Trip Distribution
In t h is ch a pt er , th e m ech a n ics of t h e secon d crim e t r a vel dem a n d m odeling st a ge -
t r ip d ist r ibu t ion - is expla in ed. T rip distribu tion is a m odel of th e num ber of tr ips th at
occu r bet ween ea ch or igin zon e a n d ea ch des t in a t ion zon e. It u ses th e p red ict ed nu m ber of
t r ip s or igin a t in g in ea ch or igin zon e (t r ip pr odu ct ion m odel) a n d t h e pr edict ed n u m ber of
t r ips en din g in ea ch d est in a t ion zon e (tr ip a t t r a ction m odel). Th u s, t r ip d ist r ibu t ion is a
m odel of t r a vel bet ween zon es - tr ips or link s. The m odeled t r ip dist r ibut ion ca n t h en be
com p a r ed t o t h e a ct u a l d is t r ibu t ion t o s ee wh et h er t h e m od el p r od u ces a r ea s on a ble
a p pr oxim a t ion .
Th e o r e t i c a l B a c k g ro u n d
Th e t h eor et ical ba ckgr oun d beh in d t h e t r ip d ist r ibu t ion m odu le is pr es en t ed firs t .
Next , t h e specific pr ocedu r es a n d t est s a r e discu ss ed wit h t h e m odel bein g illus t r a t ed wit h
da t a from Ba lt im ore Coun t y.
Lo g ic o f t h e Mo d e l
14.1
Th e t r ips on t h e dia gon a l ar e intra-zonal t r ips , t r ips t h a t origin a t e a n d en d in t h e
sam e zone. Again, to use th e exam ple below, th ere were 37 tr ips th at both originat ed and
en ded in zon e 1, 53 t r ip s t h a t bot h or igin a t ed a n d en ded in zon e 2, a n d so for t h .
M N
G Oi = G Dj (14.1)
I=1 j=1
F igu r e 14.1
14.2
equ a t ion 14.1 h olds . Th is wa s d one in t h e t r ip gen er a t ion st a ge. Bu t , it is es sen t ia l t h a t
t h e st ep sh ou ld h a ve been com plet ed for t h e t r ip dist r ibut ion t o be implem en t ed.
Ob se rv e d a n d P r e di ct e d D is tri bu ti on s
Th e obser ved d ist r ibu t ion is calcula t ed by sim ply en u m er a t in g t h e n u m ber of t r ips
by each origin -dest ina t ion com bina t ion . This is somet imes ca lled a t rip lin k (or t r ip pa ir ).
Th er e a r e n ot a n y special st a t ist ics oth er t h a n a sim ple t wo-wa y cr oss-clas sifica t ion t a ble.
Th e secon d d ist r ibu t ion , however , is a m odel of t h e t r ip d ist r ibu t ion m a t r ix. This is
u su a lly ca lled th e predicted distr ibut ion. In th is case, a simple model is used to
a ppr oxim a t e t h e a ct u a l emp irical dist r ibut ion . The t r ips origin a t ing in ea ch or igin zon e a r e
a llocat ed t o dest in a t ion zon es u su a lly on t h e ba sis of bein g dir ectly pr opor t ion a l t o
a t t r a ct ion s a n d in ver sely pr opor t ion a l t o cost s (or im peda n ce).
Anoth er way to th ink of th is is th at th e actu al distr ibut ion of crime tr ips is complex,
r epr esen t in g a la r ge n u m ber of differ en t decis ion s on t h e pa r t of offen der s wh o do n ot
n ecess a r ily use t h e sa m e decision logic. Th e m odel, on t h e oth er h a n d, is a sim ple
a llocat ion on t h e ba sis of t h r ee or , som et im es, four va r ia bles . Alm ost by definit ion , it will
be m u ch sim pler t h a n t h e r ea l dist r ibut ion . Still, t h e sim ple m odel ca n oft en ca pt u r e t h e
m ost imp or t a n t ch a r a ct er ist ics of t h e a ct u a l dist r ibut ion . Hen ce, modelin g ca n be an
14.3
extr em ely u seful an a lyt ica l exer cise t h a t a llows oth er t ypes of ques t ion s t o be as ked t h a t
a r e n ot possible wit h ju st t h e obs er ved dis t r ibu t ion .
Th e G ra v i ty Mo d e l
M1 M2
F = g ----------------- (14.2)
D2
S o c i a l Ap p li c a ti o n s o f t h e G ra v i ty Co n c e p t
Pi Pj
I ij = "----------------- (14.3)
D ij2
14.4
Pi Pj
I ij = K ----------------- (14.4)
D ij
Tri ps as In te ra ct io n s
" P i $ Aj
T ij = --------------- (14.5)
D ij
Over time, the concept ha s been genera lized an d applied to ma ny different types of
t r a vel beha vior . For exam ple, Hu ff (1963) a pplied t h e con cept t o r et a il t r a de bet ween
zon es in a n u r ba n a r ea u sin g t h e gen er a l for m of
S j8
Aij = " ----------------- (14.6)
D ijD
14.5
a t t r a ction in t h e r egion. Aga in , it can be gen er a lized t o all zon es by, first , est im a t in g t h e
t ota l t r ips gen er a t ed from one zone, i, t o an oth er zone, j,
P i 8 AjJ
T ij = " ----------------- (14.7)
D ijD
Secon d, t h e t ota l n u m ber of t r ips gen er a t ed by a locat ion , I, t o all dest in a t ion s is
obt a in ed by s u m m in g over a ll d es t in a t ion locat ion s, j:
" P i 8 $ AjJ
T ij = ------------ (14.9)
D ijD
wh er e " is a con st a n t for t h e pr odu ction s, P i 8 , but $ is a con st a n t for t h e a t t r a ction s, AjJ .
Th is t ype of fu n ct ion is ca lled a d ou ble con st ra in t m odel beca u se t h e equa t ion h a s t o be
con s t r a in ed by t h e n u m ber of u n it s in bot h t h e or igin a n d d es t in a t ion loca t ion s ; t h a t is , t h e
s um of P i over a ll loca t ion s m u st be equ a l t o t h e t ot a l n u m ber of p r od u ct ion s wh ile t h e s u m
of Aj over a ll loca t ion s m u st be equ a l t o t h e t ot a l n u m ber of a t t r a ct ion s. Ad ju st m en t s a r e
u su a lly requ ir ed t o ha ve t h e su m of in dividu a l pr odu ctions a n d a t t r a ctions equ a l t h e t ota ls
(u su a lly est im a t ed in depen den t ly).
N e g a t i v e E x p o n e n t i a l D i s t an c e F u n c t i o n
14.6
m axim ization, th e lat ter pa rt of th e equat ion is a negative exponent ial fun ction t ha t h as a
m a xim u m va lu e of 1 (i.e., e -0 = 1) (Wilson , 1970). Th is h a s t h e a dva n t a ge of m a k in g t h e
equ a t ion m or e st a ble for in t er a ct ion s bet ween loca t ion s t h a t a r e clos e t oget h er . F or
exa m ple, Cliff an d H a gget t (1988) us ed a n ega t ive expon en t ia l gra vit y-t ype m odel t o
descr ibe t h e diffu sion of m ea sle s in t o th e U n it ed St a t es from Ca n a da a n d Mexico. It h a s
a lso been a r gu ed t h a t t h e n ega t ive expon en t ia l fu n ction gen er a lly gives a bet t er fit t o
u r ba n t r a vel pa t t er n s, p a r t icula r ly by a u t omobile (Foot, 1981; Boss a r d, 1993;). F igu r e 14.2
sh ows a t yp ica l n ega t ive exp on en t ia l fu n ct ion a n d on e r ecom m en ded for h om e-ba sed wor k
t r ip s by t h e Tr a n spor t a t ion Resea r ch Boa r d a s a defa u lt va lu e (NCH RP , 1995).
Tra v e l Im p e d a n c e
Trave l Time
It h a s been r ea lized, however, th a t dist a n ce is on ly a n a ppr oxim a t ion t o imp eda n ce.
In r ea l tr a vel, tr a vel t ime is a m u ch bet t er ind ica t or of t h e cost of t r a vel in t h a t tim e va r ies
by t h e t im e of da y, da y of week , an d ot h er fact ors . For exa m ple, t r a vel a cross t own in a n y
m et r opolita n a r ea is gener a lly a lot ea sier a t 3 in t h e m or n ing, sa y, t h a n a t t h e pea k
a ft er n oon r u sh per iod. The d iffer en ce in t r a vel t ime can var y as m u ch a s t wo-t o-t h r ee
t im es bet ween pea k a n d off-pea k h ou r s. U sin g on ly dis t a n ce, h owever , t h ese va r ia t ion s a r e
n ever picke d u p be cau se t h e dist a n ce bet ween locat ion s is in va r ia n t .
14.7
Figure 14.2:
0.0020
Relative impedance
0.0015
0.0010
0.0005
0.0000
0 4 8 12 16 20 24 28
Distance (miles)
gen er a lly be s elect ed over one t h a t t a k es longer even if th e first one is lon ger in dis t a n ce.
F or exam ple, it s been docu m en t ed t h a t people will ch a n ge wor k loca t ion s t h a t a r e fa r t h er
from t h eir h ome if t r a velin g t o th e n ew work locat ion t a k es less t im e (e.g., t r a velin g in t h e
opposit e dir ect ion t o th e bu lk of t r a ffic; Wa chs , Ta ylor, Levin e & On g, 1993).
Never t h eless, sh ift ing t o a n imp eda n ce fu n ct ion a llows a t r a vel m odel t o bet t er
r eplica t e a ct u a l t r a vel con dit ion s. Most t r a vel d em a n d m odels u sed by t r a n spor t a t ion
pla n n er s u se a n im peda n ce fu n ct ion , r a t h er t h a n a dis t a n ce fu n ct ion . 1 Dist a n ce wou ld only
be m ea n ingfu l if t h e st a n da r ds wer e inva r ian t with r espect t o t ime (e.g., a m odel ca lcu lat ed
over a n en t ir e yea r , 24 h ou r s a da y). As will be dem on st r a t ed in ch a pt er 16 on n et wor k
a ssign m en t , a t r a vel t im e ca lcu la t ion lea ds t o a ver y d iffer en t n et wor k a lloca t ion t h a n a
dist a n ce ca lcu lat ion . For exam ple, if dist a n ce is us ed a s a n imp eda n ce var iable, th en t h e
sh ort est t r ips will r a r ely t a k e t h e fre ewa ys becau se t r a vel t o an d fr om a free wa y usu a lly
ma kes a t rip longer th an a direct r out e between a n origin and a destinat ion. But as m ost
people u n der st a n d, ta kin g a freewa y to tr a vel a sizeable dist a n ce is us u a lly a lot quicker
t h a n t r a ver sin g a n u r ba n a r t er ia l sys t em wit h m a n y t r a ffic light s, s t op sign s, crossin g
pedest r ia n s, cr oss t r a ffic fr om p a r k in g lots a n d s h oppin g m a lls, a n d ot h er u r ba n obst a cles.
Toda y, t h e u se of dis t a n ce in t r a vel d em a n d m odelin g h a s vir t u a lly been dr opped by m ost
tr an sport at ion plann ers.
Tr a v e l Co s t
14.9
differen t t r a vel m ode in gett ing fr om an air por t t o a h ot el on a t r ip; t h e for m er will
pr obably t a k e a t a xi wh er ea s t h e la t t er will pr obably t a k e a bu s or t r a in (if a va ila ble).
In a ddit ion t o r ea l cost s a r e per ceived cost s. For t r a n sit u ser s, pa r t icu lar ly, th ese
per ceived cost s a ffect t h e ea se a n d t ime of t r a vel. One of t h e st a n da r d qu est ion s in t r a vel
su r veys for t r a n sit u ser s is t h e t ime it t a kes t o wa lk from t h eir h om e t o t h e n ea r est bus
st op or in t r a -ur ba n r a il syst em (if ava ila ble) an d fr om t h e la st t r a n sit st op t o th eir fin a l
destinat ion; th e longer it t ak es to access the tr an sit system, th e less likely an individua l
will u se it . Similar ly, tr a n sfers bet ween bus es or t r a ins decrea se t h e likelih ood of t r a vel by
t h a t m ode, a lm ost in pr opor t ion t o th e n u m ber of t r a n sfer s. Th e r ea son is t h e difficult y in
get t in g ou t of on e bu s or t r a in a n d in t o a n ot h er . Bu t , t h e t im e bet ween t r a in s a dd s a n
imp licit t r a vel cost ; t h e lon ger t h e wa it bet ween bus es, th e less likely th a t m ode will be
u sed by t r a veler s. In sh ort , ea se of a ccess a n d con ven ien ce ar e posit ive in cent ives in u sin g
a m ode or a r out e wh ile difficult y in a ccessin g it, lack of con ven ien ce, an d even fear of bein g
vuln er a ble to cr ime will decr ea se t h e likelih ood of u sin g th a t m ode or r ou t e. 2
An ot h er per cep t u a l com pon en t a ffect in g a lik ely ch oice is t h e r elia bilit y of t h e
t r a n sport a t ion m ode. Man y offen der s a r e poor a n d dont h a ve expen sive, well m a int a ined
veh icles. If th e veh icle is n ot ca pa ble of h igher sp eed s or is even like ly to brea k down wh ile
a n offen ce is bein g com m it t ed, t h a t veh icle is n ot lia ble t o be u sed in m a k in g a t r ip or t h e
ch oice of dest ina t ion m a y be alt er ed. It is well kn own t h a t m a n y offen der s st ea l veh icles
14.10
for u se in a cr im e. F ea r s a bou t n ot bein g iden t ified a r e clea r ly a m a jor fa ct or in t h ose
decis ion s, bu t t h e r elia bilit y of t h eir own veh icles m a y a ls o be a fa ct or .
Trave l Utility
Th e con cept of u t ilit y ma y be less u seful for cr ime a n a lysis t h a n for gener a l tr a vel
beh a vior . For on e t h ing, since th e con cept is individua l, it can on ly be ident ified by
in dividu a l su r veys (Domen cich a n d McFa dd en , 1975). For cr im e a n a lysis , t h is m a k es it
vir t u a lly imposs ible t o us e sin ce it is ver y difficult t o in t er view offend er s, a t lea st in t h e
Un it ed St a t es. In a ddit ion , t h e m a t h em a t ics r equ ir ed for a r t icu la t in g a u t ilit y fu n ct ion a r e
difficu lt sin ce u t ilit y fu n ct ion s h a ve a ver y com plex for m , u su a lly in volvin g t h e bin om ia l or
m u lt in omia l logit fun ction wit h a Weibu ll er r or t er m . In t h e n ext cha pt er , br ief me n t ion is
m a de of t h is t ype of m odel.
Im p e d a n c e F u n c t i on
Th u s, for a zon a l type m odel, we ca n leave t h e gra vit y fu n ct ion a s a gener a lized
imp eda n ce fu n ct ion . For t r a vel bet ween a n y on e zon e a n d a ll ot h er zon es, we ha ve:
14.11
T i = " P i 8 G (AjJ/I ij) (14.11)
Alterna tive Mode ls: Inte rven ing Oppo rtun ities
S j$
Aji = " ----------------- (14.13)
G(S k >) D ij 8
14.12
is t h a t t r a vel bet ween t wo loca t ion s becom es a com plex fu n ct ion of t h e spa t ial en vir on m en t
of n ea r by a r ea s a n d n ot ju st of t h e t wo loca t ion s.
Me t h o d o f E s ti m a t i o n
Th e Crim eS tat t r ip d ist r ibu t ion m odel im plem en t s equ a t ion 14.12. Th e specific
det a ils a r e discu ssed below, but t h e model is iter a t ive. The st eps a r e as follows:
1.Depend ing on wh et h er a sin gly con st r a ined or doubly con st r a ined m odel is to be
est im a t ed, it st a r t s wit h a in it ia l gu ess of t h e va lu es for " or $ (or both for a dou bly
const ra ined model). Table 14.1 illust ra tes th e thr ee models.
Ta ble 14 .1
Th r e e Me t h o d s o f Co n s t r a i n i n g t h e G ra v i t y Mo d e l
S in g le c on s tra in t
Cons t r a in origin s
Ti j = " P i 8 A jJ I i j
Ti j = P i 8 $ AjJ I i j
D o ub le c on s tra in t
Ti j = " P i 8 $ AjJ I i j
14.13
2. Th e r out in e pr oceed s t o est im a t e t h e va lu e for ea ch cell in t h e origin -
dest in a t ion m a t r ix (see figu r e 14.1 a bove) u sin g t h e exist in g est im a t es for "
an d $.
Th e Crim eS tat t r ip dist r ibut ion m odu le in clud es one set u p screen a n d five rout ines
t h a t im plem en t t h e m odel:
1. Ca lc u l a te o b s e r v e d o r ig i n -d e s t i n a t io n d i s t ri b u t io n . If th er e is a file
a va ila ble w it h t h e coord in a t es for in divid u a l origin s a n d d es t in a t ions (e.g.,
a n a r r est r ecor d), t h is r ou t in e will ca lcu la t e t h e em pir ica l t r ip dis t r ibu t ion
m a t r ix;
14.14
2. Ca li b ra t e i m p e d a n c e f u n c t i o n . If th er e is a file a va ila ble wit h t h e
coor din a t es for in divid ua l or igin s a n d d es t in a t ion s, t h is rou t in e will ca libr a t e
a n em pir ica l im peda n ce fu n ct ion .
6. Co m p a r e o b s e r v e d a n d p re d i c t e d o ri g i n -d e s t i n a t io n t r ip le n g t h s .
Th is r out in e com pa r es t h e t r ip len gt h s fr om t h e obser ved (em pir ical) tr ip
distr ibut ion with t ha t pr edicted by th e model. Compa rison a re ma de
gr a ph ically, by a coin ciden ce ra t io, by t h e Kom ologorov-Sm ir n ov Two-Sa m ple
test, an d by a Chi squa re test on t he most frequent tr ip links.
D e s c r ib e Or ig i n -D e s t i n a ti o n Tr ip s
2. Origi n file . Th e origin file is a list of origin zon es wit h a sin gle point
r epr esen t ing t h e zon e (e.g., t h e cen t r oid). It m u st be inpu t a s eit h er t h e
pr im a r y or secon da r y file. Specify wh et h er t h e da t a file is t h e pr im a r y or
secon da r y file.
14.15
Figure 14.3:
3. D e s tin a tio n fi le . Th e dest in a t ion file is a list of dest in a t ion zon es wit h a
sin gle point r epr esen t in g t h e zone (e.g., t h e cent r oid). It m u st be in pu t a s
eit h er t h e pr im a r y or secon da r y file. Sp ecify wh et h er t h e da t a file is t h e
pr im a r y or secon da r y file.Sp ecify t h e dest in a t ion ID va r ia ble in t h e da t a file
(e.g., Cen su sTr a ct , Block , TAZ).
Note: all destinat ion IDs should be in t he origin zone file and m ust ha ve the sam e
na mes an d both should be cha ra cter (string) var iables.
Missi n g v a lu es. Identify wheth er th ere ar e any missing values for t hese
four fields (X an d Y coordina tes for both origin and destina tion locat ions). By
defau lt , Crim eS tat will ignore r ecor ds wit h bla n k va lu es in a n y of th e eligible
fields or r ecor ds wit h n on -n u m er ic valu es (e.g.,alph a n u m er ic ch a r a ct er s, #,
*). Bla n ks will a lways be exclud ed u n less t h e u ser select s <n on e>. Th er e
ar e 8 possible options:
14.17
An y ot h er n u m er ica l va lu e ca n be t r ea t ed a s a m is sin g va lu e by t yp in g it
(e.g., 99)Mu ltiple n u m er ica l va lues ca n be tr ea t ed a s m issin g va lues by
t yp in g t h em , s epa r a t in g ea ch by comm a s (e.g., 0, -1, 99, 9999, -99).
T yp e o f c oo r d i n a t e s ys t em a n d d a t a u n i t s. Th e coor d in a t e s ys t em a n d
dat a u nits a re listed for inform at ion. If th e coordina tes ar e in longitu des and
la t it u des, t h en a sp h er ical syst em is bein g used a n d d a t a u n it s will
a u t oma t ically be decim a l degr ees . If th e coord in a t e syst em is p r oject ed (e.g.,
St a t e P la n e, Un iver sa l Tr a n sver se Mer cat or U TM), th en da t a u n it s cou ld
be eit h er in feet (e.g., St a t e P la n e) or m et er s (e.g., U TM .).
Note : ea ch r ecor d is a u n ique origin -dest ina t ion com bina t ion . Ther e
a r e M x N r ecor ds wh er e M is t h e n u m ber of origin zon es (in clud in g
th e extern al zone) an d N is the nu mber of destinat ion zones.
14.18
1. An I D n u m b er fr om 1 t o K , w h er e K is t h e n u m b er of lin k s
ou t pu t (ID)
2. Th e fea t u r e pr efix (ODT)
3. The origin zone (ORIGIN)
4. Th e dest ina t ion zon e (DE ST)
5. Th e X coor din a t e for t h e or igin zon e (ORIGINX)
6. Th e Y coor din a t e for t h e or igin zon e (ORIGINY)
7. Th e X coor din a t e for t h e dest in a t ion zon e (DE STX)
8. Th e Y coor din a t e for t h e dest in a t ion zon e (DE STY)
9. The n u m ber of obser ved tr ips for t h a t com bina t ion (FRE Q)
10. Th e dist a n ce bet ween t h e or igin zon e a n d t h e dest ina t ion zon e.
The point graph ical out put for each object includes:
1. An I D n u m b er fr om 1 t o K , w h er e K is t h e n u m b er of lin k s
ou t pu t (ID)
2. Th e fea t u r e pr efix (P OIN TSODT)
3. The origin zone (ORIGIN)
4. Th e dest ina t ion zon e (DE ST)
5. Th e X coor din a t e for t h e or igin zon e (ORIGINX)
6. Th e Y coor din a t e for t h e or igin zon e (ORIGINY)
7. Th e X coor din a t e for t h e dest in a t ion zon e (DE STX)
8. Th e Y coor din a t e for t h e dest in a t ion zon e (DE STY)
9. The n u m ber of obser ved tr ips for t h a t com bina t ion (FRE Q)
F igu r e 14.4 sh ows t h e ou t pu t of t h e t op 1000 lin k s for t h e obs er ved t r ip dis t r ibu t ion
fr om a sa m ple of 41,974 r ecor ds for inciden t s com m itt ed bet ween 1993 an d 1997. The
zon a l model used wa s t h a t of t r a ffic a n a lysis zon es (TAZ). These wer e discuss ed in cha pt er
12. Beca u se t h er e a r e a la r ge n u m ber of lin k s (532 or igin zon es by 325 dest in a t ion zon es),
t h e t op 1000 we r e t a k en . Th es e a ccoun t ed for 19,615 cr im e t r ips (or 46.7% of all t r ips ). A
la r ger n u m ber of lin k s cou ld h a ve been select ed, bu t t h e m a p wou ld h a ve becom e m or e
clu t t er ed. Of t h e 19,615 t r ip s t h a t a r e dis pla yed in t h e m a p, 7,913 or 40.3% a r e in t r a -zon a l
t r ips . Th ese wer e out pu t by t h e r out in e a s p oin t s a n d h a ve been dis pla yed a s cir cles wit h
t h e size pr opor t ion a l t o t h e n u m ber of t r ip s. Th e r em a in in g 11,702 t r ip lin k s wer e ou t pu t
by t h e r ou t in e a s lin es a n d a r e dis pla yed wit h t h e t h ick n ess a n d st r en gt h of color of th e
lin e bein g pr opor t ion a l t o th e n u m ber of t r ips .
14.19
Figure 14.4:
Observed Baltimore County Crime Trips: 1993-1997
Top 1000 Links
All Crime Types
#
# #
Baltimore County #
#
# #
# 25 or less
#
# #
#
#
#
26 - 49
#
# #
50 - 74
# # #
#
#
#
# #
75 - 99
#
#
#
#
#
# # # #
100 or more
Top 1000 Intra-zonal observed trips
# # #
#
#
# # #
#
# # #
#
# #
#
Less than 50
#
#
# #
#
#
# #
#
#
#
# #
#
# # #
#
#
50 - 99
#
#
# # #
# #
#
#
100-14 9
# #
# # # # # #
#
#
# # # #
# #
# # # #
# # #
# # # # # #
#
#
# # # #
#
# #
# # #
# #
# # # # #
150-19 9
# # # # # #
# # #
#
#
# #
# # # #
# # #
# # #
#
#
# # # # # # # #
#
#
#
##
#
#
# #
# # #
#
#
# #
#
# City of Baltimore # # #
#
#
#
#
#
# #
#
#
#
#
#
#
#
#
#
200 or more
#
# # #
###
#
#
#
#
# #
#
# #
# #
#
##
#
# # #
# #
#
# #
#
# #
#
Baltimore County
#
# #
# #
#
# #
#
# #
#
#
#
City of Baltimore
#
# # # #
#
#
# # #
#
# #
# #
# #
# ##
# #
##
# #
# #
#
#
#
#
#
# #
# #
#
## #
#
# #
#
#
#
# #
#
#
#
#
0 10 20 Miles W E
S
Th er e a r e sever a l cha r a cte r ist ics of t h e t r ip p a t t er n t h a t sh ould be n ote d. F ir st , t h e
in t r a -zon a l t r ip s t en d t o con cen t r a t e on t h e ea s t er n p a r t of Ba lt im or e Cou n t y. Th is is a n
a r ea t h a t is r elat ively poor with a h igh n u m ber of pu blic h ou sin g project s. This s u ggest s
t h a t t h er e a r e a lot of in t r a -com m u n it y cr im es bein g com m it t ed in t h ese loca t ion s. Secon d,
t h e zon e-t o-zon e pa t t er n , on t h e ot h er h a n d, t en ds t o con cen t r a t e a t five differ en t loca t ion s
r ela t ively clos e t o bor der wit h t h e Cit y of Ba lt im or e. Th ese five loca t ion s a r e a ll m a jor
sh oppin g m a lls. Th ir d, t h e or igin s for t h ose t r ip s t o t h e sh oppin g m a ll t en d t o com e fr om
with in th e City of Baltimore. Four th , in general, th e locat ions with h igh int ra -zona l tr ips
do n ot h a ve a la r ge n u m ber of zon e-to-zon e t r ips . H owever , t h er e is on e except ion in t h e
sou t h west cor n er of t h e cou n t y.
In ot h er wor ds, t h e obser ved dist r ibut ion of cr ime t r ips is com plex, but with severa l
pa t t er n s bein g sh own. A lot of crim e t r ips occur over ver y sh ort dis t a n ces. Bu t t h er e is
a lso a con ver gen ce of ma n y crim e t r ips on m a jor sh oppin g m a lls in t h e Cou n t y.
Ca l ib ra t e Im p e d a n c e F u n c t i o n
Th e r ou t ine est ima t es a t r a vel dist a n ce fu n ct ion u sin g a one-dimen sion a l ker n el
den sit y m et h od. See t h e det a ils in ch a pt er 9. E ssen t ia lly, for ea ch r ecor d, t h e sepa r a t ion
bet ween t h e or igin loca t ion a n d t h e dest in a t ion loca t ion is ca lcu la t ed a n d is r epr esen t ed on
a dis t a n ce sca le. Th e m a xim u m im peda n ce is ca lcu la t ed a n d divid ed in t o a n u m ber of
in t er va ls ; t h e defa u lt is 100 equ a l s ized in t er va ls , bu t t h e u ser ca n m odify t h is . F or ea ch
im peda n ce poin t calcu la t ed, a one-dim en sion a l ker n el is overla id. F or ea ch in t er va l, t h e
valu es of a ll ker n els a r e su m m ed t o pr odu ce a sm oot h fu n ct ion of t r a vel imp eda n ce. The
r es u lt s a r e sa ved t o a file t h a t can be u se d for t h e origin -des t in a t ion m odel.
Not e, h owever , t h a t t h is is a n em pir ica l d is t r ibu t ion a n d r epr esen t s t h e com bin a t ion
of or igin s , d es t in a t ion s , a n d cos t s. It is n ot n eces sa r ily a good des cr ip t ion of t h e im p ed a n ce
(cos t ) fu n ct ion by it s elf. Ma n y of t h e m a t h em a t ica l fu n ct ion s pr od u ce a bet t er fit t h a n t h e
em pir ica l im peda n ce fu n ct ion .
14.21
file. If t h e file t yp e is ASCII, s elect t h e t yp e of da t a sepa r a t or (com m a ,
semicolon, space, ta b) an d th e num ber of column s.
Missi n g v a lu es. Identify wheth er th ere ar e any missing values for t hese
four fields (X an d Y coordina tes for both origin and destina tion locat ions). By
defau lt , Crim eS tat will ignore r ecor ds wit h bla n k va lu es in a n y of th e eligible
fields or r ecor ds wit h n on -n u m er ic valu es (e.g.,alph a n u m er ic ch a r a ct er s, #,
*). Bla n ks will a lways be exclud ed u n less t h e u ser select s <n on e>. Th er e
ar e 8 possible options:
An y ot h er n u m er ica l va lu e ca n be t r ea t ed a s a m is sin g va lu e by t yp in g it
(e.g., 99)Mu ltiple n u m er ica l va lues ca n be tr ea t ed a s m issin g va lues by
t yp in g t h em , s epa r a t in g ea ch by comm a s (e.g., 0, -1, 99, 9999, -99).
14.22
1. Th e n o r m a l ker n el over la ys a t h r ee-dim en sion a l n or m a l
dist r ibut ion over ea ch poin t t h a t t h en exten ds over t h e a r ea
defin ed by th e r efer en ce file. This is t h e defau lt k er n el
fun ction .
2. Th e u n iform ker n el overla ys a u n ifor m fu n ct ion (disk ) over
ea ch poin t t h a t on ly exten ds for a lim ited dist a n ce.
3. Th e qua rtic kern el overlays a qua rt ic fun ction (inverse
sph er e) over ea ch poin t t h a t on ly exten ds for a lim ited
dis t a n ce.
4. Th e t ri a n g u l a r ker n el overla ys a t h r ee-dim en sion a l t r ia n gle
(cone) over ea ch p oint t h a t only ext en ds for a lim it ed dis t a n ce.
5. Th e n e g a t i v e e x p o n e n t i a l ker n el overla ys a t h r ee
dim en sion a l nega t ive expon en t ial fu n ct ion ('salt sh a ker ') over
ea ch poin t t h a t on ly ext en ds for a lim it ed dis t a n ce
Ch oi ce o f ban dw idth . The k er n els a r e a pplied t o a lim ited sea r ch dist a n ce,
ca lled 'ba n dwid th '. F or t h e n or m a l k er n el, ba n dwid th is th e s ta n da r d
d evia t ion of t h e n or m a l d is t r ibu t ion . F or t h e u n ifor m , qu a r t ic, t r ia n gu la r
a n d n egat ive expon en t ial k er n els, ban dwidt h is t h e r a diu s of a circle defin ed
by th e su r fa ce. For a ll t ypes, lar ger ba n dwidt h will pr odu ce sm oot h er
den sit y est ima t es a n d both a da pt ive a n d fixed ban dwidt h int er vals can be
selected.
14.23
1. By t h e n u m ber of bin s. Th e m a xim u m dis t a n ce ca lcula t ed is
divided by t h e n u m ber of sp ecified bin s. Th is is t h e defau lt
wit h 100 bin s. Th e u ser can cha n ge t h e n u m ber of bin s.
3. Output (Areal) Units . Specify th e den sit y un its a s point s per m ile,
n a u t ica l mile, foot , kilom et er , or m et er . The defau lt is point s per m ile.
4. Calcu late De ns ities o r Probab ilities . Th e den sit y est im a t e for ea ch cell
can be ca lcula t ed in one of t h r ee w a ys:
Select wh et h er a bsolu t e den sit ies, rela t ive den sit ies, or pr oba bilit ies
ar e to be out put for each cell. The defau lt is probabilities.
An exa m ple of a n em pir ical im peda n ce fu n ction from Ba lt im ore Coun t y is seen in
figu r e 14.5. Th is wa s der ived fr om t h e 41,974 in cid en t s in wh ich bot h t h e cr im e loca t ion
a n d t h e offen der s or igin loca t ion wer e kn own . As seen , t h e fu n ct ion look s sim ila r t o a
14.24
Figure 14.5:
0.12
0.1
Proportion of All Trips
0.08
0.06
0.04
0.02
0
0 5 10 15 20 25 30 35 40
Distance (miles)
n egat ive expon en t ial fu n ct ion . But t h er e is a lit t le h itch a r ou n d 3 m iles wher e t h e t r a vel
lik elih ood in cr ea ses, r a t h er t h a n decr ea se. Th is cou ld possibly be du e t o t h e Cit y of
Ba lt im or e bor der wh ich a bu t s m u ch of t h e s ou t h er n pa r t of t h e Cou n t y.
S e tu p o f Ori gi n -D e st in a ti on Mo d e l
14.26
Figure 14.6:
Or igi ns. Specify th e expon en t for t h e pr edict ed origin s. The d efa u lt is 1.0.
U se a n a l r ea d y-c a l i b r a t ed d i st a n ce fu n ct i on . I f a t r a vel d is t a n ce
fu n ct ion h a s a lr ea dy been ca libr a t ed (see 'Ca libr a t e im peda n ce fu n ct ion '
a bove), t h e file ca n be dir ect ly in pu t in t o t h e r ou t in e. Th e u ser select s t h e
n a m e of t h e a lr ea dy-calibr a t ed t r a vel dist a n ce fu n ction . Crim eS tat reads
dba se 'dbf', ASCII t ext 't xt', an d ASCII d a t a 'dat ' files.
Ma t h e m a t i ca l fu n ct i on s
14.28
f(d ij) = " +$*S ij (14.14)
wh er e f(d ij) is t h e lik elih ood t h a t t h e offen der will t r a vel fr om zon e ii t o a
pa r t icula r loca t ion, j, S ij is t h e separation in dist a n ce, tim e or cost bet ween
t h e offen der s r es ide n ce, I, a n d locat ion j, " is a slop e coefficien t wh ich
defines the fall off in distan ce, an d $ is a const an t. It would be expected th at
t h e coefficien t $ would h a ve a n ega t ive sign sin ce th e lik elih ood s h ould
declin e wit h sepa r a t ion . Th e u ser m u st pr ovide va lu es for A an d $. The
d efa u lt for A is 10 a n d for $ is -1. Wh en t h e fu n ct ion r ea ch es 0 (t h e X axis),
t h e r out in e a u t oma t ically s u bs t it u t es a 0 for t h e fun ction . F igu r e 14.7
illu st r a t es t h is fu n ct ion .
-$*S ij
f(d ij) = "*e (14.15)
wh er e f(d ij) is t h e likelih ood t h a t t h e offen der will t r a vel fr om a n or igin zon e,
ii, t o a d es t in a t ion zon e, j, S ij is t h e sepa r a t ion bet ween t h e or igin zon e a n d
t h e des t in a t ion zon e, e is t h e ba se of t h e n a t u r a l logar it h m , " is t h e
coefficient a n d $ is a n expon en t of e . Th e u ser in pu t s va lu es for " - t h e
coefficien t , a n d $ - t h e exp on en t . Th e defa u lt for " is 10 a n d for $ is 1.
(S ij - Mean D)
Zij = ------------------- (14.16)
Fd
1 -0.5*Zij2
f(d ij) = " * -------------------- * e (14.17)
Fd * SQRT(2 B)
14.29
Figure 14.7:
6%
4%
Impedance
2%
0%
0 4 8 12
Distance (miles)
Linear
Figure 14.8:
6%
4%
Impedance
2%
0%
0 4 8 12
Distance (miles)
Negative Exponential
wh er e f(d ij) is t h e likelih ood t h a t t h e offen der will com m it a cr ime a t a
pa r t icula r loca t ion, I (defin ed h er e a s t h e cen t er of a gr id cell), S ij is t h e
sepa r a t ion bet ween ea ch origins zone a n d d est in a t ion zon e, Mea n D is t h e
m ea n dis t a n ce inpu t by t h e u se r , Fd is t h e st a n da r d d evia t ion of dist a n ces, e
is th e base of th e nat ur al logar ithm , an d " is a coefficient . The user input s
va lu es for Mea n D, Fd , a n d ". The defau lt va lues a r e 1 for ea ch of t h ese
par am eters.
1 -[ ln (S 2 ij)-Mean D ]2 /2 *Fd 2
f(d ij) = "* ---------------------------- *e
S 2 ij * Fd * SQRT(2 B) (14.18)
14.32
Figure 14.9:
4%
Impedance
2%
0%
0 4 8 12
Distance (miles)
Normal
Figure 14.10:
30%
25%
20%
Impedance
15%
10%
5%
0%
0 4 8 12
Distance (miles)
Figure 14.11:
10%
6%
Impedance
2%
0 4 8 12
-2%
Distance (miles)
5. Fine Tun ing Expon en ts . In a ddit ion , for ea ch fu n ct ion , exp on en t s for t h e
a t t r a ct ion a n d pr odu ct ion t er m s ca n a dju st ed. Th is a llows a fin e t u n in g of
t h e im peda n ce fu n ct ion t o be t t er fit t h e em pir ica l d is t r ibu t ion .
1. Dista nce (miles, na ut ical miles, feet, kilomet ers, and m eters)
2. Travel time (minu tes, hour s)
3. Speed (m iles per h ou r , kilom et er s per h ou r ). Speed is t h en con vert ed
in t o t r a vel t im e, in m in u t es .
4. Gen er a l t r a vel cost s (u n specifie d u n it s).
14.36
9. Mo d e l Co n s tra in ts . In calibr a t in g a m odel, t h e r out in e m u st con st r a in
eit h er t h e or igin s or t h e d es t in a t ion s (s in gle con s t r a in t ) or con s t r a in bot h t h e
origin s a n d t h e dest in a t ion s (dou ble con st r a in t ). In t h e la t t er cas e, it is a n
it er a t ive s olu t ion . Th e d efa u lt is to con st r a in des t in a t ion s a s it is as su m ed
t h a t t h e dest in a t ion t ot a ls (t h e n u m ber of crim es occur r in g in ea ch zon e) a r e
pr obably m ore a ccu r a t e t h a n t h e n u m ber of crim es origin a t in g in ea ch zon e. .
Sp ecify t h e t ype of cons t r a in t for t h e m odel.
F i tt i n g th e I m p e d a n c e F u n c t i o n
Th e O ri g in -D e s t i n a ti o n Mo d e l
14.37
specified on t h e set u p pa ge. Secon d, t h e coefficien t s a n d exp on en t s a r e a pplied t o t h e
pr edict ed origin s a n d des t ina t ion s r esu ltin g in a pr edict ed t r ip dist r ibut ion . Beca u se t h ese
t wo st eps a r e sequ en t ia l, t h ey can n ot be r u n sim u lt a n eously.
Ca li b ra t e Or ig i n -D e s t i n a t io n Mo d e l.
Ap p ly P r e d i c t e d Or ig i n -D e s t i n a t io n Mo d e l
3. As s u m e d Co o r d in a t e s f or Ex t e r n a l Zo n e . In or der t o m odel t r ip s fr om
t h e ext er n a l zon e (t r ip s fr om ou t sid e t h e st u dy a r ea ), s pecify coor din a t es for
t h is zon e. These coor din a t es will be us ed in d r a wing lines from t h e pr edict ed
or igins t o t h e pr edict ed des t ina t ion s. Ther e a r e fou r ch oices:
14.38
(i.e., 95% or m or e), t h en it is ver y u n lik ely t h a t a n y of t h e t op t r ip lin k s will com e fr om t h e
ext er n a l zone.
S a v e To p Li n k s
14.39
5. Th e X coor din a t e for t h e or igin zon e (ORIGINX)
6. Th e Y coor din a t e for t h e or igin zon e (ORIGINY)
7. Th e X coor din a t e for t h e dest in a t ion zon e (DE STX)
8. Th e Y coor din a t e for t h e dest in a t ion zon e (DE STY)
9. The num ber of predicted tr ips for t ha t combinat ion (PREDTRIPS)
10. Th e dist a n ce bet ween t h e or igin zon e a n d t h e dest ina t ion zon e.
8. Save P oints
Com pa r in g t h e pr edict ed t r ip dis t r ibu t ion t o t h e obs er ved t r ip dis t r ibu t ion , s om e
sim ila r it ies a n d differ en ces a r e seen . F igu r e 14.13 com pa r es t h e t op 1000 zon e-t o-zon e
link s for t h e pr edict ed a n d observed dist r ibut ion s. The m odel ha s capt u r ed m a n y of t h e
14.40
Figure 14.12:
Predicted Baltimore County Crime Trips: 1993-1997
Top 1000 Links
All Crime Types
#
# #
Baltimore County #
#
# #
# 25 or less
#
# # #
#
#
#
26 - 49
#
# #
50 - 74
# # #
# #
#
#
#
#
75 - 99
#
#
#
#
#
# # #
#
#
100 or more
Top 1000 intra-zonal predicted trips
# # #
#
# # #
# #
# # #
#
#
#
# #
#
# #
#
#
# #
#
Less than 50
#
#
# #
# #
#
# # #
#
#
#
50 - 99
#
#
# #
#
# #
# #
# #
100-14 9
# #
# # # # #
#
#
# # # #
# #
#
# # # # #
# # # #
# # # # # #
#
#
# # # # #
# #
# # #
#
#
#
# # # # # #
150-19 9
# # # # # #
# # #
#
# # #
# # # #
# #
#
# # # # # # #
# # # #
#
#
#
#
# #
# # #
# #
# #
City of Baltimore # # #
#
#
#
# # #
#
#
#
# # #
#
#
# # #
#
# #
# #
#
#
#
# #
#
#
#
200 or more
#
#
#
# #
# #
#
# #
# # #
# #
#
#
# # # #
# # # #
#
# #
# #
# #
#
#
Baltimore County
#
# #
# #
#
# #
# #
#
#
#
#
City of Baltimore
#
#
# # #
# #
# #
# #
# #
#
# #
# # #
# #
#
#
# #
#
#
# # #
#
# # # #
# #
# #
#
# #
##
#
# # #
#
# #
#
# #
#
#
0 10 20 Miles W E
S
Figure 14.13:
Comparison of Predicted and Observed Crime Trips
1000 Top Zone-to-Zone Trips
All Crime Types
Baltimore County
N
0 10 20 Miles
W E
S
m a jor lin ks . For t h e five sh oppin g ma lls th a t r eceived m a n y actu a l cr ime t r ips, th e m odel
h a s ca pt u r ed t h e m a jor it y of t r ip s for t h r ee of t h em a n d som e t r ip s for a fou r t h . F or t h e
ma ll in t he sout heast corn er of th e coun ty, on t he oth er ha nd, the m odel ha s not a llocat ed a
lar ge nu m ber of t r ips. Simila r ly, for a zon e n ea r t h e west er n edge of t h e cou n t y, t h e m odel
ha s allocat ed more tr ips th an actu ally occur red.
In oth er wor ds , t h e fit bet ween t h e a ctu a l dis t r ibu t ion a n d t h e m odel is n ot p er fect .
Con sid er in g t h a t on ly 1000 of t h e 172,900 t r ip lin ks (532 or igin zon es x 325 dest in a t ion
zon es) ar e sh own, t h e m odel h a s s t ill done a r ea son a ble job of cap t u r in g t h e m a jor link s
Co m p a r in g Ob s e rv e d & P r e d i c t e d Tr ip s
1. Est imat e the par am eters of th e model an d apply th em to the calibrat ion da ta
set
14.43
Figure 14.14:
# #
#
#
Less than 50
#
Baltimore County # 50 - 99
#
#
#
100-149
#
#
#
# #
#
150-199
#
# #
# #
# #
#
# #
# #
# # #
# #
#
#
#
# #
# #
#
#
200 or more
# #
# #
#
#
# # #
# #
#
# #
# #
# # #
#
# #
#
Less than 50
#
# # # # #
#
#
# # # #
# # #
#
#
50 - 99
# # #
#
# # #
# # #
# # #
#
#
# # # #
# #
# #
# # #
100-149
# # #
# # # #
# ## # #
# #
#
# # # # #
# # # # #
#
# # #
# # # #
#
# #
# # # # #
# # # # # # #
# # #
#
#
# # #
# # # # #
# #
# #
# # #
#
150-199
#
# # # # # # # #
#
#
##
#
# #
#
# # #
#
# #
# #
City of Baltimore # # #
#
#
#
# # # #
## #
#
#
##
#
#
# #
#
# #
#
# # # #
#
##
# # # #
# # #
# # #
200 or more
#
#
#
#
#
# # ##
#
# # #
#
# #
##
# # #
#
##
#
# # #
# #
# # #
# # # #
#
# # #
# # #
#
# #
Baltimore County
# #
# #
#
# #
# # #
#
City of Baltimore
#
#
# # # #
#
#
# # # # #
# #
# #
# #
###
# #
##
# # # # #
# #
#
# #
#
#
#
#
#
# #
#
## # #
#
#
# #
# #
#
#
# #
#
#
# # # # #
# #
#
#
#
# #
N
0 10 20 Miles
W E
S
3. Com p a r e t h e t r ip len gt h s of t h e obs er ved a n d p r ed ict ed dis t r ibu t ion s u sin g
two tests:
A. Th e Coinciden ce Rat io
B. The Komolgorov-Smirn ov Two-sam ple Test
E s t im a t i n g Im p e d a n c e P a r a m e t e r s a n d E xp o n e n t s o f Gr a v it y Mo d e l
14.45
4. F or t h e pa r t icu la r m a t h em a t ica l fu n ct ion , s elect in it ia l gu esses for t h e
pa r a m et er s. For ea ch m a t h em a t ica l model, t wo or t h r ee differ en t
par am eters m ust be defined:
Co m p a r i n g In t r a -z o n a l Tr ip s
Th e firs t eva lu a t ion t est is t o comp a r e t h e per cent a ge of t r ips t h a t occur wit h in t h e
sa m e zon e - int r a -zon a l tr ips. The Tr a vel Model Im pr ovemen t P r ogra m m a n u a l in dica t es
t h a t in t r a -zon a l t r ip s sh ou ld r epr esen t t yp ica lly n o m or e t h a n 5% of a ll t r ip s for h om e-t o-
wor k t r ip s; t h a t is , com m u t in g t r ip s (F H WA, 1997, ch a pt er 4). H owever , given t h a t m ost
cr im e t r ip s a r e qu it e sh or t , t h e pr opor t ion of t r ip s t h a t a r e in t r a -zon a l is lia ble t o be m u ch
h igh er . In Ba lt im ore Coun t y, for exa m ple , 19.7% of a ll crim e t r ips wer e in t r a -zon a l.
Id ea lly, th e pr edicted m odel sh ould a lso ha ve 19.7% of all crim e t r ips bein g int r a -zon a l.
Th e Com pa r e obser ved an d pr edict ed t r ip lengt h s r ou t ine is discus sed below. The
r ou t ine out pu t s t h e n u m ber of t r ips t h a t a r e int r a -zon a l in bot h t h e obser ved an d pr edict ed
dis t r ibu t ion s. A good m odel s h ou ld pr odu ce a ppr oxim a t ely t h e sa m e n u m ber of in t r a -zon a l
t r ips in t h e pr edict ed dist r ibut ion a s wh a t a ct u a lly occu r r ed.
14.46
Illu st r a ti on
F or exam ple, in t h e Balt imore Coun t y model displayed in figu r e 14.12 a bove, t h er e
wer e 8,272 int r a -zon a l tr ips in t h e a ct u a l dist r ibut ion (ou t of 41,979). On t h e ot h er h a n d,
t h er e wer e on ly 5,428 in t r a -zon a l tr ips in t h e m odel. In oth er wor ds, t h e pr edict ed m odel
a ss ign ed fewer in t r a -zon a l t r ips t h a n a ctu a lly occur r ed.
Co m p a r e Tr ip Le n g t h D i s t r ib u t i o n
Th e secon d evalu a t ion t est in com pa r ing t h e obser ved with t h e pr edict ed
dist r ibut ion is a calculat ion of t h e t r ip lengt h dist r ibut ion (see st eps below). Beca u se t h e
t r ip d ist r ibu t ion m a t r ix will t ypically be ver y lar ge, m ost cell va lu es will be zer o. Ra r ely
will t h er e be en ough da t a t o cover a ll t h e cells a n d, even if t h er e wa s, t h e sk ewn ess in
cr im e dis t r ibu t ion s will lea ve m ost cells wit h n o da t a . F or exa m ple, for t h e Ba lt im or e
Coun t y m odel, wit h 532 or igin zon es a n d 32 5 de st in a t ion zon es, t h er e will be 172,900 cells
(325 x 532). The calibr a t ion da t a set h a d on ly 41,974 ca ses. Th u s, t h e n u m ber of cells is
m ore t h a n fou r t im es t h e sa m ple size a n d it is n ot p ossible t o fill a ll cells wit h a n u m ber .
14.47
of ea ch t r ip. It is m ea su r ed by th e a ct u a l len gth (or sepa r a t ion ) bet ween t wo zon es t imes
th e num ber of cases for t ha t zone pair. For example, in figur e 14.1, th ere were 15 tr ips
fr om zon e 1 to zon e 2 an d 7 t r ips in t h e opposite d irection (fr om zon e 2 to zon e 1). Lets
assu me th at th e dista nce between zone 1 and zone 2 is 1.5 miles. Thus, ther e are 22 trips
t h a t fall in t o a t r ip len gt h of 1.5 m iles (15 in t h e dir ection of zon e 1 t o zone 2 a n d 7 in t h e
dir ect ion of zon e 2 t o zon e 1).
If t r a vel t ime is u sed, t h e ca lcu lat ion s u ses t ime r a t h er t h a n dist a n ce. For exam ple,
if a vehicle was t r a veling 30 m iles per h ou r , th en it would t a ke 3 m inu t es t o cover 1.5 m iles
(1.5 miles 30 miles per h our = 0.05 hour s x 60 minu tes per hour = 3 minu tes). Thus,
t h er e a r e 22 t r ips t h a t fall in t o a t r ip len gt h of 3 m in u t es . A sim ila r logic would a pp ly t o
tr avel cost cat egories.
G r a p h i c a l f it
On ce t h e t r ip lengt h dist r ibut ion is ca lcu lat ed for bot h t h e obser ved an d pr edict ed
dis t r ibu t ions , it is p ossible t o comp a r e t h em . Crim eS tat ou t pu t s a gr a ph sh owin g t h e fit of
t h e t wo dist r ibut ion s. In gener a l, t h ey should be ver y close. An exam ina t ion of differ en ces
bet ween t h e dist r ibu t ion s ca n in dica t e a t wh a t t r ip len gt h s t h e m odel is fa iling. This
m igh t a llow t h e pa r a m et er s t o be a dju st ed in ord er t o imp r ove th e fit on t h e n ext it er a t ion.
E xa m ples will be given below of t h e gr a ph in g of t h e t wo dis t r ibu t ion s. Bu t , it s im por t a n t
t o come u p wit h a m odel in wh ich t h e t wo dist r ibu t ion s look sim ila r .
C oi n c i d e n c e r a t i o
K fO fP
Coin ciden ce = 3 m in [-------, ---------] (14.13)
k=1 FO FP
K fO fP
Tota l = 3 m a x[-------, ---------] (14.14)
k=1 FO FP
14.48
Coin cid en ce
Coin cid en ce r a t io = ------------------- (14.15)
Tot a l
1. E ssen t ially, t h e t wo dist r ibut ion s a r e br oken int o K bins (or int er vals). Tha t
is, t h e n u m ber of t r ips in ea ch bin is en u m er a t ed (see exa m ple a bove).
2. E a ch of t h e t wo dist r ibu t ion is con ver t ed int o a pr opor t ion by dividin g t h e bin
cou n t by t h e t ot a l n u m ber of t r ip s in t h e d is t r ibu t ion . Th is st ep is n ot
a bs olut ely es se n t ia l a s t h e t es t can be cond u cte d of th e r a w cou n t s.
H owever , by con ver t in g in t o pr opor t ion s, t h e t wo dis t r ibu t ion s a r e
st a n da r dized.
Th e Kom olgor ov-Sm ir n ov Two-Sa m ple Tes t is s im ila r t o th e coin ciden ce ra t io, bu t it
exa m in es t h e m a xim u m differ en ce across a ll bin s (Ka n ji, 1993). For ea ch d ist r ibu t ion, a
cum u la t ive su m is cr ea t ed. At ea ch in t er va l, t h e differ en ce bet ween t h e t wo cu m u la t ive
su m s is calcu lat ed. The m a xim u m differ en ce bet ween t h e t wo dist r ibut ion s is t a ken a s t h e
t est st a t ist ic:
D = | Oi - P i | (14.16)
14.49
Th er e a r e t a bles of cr itical values for t h e Kom olgor ov-Sm irn ov Two-Sa m ple Test
wh ich a r e a fun ction of t h e n u m ber of in t er va ls, K (Sm ir n ov, 1948; Ma ss ey, 1951; Siegel,
1956; Ka n ji, 1993).
Illu st r a ti on
Co m p a r i n g th e T ri p s o f t h e To p Li n k s
Th er efor e, sim ply com pa r in g t h e t r ip len gt h dis t r ibu t ion m a y not a dequ a t ely
capt ur e the top links. That is, on a verage a par ticular model ma y produce a good fit
bet ween t h e pr edict ed a n d obs er ved dis t r ibu t ion s, bu t m a y d o t h is by m in im izin g er r or
a cr oss t h e en t ir e m a t r ix of t r ip pa ir s wit h ou t n ecessa r ily m in im izin g t h e er r or for t h e t op
link s.
14.50
Figure 14.15:
Comparing Observed and Predicted Crime Trip Lengths
Empirical Impedance Function
40%
Percentage of all crime trips
30%
20%
10%
0%
0 5 10 16 21 26 31
Trip length (miles)
40%
Percentage of all crime trips
30%
20%
10%
0%
0 5 10 16 21 26 31
40%
Percentage of all crime trips
30%
20%
10%
0%
0 5 10 16 21 26 31
40%
Percentage of all crime trips
30%
20%
10%
0%
0 5 10 16 21 26 31
Us ing t h e obser ved (a ct u a l) lin ks a s t h e r efer en ce, th e t est ca lcu lat es:
K (P i O i )2
P seu do-ch i squa r e = G [-------------] (14.17)
i=1 Oi
F igur e 14.19 below sh ows t h e r a n k size d ist r ibu t ion of t h e Ba lt im ore Coun t y cr im e
t r ips. Notice h ow t h e dist r ibut ion is very sk ewed for t h e t op crim e t r ip pa irs , but declines
su bst a n t ia lly a ft er t h a t . Th a t is , t h e t op t r ip lin k (wh ich wa s a n in t r a -zon a l t r ip pa ir - zon e
654 to itself) h a d 278 t r ips. The s econ d t op link (a lso a n int r a -zon a l pa ir - zon e 714 to
it self) h a d 226 t r ip s. Th e t h ir d h a d 223; t h e fou r t h h a d 205; a n d so for t h . As m en t ion ed
a bove, t h e t op 1000 t r ip lin ks a ccou n t for a bou t 47% of a ll t h e t r ip in t h e m a t r ix, but t h e
firs t 176 a ccoun t pa ir s a ccoun t for h a lf of t h a t . In oth er words , if t h e t op 150 t o 200 t r ip
pa ir s a r e exa m in ed, t h e h ighest volu m e lin k s will be in clud ed a n d m ost of t h e sk ewn ess in
t h e dist r ibu t ion will be a ccoun t ed for . Th e r em a in in g dist r ibu t ion , wh ich is n ot fitt ed, will
be less skewed.
Illu st r a ti on
An illustr at ion of how compa ring the top link s can modify a t rip distr ibut ion m odel
ca n be given. The s a m e m odel a s sh own in figur e 14.12 wa s r u n . The ps eu do-Ch i squa r e
t est for t h e firs t 176 pa irs wa s 5,832 (r ou n din g-off t o t h e n ea r est int eger). H owever, by
m odifyin g th e m ea n dist a n ce of t h e lognorm a l fu n ct ion a lower Chi squ a r e valu e wa s
obt a ined . Aft er severa l it er a t ion s, t h e lowest Chi squ a r e valu e wa s obta ined for a m ea n
dista nce of 5.2 miles (P2 = 5,448).
14.55
Figure 14.19:
300
250
Number of observed crime trips
200
150
100
50
0
0 100 200 300 400 500 600 700 800 900 1000
Rank order
Aga in , t h e t op lin k s r epr esen t s on ly on e cr it er ia ou t of t h e t h r ee m en t ion ed. A good
m odel sh ou ld bala n ce a ll t h r ee of t h ese.
14.57
Ta ble 14.2
Mu l ti p le C ri t e ri a in S e l e c t i n g a D is t ri b u t io n F u n c t i o n
Logn or m a l fu n ct ion
St a n da r d deviat ion = 4.7 m iles
Coefficien t = 1
Or igin exponen t = 1.0
Dest ina t ion expon en t = 1.06
1. If t h e t r ip dis t r ibu t ion m a t r ix is h igh ly sk ewed (wh ich will occu r wit h m ost
crim e da t a set s), t h en it s ess en t ia l t h a t t h e t op lin k s be r eplicat ed close ly.
Th is w ould t a k e pr iorit y over t h e secon d cr it er ia , wh ich is m in im izin g t h e
er r or for t h e t r ip len gt h d is t r ibu t ion , a n d t h e t h ir d cr it er ia , wh ich is
minimizing th e err or in predicting int ra -zona l tr ips.
14.58
expon en t ial t h e slope would be a djus t ed n ext. Again , th e a im s h ou ld be to
obta in t h e h ighest Coinciden ce Rat io wit h out losin g t h e fit for t h e t op link s.
Notice tha t t his hiera rchy fits th e highest volume tr ip links first, then fits t he
overa ll t r ip len gt h dis t r ibu t ion, a n d fin a lly fits t h e n u m ber of in t r a -zon a l t r ips .
Illu st r a ti on
Ta ble 14.3
Mi n i m i z in g t h e S e c o n d Cr it e r ia i n S e l e c t in g a D i s t ri b u t io n F u n c t i o n
Logn or m a l fu n ct ion
Mea n dist a n ce = 5.2 m iles
St a n da r d Deviat ion = 4.6 m iles
Coefficien t = 1
Or igin exponen t = 1.0
Dest ina t ion expon en t = 1.06
14.59
Su bsequ en t t est s va r yin g t h e coefficien t of t h e logn or m a l fu n ct ion a n d t h e
exponen t s of th e origin a n d d es t in a t ion t er m s d id n ot a lt er t h es e va lu es . Cons equ en t ly, t h e
fina l model th a t wa s selected is list ed in t a ble 14.4.
Ta ble 14.14
Logn or m a l fu n ct ion
Mea n dist a n ce = 5.2 m iles
St a n da r d d evia t ion = 4.6
Coefficien t = 1
Or igin exponen t = 1.0
Dest ina t ion expon en t = 1.06
H owever, t h e fit wa s gener a lly bet t er . Figur e 14.20 sh ows t h e t op pr edict ed int er -
zon a l t r ip lin k s a n d com pa r es t h em t o t h e t op obs er ved lin k s wh ile figu r e 14.21 sh ows t h e
t op pr edicted in t r a -zon a l (local) tr ip lin k s a n d com pa r es t h em t o th e t op obser ved in t r a -
zon a l link s. Com pa r ing t h ese m a ps t o figu r e 14.12 a n d 14.13 (wh ich m a pped t h e t op 1000
lin k s, n ot t h e t op 176), th e fit is a bit bet t er for t h e m a jor link s, wh ich is wh a t we
opt imized. The fit is not p er fect ; it pr oba bly will n ever be. But , it is r ea sona bly close.
Im p le m e n ti ng th e Co m pa ri so n s in C r i m eS t a t
The mecha nics of condu cting the test s is fairly str aight forwa rd. The th ree tests a re
im plem en t ed in t h e Comp a r e Obser ved a n d P r edicted Tr ip Len gt h s r out in e on t h e la st
pa ge of t h e Trip d ist r ibut ion m odu le.
14.60
Figure 14.20:
Comparison of Predicted and Observed Crime Trips
Top Zone-to-Zone Trips from Optimized Model
All Crime Types
Baltimore County
N
0 10 20 Miles
W E
S
Figure 14.21:
Comparison of Predicted and Observed Crime Trips
Top Intra-zonalTrips from Optimized Model
#
All Crime Types
#
# #
#
#
Less than 50
#
Baltimore County # 50 - 99
#
#
#
100-149
#
#
#
# #
#
150-199
#
# #
# #
# #
#
# #
# #
# # #
# #
#
#
#
# #
# #
#
#
200 or more
# #
# #
#
#
# # #
# #
#
# #
# #
# # #
#
# #
#
Less than 50
#
# # # # #
#
#
# # # #
# # #
#
#
50 - 99
# # #
#
# #
#
#
# # #
# # #
#
#
# # # #
# #
# #
# # #
100-149
# # #
#
# # # #
# ## # #
# # #
# # # #
# # # # #
#
# # #
# # # #
#
# #
# # # # #
# # # # # # #
# # #
#
#
# # #
# # # # #
# #
# #
# # #
#
150-199
#
# # # # # # # #
#
#
##
#
# #
#
# # #
#
# #
# #
City of Baltimore # # #
#
#
#
# # # #
## #
#
#
##
#
#
# #
#
# #
#
# # # #
#
##
# # # #
# # #
# # #
200 or more
#
#
#
#
#
# # ##
#
# # #
#
# #
# # #
#
##
#
##
# # #
# #
#
# # #
#
# #
# # #
# # #
# #
# #
Baltimore County
# #
# #
# #
# # #
#
City of Baltimore
#
#
# # # #
#
#
# # # # #
# #
# #
# #
### #
# #
##
# # # # #
# #
#
# #
#
#
#
#
#
# #
#
## # #
#
#
# #
# # #
# #
##
#
# # # # #
# #
#
#
# #
#
#
N
0 10 20 Miles
W E
S
O b s er v e d t r i p fi l e
O b s er v e d n u m b er o f o r i g in -d e st i n a t i o n t r i p s
Specify th e var iable for t h e obser ved nu m ber of t r ips. Th e defa u lt n a m e is FRE Q.
O r i g _ID
O r i g _X
O r i g _Y
D es t _ID
Note: all destinat ion IDs should be in t he origin zone file and m ust ha ve the sam e
n a m es a n d t h e sa m e a s in t h e pr edict ed t r ip file if t h e t op link s a r e t o be com pa r ed.
D es t _X
D es t _Y
14.63
P r e d i c t e d t r i p fi l e
P r e d i c t e d n u m b er o f o r i g in -d e st i n a t i o n t r i p s
O r i g _ID
O r i g _X
O r i g _Y
D es t _ID
Note: all destinat ion IDs should be in t he origin zone file and m ust ha ve the sam e
n a m es a n d t h e sa m e a s in t h e obser ved tr ip file if t h e t op link s a r e t o be com pa r ed.
D es t _X
D es t _Y
14.64
S el ec t b in s
Fi xed n u m ber
C on s t a n t i n t e r v a l
C om p a r e t op l i n k s
S a ve com p a r ison
T a b l e ou t p u t
14.65
13. Th e cr itical D value for t h e Kom olgor ov-Sm irn ov two-sa m ple t est
14. Th e p-va lu e a ss ocia t ed wit h t h e D va lu e of Komolgor ov-Sm ir n ov t wo-sa m ple
t est r elat ive t o t h e cr itical D value.
15. Th e pseu do-Chi s qu a r e t est for t h e t op link s
a n d for ea ch bin :
Fi l e ou t p u t
Gr a p h
U t i li t y o f Ob s e r v e d Tr ip D i s tr ib u t i o n Ma p
Th is infor m a t ion by it self ca n be very us efu l for police. Two a pplica t ion s will be
discussed.
C r i m e p r e v en t i on effor t s
A ma jor a pplicat ion is using th e data shown in a t rip distr ibut ion m ap t o guide
en for cem en t effort s. F or exa m ple , in Ba lt im ore Coun t y, wit h t h e crim es occur r in g a t t h e
five sh oppin g m a lls, t h e origin loca t ions can be m ore ea sily s een . Th is h a s u t ilit y for police.
F ir st , t h e police int er ven e m ore e ffectively on t h e r out es lea din g fr om lik ely origin
loca t ion s. Th ey ca n pa t r ol t h os e r ou t es m or e h ea vily a n d, p er h a ps , in t er ven e m or e
14.66
fr equ en t ly. By u s in g t h e in for m a t ion fr om t h e t r ip dis t r ibu t ion a n a lys is , t h ey m a k e t h eir
en for cem en t effor t s sm a r t er . Secon d, t h ey ca n con du ct cr im e pr even t ion effor t s m or e
effectively. By kn owing t h e lik ely origin of offend er s, in t er ven t ion effort s in t h e origin
zon es m a y h ea d off som e of t h ese in cid en t s. P r ogr a m s su ch a s w eed -an d -seed a n d a ft er -
sch ool p r ogr a m s depen d on pr ovidin g a lt er n a t ive fa cilit ies for you t h , h opin g t o r edir ect
t h em t o m or e con st r u ct ive a ct ivities. Th ese facilities ca n be placed in loca t ion s wh er e
m a n y cr imes or igina t e.
Im p r o ve d J o u r n e y t o cr i m e a n a l y si s
A secon d a pplica t ion is in gues sin g th e likely or igin of a ser ial offen der . In cha pt er
9, t h eories of t r a vel beh a vior by a ser ia l offen der wa s d iscu ss ed. Th e r esu lt in g a n a lysis
(geogr a ph ic p r ofilin g, J ou r n ey t o cr im e a n a lysis ) u t ilized in for m a t ion on t h e dis t r ibu t ion of
incidents comm itted by th e offender. On t he oth er ha nd, the tr ip distr ibut ion pa tt ern seen
in figu r e 14.4 pr ovides a pr oba bilit y m a p of offen der loca t ion s a n d gives m or e in for m a t ion
t h a n wa s evid en t in t h e J ou r n ey t o crim e m odel. Th a t m odel a ssign ed a lik elih ood of th e
offen der livin g a t a loca t ion (t h e origin ) on t h e ba sis of t h e dist r ibu t ion of th e in ciden t s.
Th er e wa s n o ad dit ion a l infor m a t ion u sed a bout like ly or igin locat ion s. Th is t r ip
dis t r ibu t ion m a p, on t h e ot h er h a n d, p oin t s t o cer t a in zon es a s bein g t h e lik ely or igin for
offen s es com m it t ed a t t he m a jor d es t in a t ion loca t ion s . Th er e is m or e s t r u ct u r e in t h is
a n a lysis t h a n in t h e J ou r n ey t o cr im e logic.
U t i li t y o f P r e d i c t e d Tr ip D i s tr ib u t i o n Ma p
14.67
a pplied t o t h e for eca st set of or igin s a n d dest in a t ion s. Th is cou ld a llow a n exa m in a t ion of
possible ch a n ges in t h e cr im e dis t r ibu t ion (a ssu m in g t h a t t h e fu t u r e for eca st s a r e cor r ect
a n d t h a t t h e t r ip d ist r ibu t ion coefficien t s r em a in con st a n t ).
F our t h , an d fin a lly, a cr im e t r ip d ist r ibu t ion m odel is h elpful in developin g c rim e
th e ory . As in dica t ed in cha pt er 11, th e t h eor y of cr ime t r a vel h a s been very elemen t a r y
u p t o now. Th e pr im a r y focu s of a n a lysis h a s been only on t h e dest in a t ion s a n d on t h e t r ip
lengt h s a s m ea su r ed by dist a n ce t r a veled. A t r ip dist r ibut ion m odel, on t h e ot h er h a n d,
a n a lyzes bot h t r ip dest ina t ion s a n d t r ip or igins , an d can inclu de a m or e sophist ica t ed
m ea su r e of imp eda n ce t h a n sim ple dist a n ce. Beca u se t h e a n a lysis is con du ct ed over a
la r ger a r ea (a ju r is dict ion or a m et r op olit a n a r ea ), t h e h ier a r ch y of cr im e t r ip s ca n be
a n a lyzed sim u lta n eou sly an d t h e int er a ct ion bet ween or igins a n d des t ina t ion s can be
exa m in ed . In sh or t , a cr im e t r ip dis t ribu t ion m od el is a qu a n t u m lea p in sop his t ica t ion
a n d com plexit y comp a r ed t o th e u su a l J our n ey t o cr im e t ypes of m odels. H opefully, it will
gen er a t e even m ore s oph ist icat ed t ypes of m odels.
14.68
t h e n u m ber of t r ips flowing fr om ea ch origin zon e t o each dest in a t ion zon e. Th e m ode sp lit
model th en breaks t hese trips int o distinct t ra vel modes.
14.69
En d n ot e s fo r Ch ap te r 14
1. Dis t a n ce ca n be u sed a s a r ou gh a ppr oxim a t ion for im peda n ce, bu t is r a r ely a good
pr edict or of a ct u a l tr a vel beha vior . For exam ple, in t h e m ode sp lit m ode t h a t will be
dis cus sed in cha pt er 15, t h e dist a n ce bet ween a locat ion a n d t h e n ea r est bu s or r a il
r ou t e ca n be us ed t o quickly select t r ip pa irs t h a t m igh t t r a vel by tr a n sit . However ,
t h e a ct u a l pr edict ion m u st be bas ed on a n et wor k calculat ion of t r a vel t ime or t r a vel
cost in t r a vers ing t h e syst em .
14.70