Professional Documents
Culture Documents
TRANSPORTATION INTERACTIONS
By Violetta Cavalli-Sforza 1 and Leonard Ortolano, 2 M. ASCE
ABSTRACT: The Delphi method is used to predict the impacts of three alter-
native transportation programs in San Jose, California. Variables projected con-
cern both land use (e.g., number of single-family housing units) and choice of
transit mode. Forecasts are made for 1990 and 2000 for four spatial zones within
San Jose. Delphi panelists are individuals familiar with land use and transpor-
tation issues in the San Jose area. A preliminary questionnaire survey is used
to set general economic conditions and land use policies that serve as a context
for specific forecasts of land use-transportation interactions. Strengths and
weaknesses of the Delphi method in forecasting land use are assessed.
INTRODUCTION
The Delphi method is a technique for obtaining and refining the opin-
ions of a group of informed individuals, referred to collectively as the
Delphi panel. Individuals participating in a Delphi exercise are selected
because of their knowledge of the issues under study. A typical Delphi
panel has about 8 to 12 members.
The use of "expert judgment" in forecasting has been investigated by
Helmer and Rescher (8) and Amara and Lipinski (1). The layperson pos-
sesses no special knowledge of the forecasting problem, and therefore
his or her judgment can be no more than a sensible guess based on the
information provided. Unlike the layperson, an expert draws upon two
additional sources of information: past experience with events similar to
the one being studied, and in-depth knowledge about the event and the
context in which it occurs. The forecasts obtained from a group of ex-
perts are more likely to span a realistic range of possibilities and are
likely to be more firmly grounded than those resulting from a larger
group of laypersons.
325
1
iH
in neighboring cities,
LL
Downloaded from ascelibrary.org by East Carolina University on 03/27/16. Copyright ASCE. For personal use only; all rights reserved.
m = 10 IQR = 5 - 15
0
4 . Future monthly t
in neighboring t
H I i ipi
5 10 15 20
5 , Future retail and service
opportunities in neighboring
11.5 IQR = B - 15
1 1III 1i i I,
6 5 lb T 20
'. Future employment opportunity n = 16
in San Jose.
m = 15 IQR = 1 2 . 5 - IE I, l l 11 III
£ ib
, Future land use policies affecting
land outside the immediate impact
area of the proposed t r a n s i t corridt
Three states are used for this component, corresponding to whether de-
mand will increase, stay constant, or decrease. For components 1 through
7, the component states are "mutually exclusive." Using component 2
as an example, the demand for electronic goods produced in the region
cannot increase and decrease at the same time. For components 8 through
10, which concerns land use and transportation policies, states are not
mutually exclusive. For example, several land use policies can be em-
ployed simultaneously.
The states indicated for each component are all quite plausible, but
they do not exhaust the possibilities for any one component. Panelists
were encouraged to add other states as they saw fit and to rate the likeli-
hood of occurrence of each state. The likelihood rating employs a 5 point
scale with 1 as "very unlikely" and 5 as "very likely." As an example,
Fig. 3 shows the likelihood ratings for states used to characterize trans-
portation policies.
The use of both importance and likelihood ratings greatly simplified
the process of formulating alternative scenarios. Suppose the results in-
dicated that panelists were not in agreement about the likely state of a
particular component, but they all agreed that the component itself was
not very important. Based on the unimportance of the component, the
exercise of deriving alternative future scenarios could ignore the panel's
lack of certainty regarding likely states. On the other hand, suppose there
328
11
600 interchange, and Route 87 from
-4 '
« S.
1-280 to downtown San Jose.
2 •
0'
1 2 3 4 5
1*1 1
1980 and to 750+ buses by 1990 with 4
service to downtown San Jose and
BART. 2
0
| ,_
1 2 3 4 5
Downloaded from ascelibrary.org by East Carolina University on 03/27/16. Copyright ASCE. For personal use only; all rights reserved.
6
aA:
1 2 3
....
4
= 17
0
1
1
2 3 4 5
7
f) Allow for future use of cars and 6 n = 11
buses on Route 87.
4
2
0 , !
1 2 3 4 5
1,1 1 ,
Fremont. 4
0
1 2 3 4 5
1 2 3 4 5 6 T
I 1 1 1 ' , I number of responses
very unlikely uncertain likely very , I at each likelihood
unlikely likely £ I level
n •= total number of responses
The final outcome of the study was a set of forecasts of land use,
commute patterns, and choice of transit mode for each of the three
transportation investment programs. The forecasts were based on three
iterations (or rounds) of the Delphi questionnaire. For most of the vari-
ables included in the study, the behavior of the medians and other sta-
tistical descriptors of the distributions of responses during rounds 2 and
3 met conventional criteria for terminating a Delphi exercise. (These cri-
teria, which are based on convergence and stability of panelists' re-
sponses, are summarized in Ref. 4.) Time and budget constraints pre-
cluded the implementation of a fourth round.
An example of how panelists' responses changed over successive it-
erations of the Delphi questionnaire is given in Table 2. The consistent
inwards shifting of the upper and lower quartiles towards the median
is typical and desirable in the Delphi method. Since the interquartile
range is a measure of variability in the response distribution, a decreas-
ing value for the range indicates lowered variability, and therefore more
agreement (convergence) in the responses. The stability of the response
distribution is shown by the small changes in the median between rounds
(as a percentage of the median), and by the inward shifts in the upper
and lower quartiles from rounds 1 to 2 and rounds 2 to 3. The tendency
to successively smaller changes indicates that constant values are being
332
36 1 109
POPULATION
85 92 m 56 66
SINGLE FAMILY UNITS MULTI-FAMILY UNITS
71 81 m 57 65
COMMERCIAL EMPLOYMENT INDUSTRIAL EMPLOYMENT
which earlier, rapid growth left scattered vacant lots and abandoned
buildings, which were later marked for redevelopment. In contrast, Ed-
envale, which was once a center of industrial growth, has nearly run
out of land; and Almaden is too mountainous and residential to be suit-
able for substantial industrial development.
Forecasts of Commute Patterns and Modal Splits.—An important po-
tential impact of alternative transportation investments is their ability to
affect the relationship between where people live and where they work.
To examine this, Delphi panelists were asked to forecast the percentage
of study area employees living inside and outside the study area (the 2
percentages sum up to 100%). An illustrative result is for the auto al-
ternative (I) in 1990. In this case, Delphi panelists felt that 46% of study
area employees would reside inside the study area. The corresponding
forecasts for the bus and rail alternatives (II and III) were 2% higher.
These commute pattern forecasts could be used together with the modal
split forecasts described subsequently to predict the number of persons
using each transit mode. This exercise, which would utilize census data
for labor force participation rates, is beyond the scope of this study.
Modal split forecasts for work trips distinguished between two types
of trip-makers: (1) People residing and working within the area (study
area commuters); and (2) people either residing within and working out-
side the study area or residing outside and working within the study
area (regional commuters).
The modal choice forecasts were requested in terms of percent of total
home-work person trips for various transportation modes. Combina-
tions of modes were included because transit trips may require more
than one mode from trip origin to destination. By asking Delphi panel-
ists to predict directly the proportion of trips for distinct modes and mode
combinations, the forecasting task was simplified. Although the com-
binations employed were limited to two transit modes, Delphi panelists
assumed that before and after using a transit system passengers would
walk, use taxicabs, bicycles, and cars, or be driven between home and
work and the transit lines. The mode combinations employed in the
forecasts were assumed to exhaust the range of commonly selected
choices, and an "other mode" category was included for modes not oth-
erwise specified. The total number of mode combinations differed be-
tween the modal choice forecasts for study area commuters and for re-
gional commuters, reflecting the availability of more modes at the regional
level than at the study area level. An illustrative result is given in Fig.
7, which summarizes the modal split forecasts for study area commuters.
On the whole, the convergence and stability characteristics of the modal
split forecasts were similar to those of the land use forecasts. However,
the median was not used as a statistical descriptor of the response dis-
tributions for the modal choice forecasts. For any one forecast date and
transportation alternative, the percentages for each mode predicted by
335
individual panelists should add up to 100%, but the medians of the panel's
responses did not typically sum to 100%. Because of this, the mean rather
than the median was used as a descriptor of central tendency of the
response distributions for the modal choice predictions. The mean has
the property that averages of the responses summed over the mode cat-
egories add up to 100%.
337
local experts, who make predictions based on their familiarity with the
local area and a knowledge of key variables and issues affecting land
use.
It is difficult to say whether judgmental forecasts, and in particular
Delphi forecasts, can be viewed as valid estimates of future land use
and transportation impacts. Nonetheless, it must be emphasized that
the Delphi approach to group forecasting seeks to obtain and refine the
judgment of experts in the subject under investigation, arid expert opin-
ion is considered in most fields to be an important source of information.
The accuracy of the forecasts, and therefore the validity of the Delphi
method for this type of forecasting, can only be checked against reality
when it happens, and then only in light of the differences between the
basic assumptions used in the study and actual actions and events.
CONCLUSIONS
The San Jose case study experience provides a basis for judging the
utility of the Delphi approach in forecasting effects of infrastructure on
land use. The shortcomings of the approach relate principally to the
challenge of preparing a Delphi questionnaire that addresses the central
forecasting issues without being excessively long or complex. Most prob-
lems in implementing the San Jose case study stemmed from the ques-
tionnaire's length. On the positive side, the case study attained a degree
of success. It was conducted with a very limited budget; and the re-
quirements, in terms of data, staff expertise, and computing facilities,
were all modest. Indeed, many planning consultants or government
planning offices could easily carry out an exercise like the San Jose study.
Moreover, the sought-after land use projections were obtained, and they
seem reasonable by themselves and in relation to each other. The ac-
curacy of the forecasts must remain an unexplored question, at least un-
til 1990 when the forecasts can be analyzed in relation to what actually
occurs between now and then.
ACKNOWLEDGMENTS
The writers owe thanks to the 17 individuals who made the study
possible by serving as Delphi panelists. We are also indebted to our col-
leagues Jarir S. Dajani and Michael V. Russo for their contributions to
the San Jose case study. Anne Martin, a graduate student in the De-
partment of Civil Engineering at Stanford University, provided many
useful comments on a preliminary draft of this paper.
APPENDIX.—REFERENCES
1. Amara, R. C , and Lipinski, A. J., "Some Views on the Use of Expert Judg-
ment," Technological Forecasting and Social Change, Vol. 3, 1972, pp. 279-289.
338
the Delphi Method," Report No. 1PM-15, Dept. of Civ. Engrg., Stanford Univ.,
Stanford, Calif., Jan., 1982.
5. Dajani, J. S., and Ortolano, L., eds., "Methods of Forecasting the Reciprocal
Impacts of Infrastructure Development and Land-Use," Report No. JPM-11,
Dept. of Civ. Engrg., Stanford Univ., Stanford, Calif., June, 1979.
6. Davis, J. M., "Land Use Forecasting: A Delphi Approach," thesis presented
to the University of Georgia, at Athens, Ga., in 1975, in partial fulfillment
of the requirements for the degree of Doctor of Philosophy.
7. Ervin, O. L., "A Delphi Study of Regional Industrial Land-Use," Review of
Regional Studies, Vol. 7, No. 1, Spring, 1977, pp. 42-58.
8. Helmer, O., and Rescher, N., "On the Epistemology of the Inexact Sci-
ences," Management Science, Vol. 6, 1959, pp. 25-52.
9. Linstone, H. A., and Turoff, M., eds., The Delphi Method: Techniques and Ap-
plications, Addison-Wesley Publishing Co., Inc., Reading, Mass., 1975.
10. Mehta, S., and Dajani, J. S., "SILUS: The Stanford Infrastructure and Land
Use System," Report No. IPM-12, Dept. of Civ. Engrg., Stanford Univ., Stan-
ford, Calif., Jan., 1981.
339