Nature of prediction and explanation

Since we will be using the terms prediction and explanation throughout this paper, it is important to examine in more detail the meaning of these terms in methodology. In closed systems, where the ceteris paribus condition is fulfilled or can be made to hold effectively, explanation and prediction are symmetrical: the model  is explanatory if the implied set of observationstatements (initial conditions) refers to an event that has already occurred; it is predictive if the implied set of observation- statements refers to a future event. 3 The time element enters into both concepts because real world processes have measurable duration and duration extends backwards and forwards. If knowledge of a specific event at a future date is desired, it is necessary to know the process by which — in due time — this event will be brought about. In the same way, if an event that has already occurred is to be explained by relating it to other prior events, a process in time is involved. In either case reference is made to specific space-time regions.

In nonclosed systems the symmetry of explanation and prediction does not hold. Before the event has happened it is not known what exogenous variables may influence the course of events. After the event has actually occurred, it is (in principle) always possible to identify all the variables that had a measurable effect on it. Thus, in nonclosed systems, hindsight is better than foresight.

In a complete explanatory model (which, being complete, is also a predictive model) the set of general laws must, in addition to empirical laws, also contain laws of the highest available order: laws which explain all other laws of lower order used in the explanation and prediction. The empirical laws (laws of lowest order) must contain terms that refer directly to observables and thus must be directly verifiable. Laws of higher order, used to explain lower-order laws, need not contain terms referring directly to observables although they may, of course, contain such terms. The same law may, in fact, appear in one predictive model as a lowest-level empirical law and in another model as a higher-order law. However, in the hierarchy of a theoretical structure, the higher the order of a general law (that is, the more steps it is removed from the lowest-order laws and the broader the scope of the explanation that it provides), the less are the chances that it will contain terms referring directly to observables. To the extent that it does contain such terms, the law remains, of course, rigorously subject to direct empirical verification. If, however, it does not contain any terms referring directly to observables, it can be only indirectly verified: empirical laws must be validly deducible from it in a finite number of steps. These lowest-order laws must be empirically verifiable.

In predictive models the higher-order general laws appear as premises from which the lower-order laws are validly deducible. This should not obscure the fact that in the actual enterprise of discovery they may yield priority to the empirical hypothesis. This is well expressed by saying that the higher-order laws explain the lower-order laws that are implied by them. By explaining the lower-level laws, the higher-level law makes possible the building of a consistent theoretical structure out of originally isolated and unrelated empirical hypotheses. Once such a theoretical structure (even the simplest one) exists, it becomes possible to derive from it new empirical hypotheses that are not obtained from observations but which, instead, predict observations. It is important to remember that the deduction of a new empirical hypothesis from an existing (and thus accepted) theoretical structure does not establish the truth of the hypothesis nor make the ultimate empirical test unnecessary. The derivation of empirical laws from more general laws is a source of tentative hypotheses, not a procedure of verification.

The fact that an empirical hypothesis, or a higher-order law explaining it, is inconsistent with an accepted theoretical structure does not strictly disconfirm the hypothesis. In fact, it is possible that the entire previously accepted theoretical structure may be disconfirmed by such an inconsistency. Actually, inconsistent hypotheses may be held simultaneously. The point is that such a situation will be considered to be unsatisfactory and, hopefully, will not be permanent. Efforts will be made to reconcile the inconsistent hypotheses or to replace them by others that are consistent with each other. Such inconsistencies are viewed as flaws in  the theoretical structure — excused by reference to the fact that science cannot be omniscient.

Thus, because we cannot be certain that any of our laws is really true, the valid derivation of an empirical hypothesis from higher-order laws cannot establish its truth. On the other hand, hypotheses obtained in this way do normally command a high degree of confidence because we obviously must have confidence in the body of accepted laws. The history of science abounds with examples where hypotheses first deduced from existing theoretical structures have later been empirically verified and have survived confrontation with actual observations. In turn, the theoretical structure that has served as premise commands increasing confidence as more empirical hypotheses — later verified and accepted — are deduced from it. The possibility of thus indirectly verifying hypotheses that for some reason cannot be directly verified is of great importance for the social sciences.

With this background we will examine the prediction problem in economics.

Source: Skyttner Lars (2006), General Systems Theory: Problems, Perspectives, Practice, Wspc, 2nd Edition.

Leave a Reply

Your email address will not be published. Required fields are marked *