ISPRS, Vol.34. Part 2W2, “Dynamic and Multi-Dimensional GIS". Bangkok, May 23-25, 2001 
I 
10 Jobs 
51 km 
O 
49 km 
CMI 
2 km 
10 Jobs 
Figure 2. An Example of Two Job Sites and Two Resident Workers Sites 
(Df ) is the weighted average of job proximity (Df ) of 
various wage groups: 
j wT 
Di =2* f -Jr D ‘>■ (4) 
8=I ' 
where g indexes individual wage groups (g=1,2,5), W| 9 
is the number of workers of wage group g, and Wj is the total 
number of workers in the TAZ. 
MEASURING JOB ACCESSIBILITY 
The job proximity index measures how far resident workers 
are from their suitable jobs, but does not indicate their true 
advantage of job access. Many factors may handicap one’s 
ability to reach the jobs. The following job accessibility 
measure intends to capture at least four factors: availability 
of vehicles, existing road network, congestion in high-density 
areas and competition for suitable jobs among workers. 
(1) Basic Model 
Hansen (1959) proposes a simple gravity model for 
accessibility: 
n 
7=1 
Hansen’s model only considers the supply of jobs. Shen 
(1998) improves the measurement by adding the demand 
side—job competition among workers. Job accessibility for 
workers of a particular wage group g is: 
*-s 
7=1 
v/ 
m 
where vj . 
k=l 
(5) 
The notations J, p and n have the same interpretations as in 
the job proximity index (1), and m is the number of resident 
worker locations. In this study, either m (number of worker 
sites) or n (number of job sites) is 882, the total number of 
TAZS, though some TAZS have no resident workers (W k =0) 
or no jobs (Jj=0). This new index re-scales Hansen’s 
accessibility to a job location j by the location’s job 
competition intensity (Vj 9 ), and V, 9 is this job location’s 
potential with regard to all workers competing for the jobs 
(W k , k=1, m). Since job accessibility emphasizes a 
worker’s ability of obtaining a job by overcoming various 
barriers, d is measured by travel time instead of air distance. 
Similar, g indexes wage groups (g=1, 2, 5), thus 
accessibility to jobs of a particular wage group (other than 
low-wage jobs), both workers (W k 9 ) and jobs (Jj 9 ) in equation 
(5) are limited to those within the wage group g. 
The job accessibility for low-wage workers (A ( 1 ) involves 
additional complexity, and will be discussed in a separate 
sub-section. Equation (5) is used to compute job 
accessibility for workers of the other four wage groups (A, 2 , 
A, 3 , A 4 and Aj 5 ). The larger the value of A, the better job 
accessibility the resident workers enjoy. 
(2) Estimating Travel Times 
One important task of implementing the job accessibility 
measure is to estimate travel times between TAZs. Let’s first 
focus on travel times by drove-alones who account for the 
majority of commuters. The CTPP Part 3 provides the actual 
travel times between TAZs where there were commuting 
trips in 1990 (i.e., 51,021 origin-destination TAZ pairs). The 
accessibility index in equation (5) uses travel times between 
all possible origin-destination TAZs (i.e., 882x882=777,924 
TAZ pairs). Estimating these travel times is implemented in 
three steps. Each step is an improvement over the previous 
one. 
The first step is to use GIS network modeling techniques to 
simulate the shortest travel times through a network 
composed of all roads (including neighborhood roads), 
where speed limits serve as travel impedance values. See 
Wang (2000) for details. 
One noticeable shortcoming in step 1 is that the simulated 
travel times are only between the centroids of TAZs, and 
assume that the travel time within a TAZ itself (intrazonal 
travel time) is 0. This is not a realistic assumption. The 
second step attempts to better define intrazonal travel times. 
Based on the 519 intrazonal commute trips in the study area, 
the average travel time within a TAZ is 11.3 minutes. 
Research reveals that the intrazonal travel times are not 
necessarily related to the area sizes of TAZs. The 11.3 
minutes may include the time a commuter spends on starting 
the car at the beginning of the trip and finding a parking 
space at the end of the trip or even the time spent on 
walking to the office. It is part of the “mental time” of a whole 
trip reported in the census survey forms. Adding the 
intrazonal time to the GIS-derived travel times between 
centroids yields the revised travel time estimations (d e ). 
Among the 51,021 existing commute trips in the study area, 
the average of d e is 25.63 minutes, very close to the average 
of real travel times 25.27 minutes. The two have a 
correlation coefficient of 0.632 (i.e., R 2 =0.399). Considering 
the large sample of observations (n=51,021), it is a 
remarkable good fit. 
The final step improves the estimation d e by taking into the 
account of congestions at both the residential (trip origin) 
and workplace TAZs (trip destination) 1 . Adding the density of 
resident workers at the origin TAZ (DEN wk ) and density of 
jobs at the destination (DEN jb ) (both in per km 2 ) as 
predictors, the regression yields a R 2 of 0.409 (n=51,021): 
d = 1.7152+0.7838 d e +0.000485DEN wk +0.0000415 DEN jb . 
(12.79) (183.39) (9.11) (28.52) 
All explanatory variables are significant at 0.0001 (t-values 
are in the parentheses). The coefficients of density variables 
may appear small. In fact, in the study area, the TAZ with the 
maximum residential density of 7,627 workers/km 2 adds 3.7 
minutes to the trip, and the TAZ with the maximum job 
density of 178,200 jobs/km 2 adds 7.4 minutes to the trip. 
Such additions are too significant to be neglected. 
This simple approach considers possible congestions only 
at the two ends of a trip. Considering traffic congestions 
during the whole trip would require the usage of more 
complicated traffic simulation packages which demand 
details of road network coding (e.g., lane capacity, traffic 
signal system, residential demographics and business 
types). That is not feasible for the research. 
292