You are on page 1of 17

TECHNOLOGICAL FORECASTING AND SOCIAL CHANGE 18, 217-233 (1980)

An Application of the Extended Cross Impact


Method to Generating Scenarios of
Social Change in Japan

M. ISHIKAWA, M. TODA, S. MORI, and Y. KAYA

ABSTRACT

Scenarios of Japanese society in 2000 are obtained by applying an extended correlational cross impact
analysis. Since the scope of the scenarios is extensive, a procedure is developed to generate the scenarios that
consists of a preprocessing of the events of interest and a step-by-step application of a cross impact method.
Three scenarios are described, based on a workshop where the procedure was applied.

Introduction
Japan has experienced tremendous social change since World War II. Its economy
has grown rapidly and the society has become highly industrialized. Its way of life has
been westernized, and people are now concerned with improving their quality of life,
despite the fact that economic growth has slowed in recent years. The drastic changes in
lifestyle have been accompanied by gradual changes in values and human relations. Faced
with these trends throughout the country, the government has recognized the urgent need
to advance integrated social planning, rather than economy oriented social planning as in
the past [I]. An important step toward realizing integrated planning is a forecast of social
change. This is an extremely difficult task, since it involves the totality of human be-
havior, influenced by a complex environment. Nonetheless, forecasting is important to
social planners, who must carry out the demanding task of planning while taking into
account the several aspects of the country.
The purpose of this paper is to present an approach to generating consistent scenarios
of social change based on judgmental forecasts, and to apply this approach to scenarios of
Japanese society in 2000.
Techniques for assessing societal changes are not as advanced as those for forecasting
short-term changes in the economy. Since techniques to construct reliable models for
forecasting are yet to be developed, human judgment is the major source of scenarios of
social change. Those who have been carefully observing a society can express their
opinions on the future trends of individual aspects of that society, but their statements may

M. ISHIKAWA is with the Electrotechnical Laboratory of the Ministry of International Trade and Industry,
lbaraki, Japan. M. TODA is with the International Institute for Advanced Study of Social Information Science,
Fujitsu Limited, Tokyo, Japan. S. MORI and Y. KAYA are in the Department of Electrical Engineering of The
University of Tokyo, Tokyo, Japan.

@ M. Ishikawa et al., 1981


218 M. ISHIKAWA, M. TODA, S. MORI, ANDY. KAYA

be neither clear-cut nor consistent. A cross impact technique [2, 31 is useful as an aid to
construct a consistent scenario based on these fragmented opinions. The analysis is par-
ticularly useful in bringing out the overall image latent in opinions concerning separate
aspects of social change. Scenarios thus constructed reveal clearly the differences among
individual opinions and hence provide a reliable basis for discussing highly uncertain
future trends.
The presently available cross impact methods for generating consistent scenarios
limit the number of events considered in order to reduce the workload of those providing
judgments. Because the scope of this study is extensive, to best cover a large number of
aspects of individual and social behavior, the number of events of interest exceeds the
capability of the cross impact methods. Our main goal is to develop a procedure for
constructing an extensive scenario. This procedure will consist of a clustering technique
and the step-by-step application of a cross impact method.
In the next section the cross impact method is described, and the procedure for the
development of scenarios is then described. Three scenarios of Japanese society in 2000
are presented as an outcome of this study, followed by a concluding discussion.

Outline of the Revised Cross Impact Method


A cross impact method is a technique for constructing a consistent system of prob-
abilities that takes into account the impact of relevant events based on experts’ judgments
of these probabilities. This section presents an outline of a cross impact method [2] that is
a revision of the method proposed by Duperrin and Godet [3].
First, it would be helpful to describe briefly the original method of Duperrin and
Godet. In this method experts estimate the following occurrence and conditional prob-
abilities for event ei (i = 1, . . , n).

p(i) = the probability of event ei occurring during a time interval

p(i /j) = the probability of event ei occurring during the time interval given the
occurrence of event ej in the interim.
p (i 17) = the probability of event et occurring during the time interval given the non-
occurrence of event e,i in the interim.

Since cross impacts are taken into account only incompletely in human judgments,
these probabilities generally contain inconsistencies as a complete probability set. The
essence of the method for resolving these inconsistencies is the introduction of the states
that specify the occurrence of the events

E, = fx,, . . .,x,0, k = 1,. .,N (1)

where N = 2” and

I if event ci occurs,
xi =
0 otherwise.
(
Each state (or, for simplicity, scenario) El; has an unknown probability rrk, where

%-,<2 0 (2)
and
C7rk = 1. (3)
SCENARIOS OF SOCIAL CHANGE IN JAPAN 219

Consistent occurrence and conditional probabilities p*(i), p * (i 1j), and p * (i iJ) can
be expressed in terms of the irk, with the constraints given by the following quadratic
criterion: Consistent (scenario) probabilities can be obtained by minimizing

J = c b(i) - p*(i)y

= I$ 7 [P(i ;JlPo’)-

I-C C {P<i1J)
i j
[l - PO'11- p*(i(j)
[l - p*G)]}2, (4)

where{p (i)}f {p(i lj)}, and{P(i PI} are the probabilities estimated by experts.
The method of Duperrin and Godet has two deficiencies [2].
(1) The quadratic form in (4) becomes positive semidefinite as the number of events
increases, and hence scenario probabilities rrIi can not be determined uniquely. To over-
come this difficulty, we proposed obtaining maximum and minimum probabilities for each
scenario by solving the linear programming problem

max (or min) rrI;, k = 1, . ., N, (5)


subject to consistency conditions.
Let the solution vector of scenario probabilities that maximizes (or minimizes) rr,; be
n(k) (or 5~~~)).
If(k) = (II,(k), . &P)), (6)
+li) = (?-r,(k),. . . n.v’“i’), (7)

Those scenarios having both large maximum and minimum probabilities &(li’ and ~k’~’
are considered likely. In the case where the minimum probabilities are all zero or very
small (this is likely to occur as the number of events becomes large), the following
“average” probability G& is used instead of the minimum probability:

Equation (8) indicates that the average probability vector (G,, . . , G,y) corresponds to
the centroid of N vertices IFi) (i = 1, . . N). Those scenarios having both large maxi-
mum and average probabilities IIk(h-) and rk- are selected as the likely scenarios.
(2) The number of variables N (=2n) in the quadratic and linear programming
problems and the accompanying computational effort increase rapidly as the number of
events rz becomes large. This makes it almost impossible to take into account simultane-
ously a large number of events, in practice more than 7. For this reason we divided the
original set of events into several subsets by a clustering method that considers similarity
and causality between events. The cross impact method was applied to each subset
sequentially, taking into account of the causality among the subsets. The clustering
method will be described in the next section, and the step-by-step application of the cross
impact analysis will be illustrated later.

Procedure for Constructing Scenarios


The procedure for constructing scenarios comprises selecting the relevant events,
determining the causality relationship between events, grouping these events using
220 M. ISHIKAWA, M. TODA, S. MORI, AND Y. KAYA

experts to assess their occurrence and conditional probabilities, and applying the cross
impact method of the preceding section to determine the likely scenarios.

EVENT SELECTION AND THE DETERMINATION OF THE


CAUSALITY RELATIONSHIPS AMONG EVENTS
At the outset events considered to be relevant to the problem of interest are collected
through discussions among experts. The causality relationships among these events are
evaluated through discussions to yield a causality matrix T. The (i, j) element of T, &I;,
denotes the strength of the impact relationship, both positive and negative, from event et
to event e.i and takes a value of 0, 1, or 2 in increasing order of strength.

CLUSTERING OF EVENTS
This section discusses a clustering method based on similarity data among the events.
Let C be a similarity matrix whose element Cij (i # j) denotes the similarity between
event ei and event ej. (Cij is here defined as/ Tij 1 + 1Tji 1.) C onventional clustering methods
[7], in which the distance between elements must be quantitative (e.g., a series of 1-O
data), are not appropriate for this case, because the similarity defined here is of a qualita-
tive nature. Hence other approaches are pursued here. A method for grouping events called
quantification theory IV by Hayashi [4] has frequently been used in sociological analysis
that deals with quantitative data. This method portrays the events on a plane under the
criterion that elements with high similarity be closely located. This approach, however,
has the drawback that most events tend to concentrate around the origin of the plane, and it
therefore might not be appropriate, especially for large problems.
In this paper a new method for grouping has been developed. This method systemati-
cally generates subsets that include events with large similarity and applies quan@xztion
theory III [4] to the structural relations among the subsets and their constituent events,
thereby grouping the events effectively. DCMPOS [5] is a similar technique for obtaining
these subsets, but it differs from the present approach in that it is heuristic and the subsets
obtained by DCMPOS are not exhaustive.
The first part of the method is to obtain all the subsets for which each average
connection ratiox (defined in (9)) is greater than a predetermined threshold r. (Henceforth
let the subsets be called closely connected.)

xi = : !jj cj’,/(mT - mi), (9)


j=l k=l
j#k

where mi is the size of subset i, that is, the number of events in the subset, and Cj, denotes
the (j, k) element of subset i. Equation (9) indicates that the average connection ratio Xi is
defined as the average of similarities between elements of subset i.
The number of possible subsets contained in the original events set is

,,c* + ,,ca + . . = ,,c,, = 2” - n, (10)


where n is the total number of events. It requires huge computational efforts to calculate
the connection ratios for all of these possible subsets unless the number of elements is
small.
We propose the following algorithm for obtaining these subsets.

Step I: Select all the subsets of size 2 each of whose connection ratio is greater than
a threshold r. This selection is not burdensome because the number of
possible subsets of size 2, ,,C’,, is fairly small. Set m = 3.
SCENARIOS OF SOCIAL CHANGE IN JAPAN 221

Step 2: Select closely connected subsets of size m by appending one element to those
of size m - 1 obtamed previously.

Step 3: If no subset is generated in step 2, go to step 4. Otherwise increase m by 1


and return to step 2.

Step 4: Delete subsets included in subsets of larger size from the set of closely
connected subsets obtained in steps 1 and 2.

The major advantage of this approach is that the amount of computational effort required
to generate all the closely connected subsets is very small as compared to the exhaustive
search needed to check all the possible subsets of size greater than 1. (The total number is
given by (l).) The validity of the above algorithm is guaranteed by the property that if a
subset of size m is closely connected, at least one of the subsets of size m - 1 obtained by
deleting one element from the original subset is closely connected.
The second part of the method is an application of quantification theory III to the
structural relation between closely connected subsets and their constituent events, thereby
obtaining scores and weights for subsets and events, respectively. A criterion for deter-
mining scores and weights is that subsets whose constituent events are similar have similar
scores and that events included in similar subsets have similar weights. This is realized by
maximizing a correlation coefficient between scores and weights. Let the score for subset
i hey, (i = 1, . . ., l), and the weight for event cj be xj (j = 1, . ., n).
The structural relation between the subsets and the events is represented by

if the subset i includes event ej,


&o’) = :,
otherwise.
(
$ = i: &(j), i = 1, . . ., 1, (11)

5’

4 = C 6(j), j = 1, . . .) n. (12)
i=I
where ni is the number of events in subset i, and dj is the number of subsets that include
event ej . By using these definitions, a maximization of the correlation coefficient between
x and y is formulated as the following eigenvalue problem:

G z = p2 z, (13)
where

(14)

(15)
The resulting weights and scores are given by

Xj = Zj/V&* j = 1, . .,n, (16)

Yj = i i: &(j)Xj 9 i = 1, . . ., 1, (17)
t j=l

where zj is the jth component of the eigenvector z corresponding to an eigenvalue.


A plot on a two-dimensional plane of weights xj corresponding to a pair of largest
eigenvalues is used to group events ej to yield subsets. This final step of grouping is
222 M. ISHIKAWA, M. TODA, S. MORI, AND Y. KAYA

usually performed by analysts’ observations of the resulting spatial representations, which


is similar to that of multidimensional scaling analysis.
Owing to the threshold r, events whose similarity with other events are small may not
be included in any subsets. In the present approach, these “isolated” events, if any, are
excluded from an application of quantification theory III. It should be noted that those
subsets that do not share events with other subsets form an independent eigenvector in the
eigenvalue problem (13). Needless to say, these isolated events and subsets contribute to
grouping a set of events.

LlNGUlSTlC ESTIMATION OF PROBABILITY


In an application of the cross impact method presented above, experts were requested
to provide one- and two-dimensional probabilities in numerical terms of the occurrence of
events. Since these events are nonrecurrent, in contrast to those evaluated by objective
probabilities (such as dice and playing-cards), it is not easy to assess the “subjective”
probabilities numerically. Moreover, the events are usually specified by linguistic ex-
pressions whose fuzziness is supplemented by those of the linguistic statements that
express experts’ images of the world. For these events a linguistic rating of probability,
such as that chosen in Table 1, is more appropriate, because it is agreeable to experts’
intuitions. For this reason we propose the following three-stage procedure: Establish a
relation between statements expressing likelihood and the numerical values of probability
by the use of the category rating method [6], select appropriate statements that yield less
variances in individuals’ interpretation, and rate the cross impact probabilities with the
help of the selected statements.
At the outset, statements expressing likelihood are selected and listed at random.
Table 1 presents these examples for 30 statements. Table 2 shows 17 objective prob-
abilities in descending order selected from examples using dice and playing-cards. In the
category rating method subjects are asked to which category (probability) each stimulus
(likelihood statement) belongs. In the procedure we present, however, the original method
is modified as follows. First, the subjects are permitted to assign a probability inter-
mediate between two adjacent values listed in Table 2. Second, a scale value for each

TABLE 1

Standard
deviation of
Average assessed assessed
probability probabilities

Almost certain to occur 0.973 0.024


Highly probable 0.848 0.093
Very likely 0.728 0.080
Likely 0.671 0.072
More likely than not 0.568 0.049
50-50 chance of occurrence 0.503 0.01 I
More unlikely than not 0.449 0.034
Unlikely 0.304 0.105
Very unlikely 0.220 0.104
Highly improbable 0.099 0.069
Almost certain not to occur 0.049 0.052
SCENARIOS OF SOCIAL CHANGE IN JAPAN 223

TABLE 2

0.981 A card drawn out of a deck of playing-cards is other than the ace of spades.
0.923 A card drawn out of a deck of playing-cards is other than an ace.
0.833 The upturned face of a tossed die is other than 1.
0.75 Neither upturned face of two tossed dice is even.
0.667 The upturned face of a tossed die is less than 5.
0.615 A card drawn out of a deck of playing-cards is less than 9.
0.583 The sum of the upturned faces of two tossed dice is greater than 6.
0.539 A card drawn out of a deck of playing-cards is less than 8.
0.5 The upturned face of a tossed die is even.
0.462 A card drawn out of a set of playing-cards is less than 7.
0.417 The sum of the upturned faces of two dice is less than 7.
0.385 A card drawn out of a deck of playing-cards is less than 6.
0.333 The upturned face of a tossed die is less than 3.
0.25 Both upturned faces of two dice are even.
0.167 The upturned face of a die is 1.
0.07 1 A card drawn out of a deck of playing-cards is the ace of spades.

probability statement is here defined as an average of corresponding probabilities, and not


of the corresponding integer weights for categories over subjects.
An experiment by 20 subjects determined the average probability and standard devia-
tion for each statement. Based on this result, 11 statements in Table 1 that had relatively
small standard deviation of the assessed probabilities were finally selected for use in the
cross impact method.

AN APPLICATION OF THE REVISED CROSS IMPACT METHOD


One remaining problem in applying the revised cross impact method described above
is how to determine probabilities where there are more than one participant. One solution
is to arrive at a consensus through extensive discussions among the participants. The
alternative is to allow differences of opinion to remain and to apply the cross impact
method for each participant separately.
In this study the second approach was adopted because of the difficulty in arriving at
a consensus among the participants. This difficulty may be attributed to the diversity in
both occupation and age of the participants.

Future Scenarios of Japanese Society


The procedure to generate scenarios we have described was applied to assess future
trends in Japanese society. The study was focused on the long-term forecast: Japanese
society in 2000. The workshop consisted of officials of the Economic Planning Agency
and systems analysts.

SELECTION OF EVENTS
Based on discussion among experts on future trends in Japanese society, 38 events
were specified after an aggregation and disaggregation of the topics discussed; a causality
matrix T of dimension 38 was filled by way of estimates of impact relations between a pair
of these events. The clustering method was applied to T, and some isolated events that are
comparatively less important were deleted to yield a set of 20 events in order to reduce the
224 M. ISHIKAWA, M. TODA, S. MORI, ANDY. KAYA

workload of the workshop participants. The events to be forecast are listed in Table 3. The
clustering method also grouped the set into four subsets. The subsets are

Group V: events concerning personal value judgments and ways of life (six events);

Grozrp W: events concerning welfare and the community (six events);

Group L: events concerning employment (five events); and

GI-oup E: events concerning education (three events).

The causality matrix T for the 20 grouped events is shown in Figure I.


In forecasting these events, the economy was assumed to expand at the present
growth rate, and no drastic change of world situations was assumed.

TABLE 3
Selected Events of Interest Describing Social Change in Japan by 2000
Group V: Events concerning personal value judgments and way of life
VI. Chunge in the socicrl norm: Presently dominant future-oriented norm, which aims at growth,
development and industrialization, continues to give way to a present-oriented norm that stresses
short-term well-being.
V2. Populurixztion of indi~?dualism: People conform less to social custom, and individualistic behavior
is much more evident (e.g., diversified job selection. as opposed to the presently popular orienta-
tion toward white-collar workers).
V3. C~CUI~Crn t/w workethic: Work is considered a mean& of the worker’s human development as well
as a means of obtaining income, and efforts are made to humanize labor relations (e.g.. flexible
work hours).
V4. Extended holiduys for workers become customary as in the United States and Europe.
VS. Adult educntion becomes popular and the number of people benefitting increases markedly.
V6. Decentruliwtion qf culture: Presently uniform culture dominated by the mass culture of large cities
is diversified as local culture flourishes.

Group W: Events concerning welfare and community


WI. Equal ernp/o.wnent opportunityfor women is significantly advanced.
W2. Socinl sen?ce employment opportunities increase markedly owing to public policy.
W3. Volunteer activities organize and cooperate with specialists to establish social service systems in the
area of health care and welfare.
w4. Equul opportunitiesjbr the physicu/Ly handicapped are established.
W5. Trtrditio,lu/,filmi/~ life i$ freqrtmt/y disrupted by divorce, career conflict, and the generation gap.

Group L: Events concerning employment


LI. Undrsiruble working conditiom remain, compensated by high income (e.g., hazardous jobs and
excessive overtime work).
L2. Substuntiul muncrgement eJ@rts to eliminute occupational diseases and mechanical injuries are
made.
L3. Shorttrge of labor .upp/y becomes an important social problem for jobs accompanied by risks to the
worker’s lives and for socially undesirable jobs (e.g., sanitation and sewage treatment).
L4. Split conong types of work becomes more evident as an elite is overloaded with responsibility,
leaving the majority to simpler tasks.
L5. Decrease of intergenerutional social mobility results in more rigid social stratification.

Group E: Events concerning education


El. Rising percentuge of‘college gruduates exceeds 45%.
E2. Incornr urzd rducution become decreasingly well correlated.
E3. Frustrution crmong the high/y educated but underemployed is aggravated by a scarcity of prestigeous
jobs.
SCENARIOS OF SOCIAL CHANGE IN JAPAN 2

W5 1 1 -2

W6 2 -2

Ll -2

L2 -2

L3 1 2 2

L4 1 1
-
L5 1 2

El 2 2

E2

E3 ‘1

Fig. 1. Causality matrix among events.

The assessed major impact relations among the subsets and among events within a
subset are illustrated in Figure 2.

CROSS IMPACT ANALYSIS


The cross impact analysis was performed groupwise, so that participants’ judgments
on conditional probabilities were requested for not more than six events at time.
The results of the step-by-step cross impact analysis are summarized in Table 4.
Likely scenarios for three participants groups are presented in the table, together with the
consistent scenario probabilities.
Since Figure 2 indicates that group V, which is related to personal value judgments,
has an impact on all the other subsets, group V was analyzed first. A consensus was not
attained among probability judgments, and the opinions of the participants were classified
into two groups. The probability data provided by the two groups yielded two most likely
scenarios, which may be denoted (1, 1, 1, 1, 1, 1) and (0, 1, 0, 0, 1, 1) for events (VI,
V2, V3, V4, VS, V6) (see Table 4).
226 M. ISHIKAWA, M. TODA, S. MORI, AND Y. KAYA

., . . . ...... . . . . . . .
Group L .-.
t :.:

*223%
Employmen L5 L 2 !.,
: :.,

;L4 Ll L3 ‘i
‘...._.__...... :’
-. .. ..._,..... ..... . .. ....

Group "

Value and Judge- :

Group W

Welfare and Communities

-.. . . ..__..............-‘.-.-’

Group E

Education

Fig. 2. Impact relations among events and event groups. See Table 3 for the description of events.
Single and double arrows denote assessed impact relations among events and event groups, respectively.

The participant group that provided probability data to yield the most likely scenario
(1, 1, 1, 1, 1, 1) for group V then gave cross impact data for the events of group W on
condition that all the events of group V will occur, the condition identical to the most
likely scenario the participant group had forecast for group V. The participants differed in
their opinions on the occurrence of events in group W, and probability data were taken for
two subgroups of the participants, yielding two most likely scenarios A and B for the
events of group W. The participant group that provided probability data for the most likely
scenario (0, 1, 0, 0, 1, 1) for group V then gave cross impact data for the events of group
W on the condition that the occurrence of the events of group V is specified by the
scenario. The data yielded the most likely scenario C-( 1, 1, *, 1, 0, 1) for group W for
this participant group, where * means that the occurrence of event W3 is indeterminate;
that is, its occurrence is quite uncertain. The IIt and ik for this scenario are five-
dimensional joint probabilities that are the marginal probabilities of the events W 1, W2,
W4, WS, and W6 relative to the six-dimensional probabilities of all six events. We
suppressed the occurrence of event W3 by *, since two six-dimensional scenarios then
resulted having large probabilities that are not distinguishable:
scenario (1, I, 0, 1, 0, 1) QTI;(Ic’= 0.2Gk = 0.17,
scenario (1, 1, 1, 1, 0, 1) ffkcn-) = 0.28;,< = 0.19.
SCENARIOS OF SOCIAL CHANGE IN JAPAN 227
228 M. ISHIKAWA, M. TODA, S. MORI, AND Y. KAYA

TABLE 5
Illustrative Example of Probabilities of
Scenario C for the Events of Group W

Estimated by participants P(i), P(i 1~)

0.65 0.69 0.66 0.72 0.41 0.85


0.61 0.65 0.67 0.88 0.47 0.80
0.48 0.48 0.45 0.50 0.53 0.45
0.69 0.69 0.67 0.65 0.56 0.65
0.13 0.13 0.13 0.13 0.15 0.02
0.79 0.72 0.71 0.70 0.32 0.70

Consistent probabilities P’(i), P *(i j)

0.65 0.68 0.68 0.70 0.48 0.80


0.68 0.65 0.68 0.79 0.55 0.73
0.47 0.47 0.45 0.48 0.47 0.45
0.70 0.79 0.69 0.65 0.55 0.65
0. I1 0.13 0.16 0.13 0.15 0.04
0.86 0.79 0.71 0.70 0.20 0.70

The occurrence probabilities p(i) and conditional probabilities p(i (j) estimated by the
participants, as well as the consistent probabilities p*(i) and p*(i lj) that yielded scenario
C for the events of group W, are shown in Table 5 as an example of cross impact data for
the workshop.
The occurrence of event groups L and E was subsequently analyzed by the three
participant groups. In providing cross impact data for the events of groups L and E, each
participant group assumed the occurrence or nonoccurrence of the events of groups V and
W according to scenario A, B, or C of the participant group. There resulted a few
full-dimensional scenarios (five dimensional for group L and three dimensional for group
E) whose probabilities IIklk(k)and G,( are large and not distinguishable from each other, as
in scenario C for group W. For these scenarios the occurrence of one or two events was
suppressed (events denoted *), and the corresponding marginal probabilities were com-
puted for scenarios of a smaller dimension.
The overall probabilities KI and G of the occurrence of scenario A, B, or C were
obtained by multiplying the scenario probabilities n(i) and Gk of event groups. These are
shown in Table 4, where II and G are the products of II($) and Gk, respectively. Although
If and g for each overall scenario are rather small, they are quite large as compared to
2-20, which is the average probability of the scenarios specifying the occurrence or
nonoccurrence of all 20 events.

THREE SCENARIOS OF JAPANESE SOCIETY IN 2000


The results in Figure 1 were interpreted to yield the following three scenarios. In
these scenarios four aspects (i.e., four event groups) were integrated, supplemented by
discussions throughout the study of overall trends in Japanese society in the next 20 years.

Scenario A. Japanese society has been highly industrialized owing to rapid economic
growth in the past, which has slowed in recent years. Slow but stable economic growth
will continue in the next 20 years.
SCENARIOS OF SOCIAL CHANGE IN JAPAN 229

The future-oriented norm that dominated during the period of rapid economic growth
has been giving way to a present-oriented norm that stresses the enjoyment of daily life.
This trend will continue to alter social values, as is observed in other industrialized
societies.
Change in the work ethic will be clearly seen; work will be considered a means of
workers’ human development as well as a means of obtaining income. Consequently,
efforts will be made to humanize labor relations. In particular, working conditions will be
humanized by, for example, the popularization of flexible work hours. Individual desires
for human development will demand a significant increase in adult education.
These trends in attitude toward individuals’ desire to enrich their total life will
diversify their way of living. People will be less inhibited by social customs and tra-
ditions, and their behavior will become more individualistic. This individualistic behavior
will become evident in job selection, in preference of clothing, and so forth. Another
consequence will be a decentralization of culture, and the present mass culture, dominated
by large cities, will be diversified as local culture flourishes.
The increasing acceptance of individualism will lead to the elimination of various
kinds of social discrimination. Equal opportunity for women will be advanced in employ-
ment, and this will go in hand with a decrease in the difference in career attitudes between
men and women. It is expected that handicapped workers will find appropriate jobs.
Employment opportunities in social service will be substantially expanded. This will help
utilize the capabilities of women, who are underemployed, and meet various welfare
needs. In particular, social care for senior citizens must be improved as the aging of the
population is unquestionably accelerated to approach the population pyramids of mature
nations. The expanded social services will make use of volunteers, because volunteer
activities will become popular as extended holidays become customary, as in the United
States and Europe, and people are more aware of social needs. It is expected that some
kind of social service systems will be established in the areas of health care and social
welfare where specialists and volunteers cooperate.
Another consequence of individualism will be the fall of traditional morals. Kinship
and communal relationships will continue to lose their traditional force in maintaining the
unity of communities. Family life in its traditional sense will be more frequently disrupted
by divorce and career conflict, and generation gaps will increase.
As the present-oriented norm gradually penetrates society, employment systems will
inevitably vary to conform to the change of value. Undesirable working conditions, such
as hazardous jobs and excessive overtime, will be decreased, because even high income
for these jobs will not attract workers who do not consider work only as a means of
income. In order to overcome such difficulties in employment, substantial management
effort will be made to eliminate occupational diseases and mechanical injuries. However,
a shortage of labor supply may become an important social problem with respect to jobs
that are accompanied with risks to workers’ lives as more and more people tend not to
choose such jobs. Labor for such socially unwanted jobs as sanitation and sewage treat-
ment may become in short supply for similar reasons. In any case, the distinction between
two types of work will become more evident as an elite becomes overloaded with re-
sponsibility, leaving the majority to simple tasks.
Although it is not certain whether the desire for higher education will continue to
escalate, the correlation between income and education will become less evident. Con-
sequently, the value of higher education as a means of obtaining a higher income will
decrease.
230 M. ISHIKAWA, M. TODA, S. MORI, AND Y. KAYA

Scenurio B. Scenario B is almost the same except that it is based on the opinion that
individualism does not necessarily contradict close kinships and communal relationships.
The following two paragraphs, describing welfare and community life, replace the sixth
paragraph of Scenario A. The rest is the same as Scenario A.
The increasing acceptance of individualism will go hand in hand with a revival of
kinship and communal relations, which will be recognized as basic human relations. The
revival will result in an increase in communities that resemble traditional rural type
communities but will be distinctly different from them in that individualism is far more
respected.
In these communities members will cooperate to improve communal welfare; in
particular, that of the socially handicapped will be supported by volunteer activities. In
order to organize and respond to these communal efforts, social service will be signifi-
cantly expanded by public policy. Equal opportunities for women as in the United States
and Europe will not be realized, and instead men and women will continue to function in
their traditional roles.

Scenario C. Scenario C described a society in which social development will basically


continue as in the past. Even though some new phenomena will be observed because of
industrialization and the influence of foreign countries, people’s sense of values and way,
of thinking will not easily be altered. The society is briefly sketched below.

Although the slow economic growth of recent years will continue in the next 20
years, the preference for future-oriented norms that has been a driving force behind rapid
economic growth in the past will continue to be dominant. Individualistic behavior will
become evident as a consequence of industrialization and the influence of foreign coun-
tries. Equal opportunities for women and the handicapped will be advanced for similar
reasons. Social services will be expanded by public policy to supplement the weakening of
the communities that have traditionally supported the welfare of their members. Volunteer
activity will not greatly supplement these social services, because individualism will be
expressed in a deemphasis on human relations and an increase in egoistic attitudes. This
will also result in more frequent disruption of traditional family life, due to divorce, the
generation gap, and so on.
The importance of a career in an individual’s total life will not decrease, and compe-
tition in career life and higher education will become more intense. Although social
mobility due to higher education will continue to characterize an active society, over-
education will become a sociai problem as the highly educated become significantly
underemployed. In this context, high-income jobs as compensation for undesirable work-
ing conditions will not decrease, although substantial management effort will be made to
eliminate occupational diseases and injuries.

Concluding Remarks
Scenarios of social change in Japan were obtained in an application of cross impact
analysis. Since the scope of the analysis was extensive, the number of events considered
exceeded the capability of correlational cross impact analysis. A procedure was therefore
developed to generate the scenarios, consisting of a preprocessing of the events and a
step-by-step application of cross impact analysis.
A cross impact method is based on judgmental data and is thus inevitably subjective.
SCENARIOS OF SOCIAL CHANGE IN JAPAN 231

TABLE 6
Example of Relation Between Events and Subjects

Subset i 1 2 3 4

I V V
2 v v
3 V v v

V indicates that subset i includes event e’ [f&(j) =


I]. Otherwise the subset i does not include event e’
[G(i) = 01.

However, we believe that the method is useful in generating consistent overall scenarios
out of fragmented judgments of various aspects of social change. Moreover, the method
reveals explicitly the differences among individual opinions and is therefore useful in
focusing discussion on these differences. For example, in this study we discovered differ-
ences of opinions while analyzing groups V and W. The differences with respect to group
V turned out to yield totally different overall scenarios, whereas those with respect to
group W brought out different opinions concerning the notion of individualism.
The procedure by which the scenarios are developed is significantly dependent on
linguistic statements, and hence on individual interpretations of these statements. The
scenarios and associated cross impact data should therefore be understood as tools for
forming a consistent image of and clarifying the differences in individual opinions. They
thus can help focus discussion on critical aspects of social change. The procedure should
prove to be useful for social planning when used with an understanding of its limitations,
because a powerful objective method is not currently available to forecast the uncertain
future of complex social systems.

Appendix: Quantification Theory III


This appendix presents details of the method and an illustrative example.
Suppose the structural relation between events and subsets is given by Table 6.
Rearranging rows and columns of Table 6 under the criterion that subsets of similarly
constituted events be closely located and that events included in similar subsets be closely
located, we obtain Table 7. In the rearranged table each V is located near the diagonal
running from upper left to lower right. From this property one can easily see that the

TABLE I
Example of Rearranged Relation Between Events
and Subjects

Event eJ

Subset i 2 1 4 3

2 v v
1 v v
3 v v v

Symbols are as in Table 6.


232 M. ISHIKAWA, M. TODA, S. MORI, AND Y. KAYA

criterion for rearrangement is tantamount to maximizing the correlation coefficient be-


tween the scores yi for subsets and the weights xi for events.
The correlation coefficient p can be represented as

where

(19)

N =
i=l
$’n,. (22)

The coefficient p 1smaximized by solving the following simultaneous equations:

3P
_- - 0
j = 1, ., I, (23)
dXj '

dp-0
__ j=l ,.. .) n. (24)
'Yi ’
Assuming without loss of generality that an average of x is equal to 0,

X = -L f: i $(j)xj = 0, (25)
N i=i j=i

we can substitute (18)-(22) into (23) and (24) to obtain (13).


In the above example the resulting eigenvalues and correlation coefficients are { 1.O,
0.540, 0.154, 0} and ( I .O, 0.735, 0.392,0> respectively. In general, the event weights in
(16) corresponding to the largest eigenvalue are most meaningful for the purpose of
clustering. However, since the event weights corresponding to the first eigenvalue are the
same over all events, and hence trivial, those corresponding to the second and the third
eigenvalues are used for clustering, as displayed in Figure 3. Figure 3 implies that events 1
and 4 form a cluster, since these events are located close together. Events 2 and 3 form
singleton clusters since they are located apart from other events.
SCENARIOS OF SOCIAL CHANGE IN JAPAN 233

third axis

i-l.5

-2

0.5 second axis

'1

'4

Fig. 3. Example of two-dimensional map of events.

The authors wish to acknowledge their gratitude to R. Mikita, H. Tani, and S.


Humamatsu of the Economic Planning Agency for valuable discussions.

References
1. Advisory Board for Integrated Public Policies, Toward inregrated public policies (in Japanese: Sogo Shakai
Seisaku o Mofomete), Japanese Economic Planning Agency, Tokyo, 1977.
2. Y. Kaya, M. Ishikawa, and S. Mori, A Revised Cross Impact Method and Its Application to the Forecast of
Urban Transportation Technology, Technol. Forecast. Sot. Change 14, 243-257 (1979).
3. J. C. Duperrin and M. Godet, SMIC 74-A Method for Constructing and Ranking Scenarios, Futures 7,
302-312 (1975).
4. C. Hayashi: On the Prediction of Phenomena from Qualitative Data and the Quantification of Qualitative
Data from the Mathematics-Statistical Point of View, Ann. Insr. Stat. Math. 3, 69-98 (1951).
5. C. T. Moore (ed.), Emerging Methods in Environmental Design and Planning, MIT Press, Cambridge,
Mass., 1970, Ch. 14.
6. R. D. Lute et al. (eds.), Handbook of Mathematical Psychology, Wiley, New York, 1963.
7. M. R. Anderderg, Cluster Analysis for Applications, Academic, New York, 1973.

Received 8 February 1980; revised 17 October 1980

You might also like