Professional Documents
Culture Documents
Subjective Projection Method
Subjective Projection Method
Subjective Projection Method
Abstract
The state of the art in subjective methodology is summarized? In addition, viewpoints and insights into
the nature of forecasting and subjective methodologies are offered. All are based on the fundamental
idea that subjective data and processes are built into the very nature of forecasting. An examination of
the research in subjective methodologies is made and found to be pursuing a common strategy--a
strategy that parallels the multi-indicator development in objective methodology. This is termed the
"SIG-process" strategy. It is further examined and articulated and found consistent with past and
present research, expanding applications of subjective methodology, and future innovations incor-
porating computer and computer-related technology.
FRITZ R. S. DRESSLERiS a member of the staff of the University City Science Center, 3508 Market
Street, Philadelphia, Pa. 19104. He has an extensive background in industry and government and as
a consultant in technological forecasting and economic analysis. He is with the University City
Science Center, Philadelphia, Pennsylvania.
i Based on research performed in support of health and behavior projects at the University City
Science Center.
Copyright © 1972 by American Elsevier Publishing Company, Inc.
428 FRITZ R. S. DRESSLEK
emphasizes the measurement step. This is reasonable considering the different require-
ments and aims of each type of forecasting. Exploratory forecasting "explores" into
the future with intuitive or formal projections. The emphasis is into the future using
known data (measurements) as a point of departure. On the other hand, normative
forecasting is oriented to establishing a basis for resource allocations and takes as input
given goals. It searches for the underlying data patterns (relationships) between
allocated resources and generated technology, based, primarily, on past known
or assumed development histories. Projection, if any, is minimized in the rush to
apply highly-refined analytical techniques to define an " o p t i m u m " set of resource
allocations.
In summary, the anatomy of a forecast is well described by Wald as a "cognitive
model in operative use." That is, forecasting brings into play, in some measure, sub-
jective data established, in part, by observations in experiments and in non-experimental
real-life events.
1. Maximize the probability of being within the opinion area of subjective capability
(use experts in a group to moderate the influence of individual speculations).
2. Maximize group cooperative phenomena (use iterative procedures incorporating
information feedback).
3. Minimize the recognized negative influences of human interactions (retain some
form of anonymous response).
Two major thrusts to apply SIG-processes have evolved; one features controlled
face-to-face group interaction; the other allows no face-to-face group interaction.
Over the last twenty years, a number of studies [7] have detailed the shortcomings
of traditional group processes involving face-to-face interaction. Among the short-
comings, the most serious were found to be:
Given the above shortcomings, one simple approach would be to exclude face-to-
face interaction but still maintain the "group-of-experts rule." Such an approach has
developed. Although its foundations can be traced to work in the early 1950's, its name
--Delphi technique--did not appear in print until 1959 [8]. There is no face-to-face
interaction in the Delphi technique. Instead, formal questionnaires or other formal
communication channels, such as on-line computer communication, are used to convey
information between individuals and thus preserve a double anonymity. The respon-
dents are not known to one another, nor are any estimates associated with any respon-
dents.
The Delphi technique received a great boost in general interest with the publication
of a Rand Corporation Report by Helmer and Gordon [9] in 1964 on a study of
forecasting technological events. This publication was well timed and coincided with
a surge of iDterest in long-range exploratory forecasting, with, of course, a need for
an efficient and systematic technique to elicit expert opinion.
Researchers in subjective methodology are well aware of the shortcomings of
traditional group processes. They are also well aware of the powerful and beneficial
aspects of face-to-face interaction--the essence of the traditional group process. These
benefits include:
430 FRITZ R. S. DP,.ESSLF_~
1. Immediate data feedback
2. Sense of group "presence" and competition
3. Non-verbal communication
4. Clarification of misunderstandings.
The potential value of these benefits suggests that group processes contain the face-
to-face feature, but include procedures to diminish the negative aspects of face-to-face
confrontations. (Such attempts are also seen, for example, in panel discussions in-
corporating time-limits and a controlling moderator). Developments along this line
have been far more informal and difficult to identify than in the Delphi technique.
However, we can take the group process model reported by Delbecq and Van de Ven
[10] as representative of the state of the art using controlled face-to-face interaction.
It is termed the Program Planning Model (PPM) and has had wide application in
business, industry, government, and education. For example, Gustafson [11 ] has used
PPM as a conceptual framework to design a technique to elicit estimates of relationships
between diseases and indicator complexes. In this context PPM is transformed into
separate and unique application termed the Estimate-Talk-Estimate (E-T-E) technique.
Obviously, the ideal SIG-process would be one that contained all the benefits of
a face-to-face interaction while at the same time avoiding all negative affects. The trend
of research suggests that such an ideal is the goal towards which the various techniques
are moving. For example, Gordon reports [12] recent experiments with mini-Delphi
in which only a portion of the anonymity of Delphi is preserved. Turoff has developed
his Delphi conference [13] in which participants interact via an on-line computer using
teletype units or CRT's for message transmission. His idea is to automate the data
transmission and analysis normally used in Delphi and thus secure some of the speed
and communication advantages of direct interaction.
The trend of group-process developments as well as recent experimental and sug-
gested applications argue that SIG-Processes are a means to create effective com-
munications rather than a means to create "correct answers." This generalization is
in part the basis for adopting the collective term, SIG-process. This generalization is
also consistent with the overall movement of future research, a movement that presses
forward under the sheer need to obtain and use a broad range of information concerning
the future. It is not surprising then that SIG-processes have been used to generate ideas,
develop utility functions, collect data, and obtain value judgments--all of which have
no "correct answer."
l'ft. SIG-Processes
As shown, SIG-processes can be organized along two major thrusts, each producing
operational techniques: (1) no face-to-face interaction (e.g., Delphi) and (2) structured
face-to-face interaction (e.g., PPM and E-T-E). This suggests that the degree of
interaction between the individuals making estimates is an important criterion to
characterize SIG-processes. For example, brainstorming2 with its free face-to-face
interaction can be taken to represent maximum group interaction. On the other hand,
Delphi is highly structured and has no face-to-face interaction and thus can be taken
to represent minimum interaction. As a conceptual aid, consider an "interaction
continuum" ranging from maximum to minimum along which all possible SIG-
L INTERACTION d
I- -I
BRAINSTORMING CMAX) BASIC DELPHI(.MIN)
~ E-T -E
Z
/'-'~ _ MI N I ~
k.. ottp,, "+ _'r
DIRECT
I
DELPHI~
INODIRECT
DISC USSIONSIDISCUSSIONS
NO ET,"
+1
I
I
I
FTF~ INTERACTION INTERACTION
FTF~FACE'TO" FACE
Fig. 1. Simplifiedmappingof selectedSIG-processes.
I. An executive group that has been established to plan and manage the entire
process, and
2. An expert group composed of carefully chosen individuals whose opinions are
to be obtained and refined during the Delphi process.
The basic pattern of an executive group and a distinct expert group is found in all
SIG-processes. Suitability of application is assured via a situation-determined protocol,
thus the groups are combined in a wide variety of ways. However, all protocols seek
simply to create communication between the interacting "units of individual activity"
rather than overlay an impeding bureaucratic procedure.
Do SIG-processes "work?" Which protocol is best for a given situation ? At this
time, both questions have not been fully assessed. But, most researchers agree that
SIG-processes "work" in that they generally produce a narrowing of opinions and
a shifting of the median as the questioning proceeds.
Dalkey [14] recently verified these results in a series of detailed experiments using
Delphi. A large number of UCLA students were given a questionnaire using "almanac-
type" questions with known answers against which the group consensus could be
checked. Admittedly, questions of this sort differ from estimates about the future, but
they afforded an opportunity to observe some of the characteristics of group inter-
actions and behavior under controlled conditions.
Do Dalkey's results reflect the performance of Delphi when the questions under
study relate to the future? No hard answer can be given; however, in Campbell's
short-range forecast [15] of future economic conditions made by Delphi and "normal"
methods, the Delphi forecasts proved to be more accurate.
Ament [16] has compared the 1964 long-range forecasts made under Helmer and
Gordon to actuality and found that of the 22 events forecast to occur by 1970, 15
have occurred, 5 have not, and 2 are uncertain. Again, not a hard answer; but, one
that tends to support the Delphi technique as useful for making exploratory forecasts.
The accuracy question has proven very difficult to evaluate. For example, Ament
found it necessary to convene a special panel to assess if the events discussed in the
Helmer and Gordon Delphi had indeed occurred, and if so, to what degree.
Research has also focused on the protocol of SIG-processes to determine if some
underlying phenomenon is present. For example, Dalkey found that the estimates
obtained by Campbell followed the same lognormal distribution as those given to his
"almanac-type" questions. Martino [17] has analyzed the data obtained in the responses
to the Delphi conducted by North and Pyke at TRW. He found a similar distribution.
These findings support the belief that the Delphi technique is not a chaotic process
but is in fact well-behaved and can be described quantitatively.
What is known about the accuracy and protocol of other SIG-processes ? Gustafson
[18] has performed controlled evaluations of E-T-E and three other techniques con-
sidered applicable to making estimates (measurements) in medical diagnosis. He found
the E-T-E protocol superior to Delphi for his estimating task. Do these findings
invalidate the SIG-processes concept ? The answer seems to be no if the difference in
forecast situations is taken into account.
Gustafson used only eight questions. All dealt with the heights of individuals, male
and female, and were, therefore, all within the scope of the respondenrs daily ex-
periences. Gustafson was definitely making measurement-type estimates in the area
of opinion.
SUBJECTIVE METHODOLOGY IN FORECASTING 433
On the other hand, the majority of Delphi applications deal with full forecasts
(measure and projection) or, in the case of Dalkey, with questions far removed from
the respondent's daily experiences, thus moving into the area of speculation. For
example, Dalkey asked such questions as, "What was the total motor fuel consumption
in millions of gallons in 1965 in Massachusetts?"
One way to view the Dalkey and Gustafson results is to say that each showed the
necessity of matching protocols to the nature of the required estimates.
If an estimate falls within the scope of experience and training of a set of motivated
individuals, use the E-T-E protocol to obtain the best answer. If, on the other hand,
an estimate is remote to the experience and training of a set of individuals or is in
the nature of an exploratory forecast, then use the Delphi protocol to diminish in-
dividual speculations and avoid the risks of direct face-to-face interaction. In this case
the forecast situation is such that there is little or no gain in applying the positive
aspects of face-to-face interaction--so why risk exposure to the negative aspects with
such a meeting.
The accuracy and protocol of SIG-processes have not been fully evaluated. However,
the limited, controlled work that has been performed supports the view that SIG-
processes are a means to create effective communication between individuals to
promote a common strategy for collective problem solving rather than create "correct
answers."
i FORECAST METHODOLOGY i
I
I i
I,UD'ECT"E I
J
MULTI-
I I
ZERO
INDICATOR INDICATOR
'~ORCANIZED JUDGMENT I I
(SlG- PROCESSES) FORMULATION CLASSIFICATION
I I
I I I I
MULTI" ZERO MULTI" ZERO
INDICATOR INDICATOR INDICATOR INDICATOR
Classification and formulation methods may utilize the same data. But note how
they differ. The formulation method assumes that all items in a group "respond" the
same way to changes in any given indicator, while the classification method makes
allowance for the fact that an indicator may "affect" different items in different ways.
The formulation method assumes that there is no interaction between indicators, while
the classification method is not limited to this assumption.
Figure 2 contains the terms "Multi-Indicator" and "Zero-Indicator." These terms
are used, respectively, to designate a method based on the ability to organize a number
of indicators and a method representing the limiting case of no indicators.
The multi-indicator case is often called causal; referring to the fact that a number
of indicators (causal variables) have been identified and are used in some organized
way in the forecast. The zero-indicator case is often called simple; and uses only the
item of interest (dependent variable); usually, an analysis is carried out to identify
its regularities over time.
Causal methods go beyond the dependent variable to consider indicators which
cause changes in the dependent variable. An attempt is made to (1) relate certain
indicators to the dependent variable; (2) forecast the indicators (simple forecast); and
(3) infer the value of the dependent variable from changes in the indicators. The key
assumptions are that the indicators can themselves be forecast relatively accurately
in comparison to the dependent variable and that established relationships will remain
constant over time.
The range from multi-indicator to zero-indicator, that is, from causal to simple can
be viewed as a "continuum." The exact number of indicators, if any, that should be
SUBJECTIVE METHODOLOGY IN FORECASTING 435
used will depend on the nature of the situation, the available data, and the required
estimates.
Clearly, Fig. 2 serves as an organization of forecast (measurement and projection)
methods in its own right. For example, using Figure 2 we can relate forecast methods
as shown below: 4
Common Terminology
Figure 2 Terminology (State of the Art)
Subjective Multi-Indicator SIG-Processes*
Objective Formulation Multi-Indicator Econometric or Regression
Objective Formulation Zero-Indicator Time Series
Objective Classification Multi-Indicator Configuration
Objective Classification Zero-Indicator Association*
* These names developed during research at the University City Science
Center.
It is difficult to conceive that under any circumstances a single item can be isolated
in the "subjective capability" of individuals; therefore, the subjective/zero-indicator
method is considered impossible to obtain and is, therefore, not given in the above
listing. Similarly, all listed methods are difficult to conceive as isolated or "pure"
methods. In reality, all borrow some support from each other and emerge only as the
main theme in any applied situation. Nevertheless, the listed methods serve to display
an overall approach to forecasting and thus aid in establishing a subjective methodology
distinct from an objective methodology. Note that through the application of SIG-
processes, subjective methodology has developed organized and comprehensive tech-
niques (multi-indicator) similar to objective methodologies.
References
1 A number of sources can be given to outline the methods and techniques of forecasting, for example,
R. Ayres, Technological Forecasting, McGraw-Hill, New York (1969); (ed.), J. Bright, Technological
Forecasting for Industry and Government, Prentice-Hall, Englewood-Cliffs, New Jersey (1969).
2 A book distinguished by its internationalscope, E. Jantsch, TechnologicaIForecasting inPerspective,
OECD Publications, Washington, D.C. (1967).
3 H. Wald, "Forecasting on a Scientific Basis", proceedings of an international summer institute
held in Cuvia, Portugal, September 1966, Centro De Economia E Financas (1967).
4 E. Roberts, Technological Forecasting 1, 113 (1969).
5 For a summary of exploratory and normative forecasting, see, for example, Jantsch or Roberts,
ibid.
6 N. Dalkey, The Delphi Method: An Experimental Study of Group Opinion, The RAND Corporation,
Santa Monica, California (June 1969) (RM 5888-PR).
7 A number of sources are available, for example, S. Asch, "Effects of Group Pressure Upon the
Modification & Distortion of Judgments", Readings in Social Psychology, 3rd ed., Holt, Rinehart,
and Winston, London (1958); H. Kelly and J. Thibaut, "Experimental Studies of Group Problem
Solving and Process", in Handbook ofSocialPsychology, Vol. 11(G. Lindzey, ed.), Addison-Wesley,
Reading, Mass. (1964).
8 0 . Helmer and N. Rescher, On the Epistemology of the lnexact Sciences, The RAND Corporation,
Santa Monica, California, R-353 (February 1960); also in Management Science 6 (1959), pp.
25-52.
9 T. Gordon and O. Helmer, Report on a Long-Range Forecasting Study, The RAND Corporation,
Santa Monica, California, (AD 607-7777) September 1964.
10 A. C. Delbecq, and A. Van De Ven, A Group Process Model for Problem Identification andProgram
Planning, University of Wisconsin, Madison, Wisconsin (October 1970).
11 D. Gustafson, and R. Ludke, Information Manual on Computer-Aided Medical Diagnosis Study
for Ph.vsician Likelihood Ratio Estimators, University of Wisconsin, University General Hospital
Aided Medical Diagnosis Project, Madison, Wisconsin (February, 1970).
12 T. Gordon, The Current Methods o f Future Research, The Institute for the Future, Middletown,
Massachusetts, 1970.
13 M. Turoff, Technological Forecasting 3 (1972, to be published).
14 See Ref. 6, also: S. Brown, S. Cochran, and N. Dalkey, The Delphi Method, H: Structure of Experi-
ments, The RAND Corporation, Santa Monica, California (June 1969) (RM 5951 PR).
15 R. Campbell, Methodological Study o f the Utilization of Experts in Business Forecasting, Ph.D.
dissertation, University of California, Los Angeles, California (September 1966, unpublished).
16 R. Ament, "Comparison of Delphi Forecasting Studies in 1964 and 1969", Futures, March 1970.
17 J. Martino, Technological Forecasting 1, 355-358 (1970).
18 D. Gustafson, R. Shukla, A. Delbecq, and G. Walster, A Comparative Study of Differences in
Subjective Likelihood Estimates Made by Individuals, Interacting Groups, Delphi Groups and Nominal
Groups, The University of Wisconsin, Madison, Wisconsin (January 1971).
19 H. North, and L. Pyke, 1EEESpectrum 6, No. 1 (1969).
20 Turoff, ibid, see also, M. Turoff, "A Synopsis of Innovation by the Delphi Method", paper pre-
sented at the 1970 ORSA meeting held in Detroit, Michigan.
21 F. Dressier, Subjective Methodology, UniversityCity Science Center (internal report), Philadelphia,
Pa. (November 1970).
22 For a review of the literature and the method see, M. Turoff, An Alternative Approach to Cross
lmpact Analyses, copies can be obtained from M. Turoff, Office of Emergency Preparedness,
Office of the Assistant Director for Resource Analysis, Executive Office of the President, Washing-
ton, D.C.
23 V. De Santi et al., Genie, Office of Economic Opportunity, Executive Office of the President,
Washington, D.C. (June 1971).
24 Roberts, ibid.
25 M. Turoff, "Delphi and its Potential Impact on Information Systems", to be given at the Fall
Joint Computer Conference Proceedings (1971).
26 I. R. Hoos, A Critical Review of Systems Analysis: The California Experience (NASA Rpt. No.
CR-61350), prepared by University of California at Berkeley for NASA, Washington, D.C.
(December 1968).