international relations (5). Why has it been used to such a small extent in the social sciences? Perhaps the reason lies in the insufficiency of the available data. If this situation improves, the problem of validation will become less severe. The results if the post-diction can be evaluated if and when we use a differentiated interpretation of regularity (see 3.3).

Today, a subject area that could be approached with the technique of post-dictio is higher education. The development of a university, a faculty, or a department can be reconstructed to quite an extent on the basis of written documents. The system of higher education shows a certain degree of democratic control, and therefore a certain degree of openness. We need additional data, but these can be obtained with no more difficulty than we are normally confronted with in social research.

4.2 Consequential Analysis

Consequential analysis may be considered as a prediction containing many conditions. If consequential analysis is employed in a field in which this type of prediction has been used and validated (e.g. by post-diction on a regular basis, its results may be extremely relevant, especially from a practical point of view.

4.3 Anlaytic, Educational and Heuristic Functions

Not only analytical models but educational and heuristic models, too, should demonstrate a high degree of validity in order to be useful.

Educational models provide the student with an abstraction of a syste he will have to deal with later. If this model does not contain the main characteristics of the social system it describes, the educational process is bound to be less than optimal. Of course we "distort" reality in educational models, for instance by changing the time-decision. But how can we design an adequate educational model with an optimal "distortion of reality without knowing this social reality to an adequate degree of detail and validity?

4.4 Theoretical and Operational Levels

In the empirical sciences (including social sciences like political sciences or sociology) we distinguish between statements or propositions on a theoretical level and on an operational or experimental level (6).

Propositions on a theoretical level have a relatively high degree of abstraction. They cover a great number of phenomena, that may differ with regard to time, location, etc.

Propositions on an operational level are restricted to phenomena of a relatively low level of abstraction. They contain statements that give information on a relatively high level of verification. To meet this standard of verification they are confined as a rule to objects that have quite narrow limits as to time and space.

Theoretical propositions cannot be confirmed empirically in a direct way. They have to be "translated" into operational statements, and their confrontation with social reality (attempts to falsification) takes place on the operational level.

Analytical simulations that aim at a high degree of empirical validity have a tendency to keep to a low level of abstraction. А model in the field of international relations simulating the development of one international conference on disarmament may be able to predict (by post-diction for instance) some characteristics of the phases of this conference. Maybe in the long run the level of abstraction of models like this can be increased a bit. But a predictive model of "the" contemporary international negotiation, or of the development of "the middle-size town in the United States or in Europe is an utopia. General analytical models are like theories: their "predictive" power is limited. We must translate their general statements into operational statements, that can be put to a test in experimental situations.

The difference between theoretical and operational models has consequences also for the use of simulation in the field of education and training. Education aims at a general preparation of the participants (students etc.) for a great variety of activities. This implies, that educational models cannot have a high degree of direct predictability. Their validity is examined by testing operational statements that have been derived from the general statements. Training has a different perspective. It aims at preparing the participant for a restricted set of activities. Simulation models developed for training purposes show a low degree of abstraction. Therefore, their "validity' with regard to their reference-system can be determined in a much more direct process of verification.

The problem of the data behind simulation models is located at the operational level of this model. But of course the theoretical level is involved also, especially as soon as theoretical models have to put their statements to an empirical test after they have been "translated" into operational statements.

Our essay puts forward a number of hypotheses: a) in the development of simulation models the data constitute

the most serious drawback;

b) we have to reconsider the regularity and predictability of

social systems;

c)

we have to develop and use new approaches in data collection and in validating the obtained data; some new approaches are available;

d) better data will provide us with better conditions for post

diction and consequential analysis;

e) the empirical basis as a weak link in the development of

simulation models does not only affect analytical simulations but educational and heuristic simulations also.

If the empirical basis does indeed constitute the weakest link in the chain of model building, improvement of this basis should receive high priority. This might comple the simulationists to expeditions into fields which they at present leave mainly to others.