340
A Further Comment on SMIC
74
book to see how we treat this problem and the way in which the results are used. 4 In practice, the SMIC method provides the most probable images for the phenomenon being studied in the future. Once the final and intermediate images have been selected, the method of scenarios consists of exploring and describing, whilst taking into account the mechanisms and trends identified during the retrospective study, the most coherent paths leading to those images (see Figure 1). Finally, the main role of the SMIC method is to assist us in identifying the most probable futures which then
A FURTHER
COMMENT
become the scenarios.
object
of the
method
of
References
Duperrin and M. Godet, “SMIC 74 -a method for constructing and ranking scenarios”, Futures, 7 (4), August 1975, pages 302-3 12. 2. R. B. Mitchell and A. Tydeman, “A note on SMIC 74”, Futures, 8 (l), February 1976, pages 64-67. 3. J. C. Duperrin and M. Godet, “Prospective des systtmes”, Library of theses: Faculty of Science (PARIS VI) and Faculty of Economics (PARIS I). 4. M. Godet, Crise de la @!vision et essor de la firospective : exemples et mbthodes (Paris, les PUFF, forthcoming 1977). 1. J. C.
ON SMIC 74
R. B. Mitchell and J. Tydeman WE agree that our suggested method for finding consistent probabilities, as a first step towards finding scenario probabilities, is theoretically inadequate in that it may, in odd cases, not allow a feasible solution for the Ilk. The reasons for this inadequacy are of fundamental importance and may cause further questioning of the SMIC method. It should be noted that in most practical situations the solutions will be identical. The difficulty with our suggested programme is not the number of constraints, as implied by Duperrin and Godet,l but the inability of expressing, in terms of the single and paired event probabilities, the overall constraint that requires the sum of all scenario probabilities to be unity. Pairwise consistency does not imply overall consistency. R. B. Mitchell is College Fellow in Operations Research at the Canberra College of Advanced Education, Australia; J. Tydeman is Senior Lecturer in the Administrative Studies Programme at the Australian National University, Canberra, Australia.
Our programme does find the best fitting pairwise probabilities in the context in which the respondent has and if these do not been working, permit a feasible set of scenario probabilities then perhaps this fact should be brought out and not buried in the technique. Faced with such an outcome the respondent may well elect to change some aspects of his estimates rather than others. For instance, in the example of their reply, Duperrin and Godet have removed what appears to be an essential feature of the respondent’s estimates, namely, that events 1 and 2 are mutually exclusive. This raises the whole question of the relationship between the results of the techniques and what the respondent really believes, which we specifically excluded in our note.s In general, we feel that there should be as much interactive feedback as possible to ensure that the results do reflect, as closely as possible, the respondent’s views. The major points of our note, regarding multiple solutions and the
FUTURES
Augurt 1976
Further Comments on Cross-impact Analysis
introduction of a criterion to make a choice, have been accepted. It is unfortunate that Duperrin and Godet did not see fit to discuss in their paper their earlier work on this issue. Indeed, they implied that the solution was unique.s In relation to the criterion selected by Duperrin and Godet to choose between the solutions, we offer the following points : l
l
The criterion of maxk (max Ilk) is just as arbitrary as any other. We believe that the circumstances of a study might suggest other criteria as appropriate for that study. This criterion does not necessarily yield a unique solution.
l
341
This criterion is not a linear function of the II& and will, in general, require up to 2n runs of the simplex algorithm, maximising each IIn in turn.
References “SMIC 74-a reply from the authors”, Futures, 8 (4), August 1976, pages 336-340. 2. R. B. Mitchell and J. Tydeman, “A note on SMIC 74”, Futures, 8 (I), February 1976, pages 6467. 3. J. C Duperrin and M. Godet, “SMIC 74 -a method for constructing and ranking scenarios”, Futures, 7 (4), August 1975, pages 307, 311, 312. 1. M. Godet,
FURTHER COMMENTS ON CROSS-IMPACT ANALYSIS P. Kelly THE method of cross-impact analysis suggested by Drs Duperrin and Godetl has a strong appeal to those of us who are required to advise the decision makers of industry and government because it appears to combine the knowledge and experience of “experts” with the rigour and objectivity of mathematics. Such a method should, in principle, answer the criticisms levelled at both the systematic approach to future studies, that it ignores the imponderables of human behaviour, and the intuitive approach, that it is intrinsically imprecise and subjective. Unfortunately, if such a bridging exercise fails to deliver the goods it is liable to do more harm than good, for it will help to exacerbate the already visible and growing rift between the “soft” sciences and the “hard” sciences, just as the ultimate failure of the scientific historians of the late 19th The author is with Unilever Research, Port Sunlight Laboratory, Wirral, Merseyside L62 4XN, UK.
FUTURES
August 1976
century led to the reversion of history to an exercise in rhetoric during the period between the wars. In the meantime the application of specious scientific principles to the analysis of expert opinions which are no better than guesswork inevitably fosters massive self-delusion in both parties, the one impressed by the elegance of quadratic programming algorithms, the other by the reduction of his shrewd assessments of event likelihoods to a seemingly precise description of things to come crisply delineated on a computer printout. Mitchell and Tydeman2 as well as the present correspondent in an unpublished paper, have indicated that the method of Duperrin and Godet does not provide a unique solution to the problem of computation of total scenario probabilities from given data on event and event pair conditional probabilities. It is not necessary to rehearse the arguments of Mitchell and Tydeman, which are sound in so far as