environment. Before any data were rec.ordc.d, block length was measured and each of the five possible start and end points were marked. In addition. the combinations of s t art and end points were into b1ue or red as designed Field Surveys The flfia1 sample of 86 survey panicipants was recruited through a temporary stafrmg agency. The initial target sample size was 96 so that a tota l of24 would participate on each of the four suiVey days with 12 on each bus. Ten did not show up for all four days com bined. This approach to selecting participants gave greater certainty in the number of recruited participants who a ctually showed up Given the fact that completing the field suiVeys for any given participant took about S hours, recruiting voluntee r s through random sampling of residents in the study area would not have worked as well. Field surveys were conducted toward the end of April2002. Prior to departing a central location each day, participants were given verbal instructions and a participant identjfication number (PJN) at random. The PlN;S were numbered consecutively from I After the briefmg, those participants with even PJNs boarded one bus and the others the other bus At each block, the participants from the same bus were divided into two groups of around 5 to 6 in each group. Two survey workers brought one group to the blue start point and the olher to the red point. These two su!Vey workers were supervisors for the two start points. Both of them recorded the PINs of those participants at their start point. Both were also responsible for distributing and receiving the instruments and checking whether the instruments were filled properly. One of them was a timer as well who not only determined when to start and end a particu l ar cros sing scenario but also recorded times. In addition, 1he timer had a shee t with an six blocks that color coded the locations of the two start and end combinations and mid b l ock crosswalk marking At the same time, one survey worker brought one red flag and one blue flag to mark the end points for the two combinations. Another brought the orange traffic cones to the appropriate loca t ions if a mid-block crosswalk was requi_red. Two survey workers got into position for collecting turning movements at inte. rsections and two others for turning movements at mid b l ock l ocations (including driveway vol'umes and U tums) as well as large vehicles (i.e., truckst buses, and vans that are larger than regular household vehicles). These were the data collectors, who used pre-developed fonns for these dynamic data Once everyone was in position, the timer signaled to e veryone when to start a crossing scenario. Once started, the participants were given three minute-s t o obsene the street environment as indica te d on the survey instrument and were asked to itll the instrument right after being instructed to stop observation Meantime. the data collectors were recording turning movemenls and the number of large vehicles Also. pre-laid traffic counters were recording directional traffic volumes by speed ranges. four two-way radios were used for communications. Once each g roup was done with its f1tst crossing scenario for the block, the two groups then switched locations with each other. Once bod> scenarios were done for a given block, everyone boarded the bus and traveled to the next b lock. 6
Dataset A dataset was developed from the survey scenarios, static data, dynamic data, and stated preferences. It contained a total of 1,028 observations (out of 1,032 possible observations) and 42 independent variables. Among the variables, 6 are nip characteristics, traffic characteristics, roadway characteristics for crossing, and traffic control characteristics, respectively Nine are roadside characteristics that are measured separately for each side of the blocks. The flrst two columns of Table 1 explain thtse variables Blocks The 48 blo cks had a range of combinations of the potential detenninants considered. The average. length was 618 feet with a minimum of232 a maximum of 1,300 and a standard deviation of314 feet Ther e were 15 blocks on a 2-lane road, 16 on a 4-lane road, and 17 on a 6 -lane road. Sixteen ofthes e blocks were undivided; 20 were with restrictive medians (taised or grassy); and 12 were with pai.nted medians. Crosswalk marking was present at both intersections for 7 blocks, at one intersection for 24 blocks, and at none of the intersections for 17 blocks. Participants The sample of participants had more females than males but bad a reasonable spread by age and househol d income. The 65+ age group crossed far fewer roads thin the younger groups on the day before survey, ranging from one third of the average crossing by the 25-44 group to one half of the average by the other groups. On the day before survey, the female participants crossed roads one and halftimes versus and quarter times by the male participants. Few of the participants perceived themselves having difficulty with walking at normal speed. These were evenly distributed between the two genders. Far more of them perceived h aving difficulty with walking at higher walking speeds, however, especially with the 45-64 group. Every one of the 13 participants who reported no difficulty at normal walking speed but reported difficulty at higher walking speeds was female. Descriptive Statistics All potential independent variables were examined for correlation. For example, traffic volume is positively correlated with green time and crossing distance for each intersection opti on with a correlation coefficient s1ightly over 0.5. This information was then used later in model estimation. In addition, individual independent variables were examined for the reasonableness of their mean, standard deviation, maximum value, and minimum value. ESTIMATION Hypotheses Hypotheses were fonnulated for a statistical model and expected directions of effects of the independent variables. Statistical Model It was hypothesized that the most appropriate statistical model is the nested logit model (31). It is natural to view the six potential options for street crossing as rwo distinctive groups: those related to cross at intersections and those related to crossing at mid block locations. That is, the nested logit model has a two-level structure. The top level has two branches: intersections (1) and mid-block locations (M). The bottom level has two options in the intersection bran c h (A and E) and up to four options in the mid-block branch (B, C, D, F). l11dependent Variables The hypothesized direction of effects of independent variables was based on a basic specification of the utility functions. This specifi cation involved two aspects. First, aU variab1es were to be entered linearly to reduce the complexity of the model. Second dte specific utility functions to which a particular independent variable may enter were determined. Several criteria were used for this purpose. One criterion was whether an independent variable is constant across the options (e.g roadside walking varies but not total traffic volume). One criterion was whether an independent variable is dcfmed for each crossing option sig.nali2ation is defmed for intersections only). Another criterion was whether a specific direction of effects could be hypothesized (The width of shoulders or bike lanes is likely to increase the probability that pcdesrrians choose options that require roadside walking but is likely to decrease the probability that they choose options that do not require such walking). Based on this specification, hypotheses were formulated for each independent variable. Tabl e I also shows the specificat i on and hypotheses for individual options and the two branches. 7
Table I. Variables and Hyp<>theses Va r iables Hypotheses Indiv i dual Options Branches Descrip tion Unlt A E B c D p I M Trip Wal ki n g dist ance Feet along roadsides -----Crossing d istance Feet o n travel lanes ----S t art and end at m i d b l ock locatio n s I if true; 0 oth erw ise + Stan at mid-b lock & end at intersection l if true; 0 otherwise + S t a rt at intersection & end at midblock I if true; 0 otherwise + S t a n and end at intersections I i f true; 0 ot herwise + 'J'rafflc Traffic volume Vehicles per hou r M i d b l ock running speed Miles per hour -Vehic-le mix Percent trucks ----D r i\'cway volumes Ve hicles pe r hou r ----M i d -bloc.k U -t u ms Vchic.Jcs per hour --I n t ersec tion turnings Vc:hidcs pe r hour -Roadway C r o ss ing Right-turn l ane I if present; 0 otherwise Left-turn lane I if present: Ootherwise --Accclcrmion lane 1 if p r esent; Oothc r w isc --Crosswal k rmuking I ifmaliced; Oother\lt ise + + + + + + Restr i ct i \'e medians Width i n feet + + + + + + med i ans W i dt h i n feet ----Roadv.'a)' Ro,ulside Driveway frequency ( nearside) Number + + S idewalk (neart i dc) I i f pre s e nt; 0 otherwise -Buffer (nears i de) l if present; 0 otherwi s e -Barriers i n buffe r (nearside) I i f pre s e n t; 0 otherwi s e Curbed roadside (n earside) I i f curb<:d; 0 otherwise Wid t h o f outside lane (nearside) Fee t W i dth
Model Estimation Model estimation was a complex prOCe$S because of t he large number of variables and multiple utility functions involved. Model estimation followed twoscagcs and multipl e s t eps. The fusl stage resulted in a basic model !ha l included only those characteristics thai were explicilly shown i n the instruments: traffic signals, pedestrian signals., cross,vaU c s, relative cro$$ing distance, relative roadside walking distanc.e> and the location of the start and end points These characteriseics were highly significant and showc
Table 2 N ested Logi t Mod el of Pedestrian Street Crossing Behavior (!statistics in paren t heses) 1 Coeffic ie n t Individual Opti ons Branches Variable Definition lnter Mid -Intersect ions M i d-b l ock sections b l o c k A E D c D F ( M Alternat ive-specific I 2.2079 1.7266 1.3875 2 2332 con s t ant (7.34). (5.SS) (3.44) (4.2 0) Wal king d i stance Feet along roadsides -0.003 4 -0.0034 0.0034 -0.0034 .0034 0 0034 (-1!.65 ) (-! !.65) ( !.65) (l !.65) (-1!.65) ( 1!.65) C r ossing distance Feet on tra\'e llanes -0 0027 -0 0027 0 0027 0 0027 -0.0027 0 0027 ( 31) ( -2.31 ) ( 2.31) (-2 31) ( 2 31) (. 3 1 ) S ta rt and e n d at I i f true; 0 ot h erwise !.5722 mid-b l ock locations (3.14) Start at mid b l ock & 1 i f 0 otherwis e 0 8 415 end at intersecti o n (2.32) Tnffievolum e Vehicles pe:r hou r 0003 0 0003 -0.0 0 03 0003 ( !.77) ( -!.71) (.77) (-!.77) Crosswalk mark i n g 1 i f marked; 0 otherw ise !.0002 !.0002 0.7891 0.7891 0.7891 0.789 1 ( 4 .30 ) (4.30) (4.02) (4. 02) (4 02) ( 4 .02) Wid1h of nearSide Feet if p r esen t; 0 -0. 0728 -0.0728 s hould e r /bike lane otherwise (!.22) (-1.22) Width of farsidc feet i f present ; 0 .().0923 -0. 0923 shoulder/bike lane otherw i se (-1.42) ( 1.42) Traffic signal I if prescnt;O otherwise 0.7502 0 7502 (3.42) (3 42) P ede stria n s ig na l I l fprtsent; 0 otherwise 1.23 50 1.2350 (4.34) (4 .. 34) Inctusi\'e va l ue : 11 e Ln( e0"'+ eut) 0.1585 Intersec t i o n s (7 05) Includ e value: Mid 0.8342 blo ck Ln(eu" + eu< -1 eUo + e li') (5 87) Uti li ty f u nction t Coefficicnl) UA U o Uo Uc U o u u U M Number of Observations 1 ,028 Number Cases 4,3 34 Log likelihood with constanlS on l y -1769.605 log l i k elihood at converge n ce -963.72 8 Unadju s ted p -1 0 455 Adjusted p' 0 453 1 NLOGIT 3 0 of E conometri c Software, Inc. was used to estimate th i s model with full informa t ion maximum l ikelihood. The RU 1 nonnalization was used for the scale parameters. T h e nested lo git model has two le\ els wi th variable optio n s acros.s observat i ons The top level has two branches: intersect i ons and m i d block l ocations The bottom le -vel has two opt i ons in th e i ntersection branc h ( A and E) and u p to fou r options in the m i d -bloclc: b raoch (B, C D, F). A = Cross ing a t the l eft intersection (left intersectio n); B = Cross i ng at a mid block s tart p o int a t a right angle ( c ross first and wal k later) ; C Cross i ng with a jaywalk between the stan and end points (jaywa lk); 0 .._Walking to the opposite o f a mid-block end point a n d c r ossing there at a right angle (walk first and cross later); E =Crossing at
With respectlO roadside walking, the elast i city is -1.547 (A-le ft intersection), -1.853 (-right intersec t ion), -0.243 (B-c r oss first and walk later), 0.345 (D-walk first and cross later), and 0.232 (Fmidblock crosswalk). The probability of an intersection being chosen is highly responsive. An of 10 percent in roadside walking cou l d reduce the probability by 15 to 18 percent. ln contrail, the probability of any mid-block option being chosen is irrcsponsive. With respect to traffic ''olume, the elas t icity is -0.197 (B-cross fU"St and walk later), -0.273 (C-jaywalk),0 .134 (D-walk frrsl aod cross later), and -0.059 (F-midblO<;k crosswalk). are less likely to choose mid-b l ock options when traffic volume increases. This impact, howeVe r is irrc spons iv c Furthennore, the values for options B (cross first and walk later), D (walk first and c r oss later), and F (mid-block crosswalk) are several times higher in magnitude than those with respect to crossing distance but lower in magnitude than those related to roadside walking distance. For option C (jaywalk) however, the elasticity with to traffic volume is only half oflhat in magnitude as crossing distance. To present the formula for probability calcu lations, let Uo (0 =A, E; B, C, D, F; I, M) be the sum oftlte produclS of all variables in the fJist column with the parameter values for optio n 0 on the side columns in Table I. No t e that the indusive values are V1 = Ln(e + e "") and v., = Ln(e"' + e"< + eu+ e ')for the intersection and mid-bloc k branches. respectively. The probability of a crossing op tion being chosen is the product of its marginal and conditional choice probabilities. The conditional probabi l ity represen t s tlt e probability of choosing a particular crossing option once the choice has been made berween intersections or mid-block options. With intersections being chosen(!), for example, the conditional probability of intersection k (k = L, R) being chosen is given by P(k/ I)= e"t ev' With mid-block options being chosen similarly, the probability o f midblock option m (m = B, C, D F) being chosen is given by P(m I M) = e"J c The margina l probability represents the probability of choosing intersections or mid-block op tions. Specifically, the probability of either being chosen (J I M) is P(J) = e _u, I cv where V = J.n(cu + eu'}A). DISCUSSION Limitations Before discussing potential implications, it is crit i cal to understand the simplifications made as pan of the research. One simplification is that the dynamics of traffic conditions and pedestrian's street crossing behavior are modeled away. The model relates the average traffic conditions during a three-minute ]X'riod with how a pedestrian may have chosen to cross a street block under such average conditions Whether s afe traffic gaps are avail able can change quickl y ove r t i me and across locations along a street b lcx:k. Suc h temporal and spatial dynamics in traffic conditions lead to dynamics in the sttcct c rossing behavior of pedestrians as well. This simplification falls short for understanding cc::rtain crossing behavior, such as mid-block dash, i.e. siruations where the pedestrian unexpectedly appeared in front of a motorist while the-pedestrian was ru1ming and the motorists view was not obstructed (.J5). Another simplification is that it ignores the role of time constraints. Relative to o ther direct attributes. time and its predictability woul d be-come far mor e important 10 a pedestrian w h e n he has a tight time constraint. As a result, he may take riskier crossing op tions. By exc luding time-constnl.ints, the usefulness of the model is reduced in understanding the behavior of transit users in trying to c-atch a c oming bus on the other side of the road. The exclus i on is made partly because o f the difficulty in modeling time constraints. Implications Implications relating to rese-arch, planning toots, and engineering so lutions are discussed. Research Methods. A number of implications can be drawn that have both currenr and lasting value to researchers. These include: 1. The results show that pedestrian street-crossing behavior can be reasonabl y modeled with indirect factors that can be direct ly measured in practice. In this case, the indirect factors describe the street environment. Howe\er, an otherwise similar model based on dire<::t factors alone fits the reported pedestrian street .. crossing behavior better In fact t h e adjusted p' increased from 0.453 to 0 .552. The direct factors measure perceived safety, time, and predictability on a scale from I (le3.1t favor able) to 10 (most favorable). The data \'.ere co11ected from the respondent s in the field jus t after they stated their crossing preference for each crossing s cenario. II
2. Excluding personal attributes from the preferred model appears to have small impacts on the model. An alternative model with added personal anributc s was estimated The addition improved the preferred model with an increase in the-adjusted p2 to 0.471. The elasticity with respect to roadside walk ing was compared. for example, and it increased from -1547 to -1.593 forthe let\ intersection and from -1.853 to -1.901 for the ri ght inte rse ction. 3. 11le r eponed results earlier show that the nested logit model fits the stated pedestrian street-crossing behavior better than the conditionallogit model. 4. The quasi-stated prc:fcrcncc approach provides an alternative to the standard stated-preference approach. S. The survey design provides an example of modeling th e continuum of street crossing options in real Jife with discrete methods. Planning Tools. 11te existing too ls for determining pedestrian level of service are based on simple regression models that predict pedestrian perc eptions of quality of service with the stree t environment. The estimated model from this res earch could ptO\'ide a new approach that is based on pedesrrians overall satisfaction with street crossing. SpecificaUy, the estimated utility functions can be combined to provide a meaningful measure of the overall satisfaction from crossing specific bJocks: V ln(e0 + e0ot), This con c ept is similar to using the denominator of a logit destination choice model as an acceftSibiHty measure -(36). More important, this new approach to determining pedestrian level of service is also a behaviorally sound way to measure level of service across different modes equally. The Na tional Corporate Highway Research Proi!J'&m has planned a research project to look for a unified approach for equal measurement oflevel of service across modes (37). E-ngineering Solutions. TI1eestimated model may be used to simulate how certain engineering solutions may influence how pedestrians cross streets. I. The model can be used to determine the circumstances under which pedestTians are more likely to go lOan intersection or a mid-block crosswalk. With some basic assumptions, curves may be developed to show how different combinations of selected aspects of the street envirorunent influence the likelihood that a typical pedestrian would choose an intersec t ion or a mid block crosswalk in daytime conditions. 2. The model can also be used to dete nnine how marking a midb1ock crosswalk may discourage pedestrians from taking risky options. 3. Transit stops are often the destination of pedestrians crossing a street. \Vhcn these stops arc located inappropriately. transit users may be more likely to take risky options for crossing. F or given origins the model can help understand how the destination within a block can influence the likelihood of pedestrians to take risky options. The same i_mplic:ation also applies to l ocating walkways from major activity cente r s. newspaper boxes, vending machines, etc. The aetuaJ simulation requires additional space to explore and may be c arr ied out in a later paper. ACKNOWLEDGEMENTS The Florida Department of Transportation provided funding through the National Center for Transit Research. We like to thank District Seven of the Department for performing the traffic counts and the HARTline for providing bus transportation. REFERENCES I. Rouphail, Nagu M., Joseph E Hummer, JosephS Milazzo 11, and D. Patrick Allen (1998) Recommended Procedures for Chapter 13, "PedestTians," of the HCM, FHWARD-98-107. 2. Hidas, Peter, Kolita Weerasekcra, and Michael Dunne (1998), Nega tive Effects of Mid-block Speed Control Devices and their Importance in the Overa\llmpact ofTraffic Calming on the Envirorunent, Transponation Rescareh-D 3, pp 41-50 3. John, and Jalal Abduljabbar ( 199 3) Crossing the Road: A Method of Assessing Pedestrian Crossing Difficulty, TJ'Offic Engineering and Contro134, pp. 526-532. 4. Knoblau ch, Richard L., Martin T. Pietrueha, and Marsha Nitzburg (1996), F ield Studies ofPcdestri an WaUcing Speed and Start-up Time Transportation Researeh Record 1538, pp. 27-38. 5. Oppenlandcr,Joscph C. (1996) Correlation of Volumes with Accepta b le Gaps in Pedestrian Signal Warrant, in Compendium ofT ecbnical P apers for the 66'" ITE Annual Mee ting pp. 158-162. 12
6. Connelly, Marie L., Helen M. Conaglen, Bany S. Parsonson, and Robert B Isler (1998), Child Pedestrians' Crossing Gap Thresholds, Accident Analysis. and Prevention 30, pp. 4 4 3A53. 7 Mila2zo II, J osephS Nagui M. Rouphail, J o seph E. Hummer, and D. Patrick Allen ( 1 998), Quality of Service for lnleiTIIpted Pedestrian Facilities in the 2000 Highway Capacity Manual TRB Paper No. 99-0 131. 8. Sisiopiku, Virginia P and Darcin Akin (2000 ) Assessment of Pedestrian Crossing Options, Transportation Resear c h Board Paper No. 00 -1494. 9 Sharples, J.M., and J.P. Ple t cher (2001), Pedestrian Perceptions of Road Crossing Facilities, Trans port Research Laboratory, the Scottish Executive Central ReseaTCh Unit, Edinburgh. 10. Baltes, Michael R., and Xuebao Chu (forthcoming) Pedestrian Level of Service for Mid-block Street Crossing, Transportation Resear ch Record. II. Crider, Linda B., J odi Burden, and Feng Han (2001), Multimodal LOS Poin t Level of Service Project-Final Report, Department of Urban and Regional Planning, University of Florida, Gainesville Florida. 12. Sarkar, Sheila (1993) Determinat io n of Service Levels for Pedestrians, with European Exampl es, Transportat ion Research Record 1405, pp 35. 13. Sarkar, Sheila (1995), Evaluation of Safely for Pedestrians at Macr o -and Microlevcl s in Urban Areas, Transportation Research Record 1502, pp. 105-118. 14. Kochevar Robert A., and Na2ir Lalani (1985), How Long Should a Safe Pedestrian Clearance Interval Be? IT Journal 55, pp. 30-32 and 49. 15. Lalani) N. (2001), Alternative Treatments for AtGrade Pedestrian Crossings Institute of Transportation Engineers. Washington, D.C. 16. Zegeer, C.V., J.R. S t ewart, H Huang, and P. Lagcrwcy (2001) Safely Effects of Marked versus Unmarked Crosswalks at Uncontrolled Locations: Analysis of Pedestrian Crashes in 30 Cities, Transportation Research Record 1773, pp. 56-68 17. VanHouten, R., JEL Malenfant, and D McCusker (200 1 ), Advance Yield Markings: Reducing Motor VehiclePedestrian Conflicts at Mult ibne Crosswalks with Uncontrolled Approach, Transportation Research Reeord 1 773, pp. 69 74. 18. Huang, Herman F and Michael J Cynecki (2000), Effects of Traffic Ca lming Measures on Pedestrian and Motorist Behavior, Transportation Research Record 1705, pp. 19. Schnell T., F. Aktan D.V. McGehee, M. Dvorak, J. Hunt, A. Reyes, and D Sorak (2001), Pedestrian Visibility under Automobile Low Beam Headlight JJiumination: With and Without Headlight Covers. Transportation Research Rec ord 1773, pp. 114. 20. Raymond, Paub D., and Richar d L. Knoblauch (2000), the Effects of Crosswalk Markings on Vehicle Speeds, TRB Paper No 00-1602. 21. Gordon, Stewart, and H. Douglas Robertson (1988), A Stt1dy of Driver Noncompliance with Traffic Signals, Transportation Research Record 1168. 22. Grayson, G B. (1987), Pedestrian Risk in Crossing Roads: West London Revi s i ted, Traffic Engineering and Control28, pp. 27-30. 23. TI>ompson, Lany A. (1996) Pedestrian Road Crossing Safely, CPL Bibliography 331/3321333, Journal of Planning Lit erature II, pp. 263-300. 24. Palamar!hy, Srinivas, Hani S. Mahmassani, and Randy B Mac hemehl ( 1994), Models of Pedestrian Crossing Behavior at Signalized Intersections, Recse.arch Report 1296 1 Center for Transporta tion Research, University of Texas at Atlstin. 25. Hamed, Mohammad M (2001), Modeling Pedestrians Crossing Behavior: Some Empirical Evidence, in Road Safety on Three Continents Proceedings of !he Conference, 19-21 Se. ptember 2001, Moscow, Russia. 26. Himanen, V., and R Kulmab ( 1988), An Application of Logit Models in Analyzing the Behavior of Pedestrians and Car Drivers on Pedestrian Crossings, Accident Analysis and Prevent ion 2, pp.l87 -197. 27. Chu, Xuchao (2002), A Nested Logit Model of Pedestrian Street Crossing Behavior, prepared for presenta t ion a t the 49\h Annual North American Meetings of the Regional Science Associa tion international, San Juan, Puerto Rico, November 13-16,2002 28. Chu, Xuehao (2002), A lternative Preference Methods, prepared for possible presentation a t !he 10 International Conference on Travel Behavior Research, Lucerne, August 2003. 29. Chu, Xuehao, and Michael R. Baltes (forthcoming), Measuring Pedestrian Level of Service for Mid-block Street Crossing: The Selection of Potential Determinants, Transponation Research Record. 30 Landis, Bruce W Vcnka t R. Vattikuti, Russell M. Ottenberg, DouglasS. Mcleod, and Martin Outt enplan (2001), Modeling !he Roadside Walking Environment A Pedestrian Level of Service, TRB Paper No. 0 1-0511. 31. Greene, William H. (1990) Econometric Analysis, MacmiUan Pub lishing Company, New York, NY. 13
32. Koppelman, FrankS and Chieh-Hua Wen (1998), Alternative Nested Logit Models: Structure, Properties and Estimation, Transportation Research B, 32, pp 289-298. 33. Hunt, Gary L. (2000 ) Alternative Nested Logit Model Structures and the Special Case of Partial Degeneracy, Journal of Regional Science, 40, pp. 89-113. 34. Hensher, David A., and William H. Greene (2002), Specification and Estimat i on of the Nested Logit Model: Alternative Nonnalizations, 36, pp. 1 J 7. 35. National Center for Bicycling and Walking (2002), Educa t ion and Public Awareness Campaigns, http://www.bikefed.orgleducation and public awareness carnpaigns.htm 36. Ben-Akiva, Moshe, and Steven R. Lerman (1985) Discrete Choice Analysis: Theory and Application to Travel Demand, the MIT Press Cambridge, Massachusetts. 37. National Cooperative Highway Research Program (2002), Mullimodal Arterial Level of Service Projeet3-70 FY 2003, http://www4.trb.orgltrb/crp .nsf/NCHRP+proj eets. 14