The history of many fields of science shows a characteristic pattern. There is a time in which the science goes through a philosophic stage in its development;­the emphasis is on theory, on general concepts, and on the questioning of the fundamental assumptions and methods by which knowledge has been accumulated. At the end of such a philosophic stage often stands an agreement on some basic assumptions and methods-though not necessarily on all of them-and a shifting of interest to the application of these methods to the gathering of detailed facts. The philosophic stages in the development of science define the main lines of interest; in the empirical stages these interests are followed up. Philosophic stages in the development of a particular science are concerned with strategy; they select the targets and the main lines of attack. Empirical stages are concerned with tactics; they attain the targets, or they accumulate experience indicating that the targets cannot be taken in this manner and that the underlying strategy was wrong.

In any case, every empirical stage ends with the need for a revision of fundamental concepts and underlying strategy. If these were inadequate, the revision must come soon. If they were adequate for a time, the revision must nevertheless come later, for the very success of the concepts, methods, and interests adopted will lead in time to an accumulation of data and problems that will point beyond the interests and methods by which they were discovered. In the end, every empirical stage will have bitten off more facts than it can chew, and scientists will have to turn to a new philosophic stage for more powerful analytic equipment.

The test of this conceptual equipment must be twofold: it must be operational, that is, it must lead to inferences capable of being confirmed or refuted by repeatable physical operations; and it must be fruitful, that is, it must lead to new observations and experiments, and eventually to further developments in theory. The stage of philosophic or conceptual emphasis must again lead to empirical progress, and the progress eventually leads once more to new fundamental problems.

The social sciences today perhaps are approaching another "philosophic crisis"-an age of re­examination of concepts, methods, and interests, of search for new symbolic models and/or new strategies in selecting their major targets for attack. At the same time, their data have increased in quantity, and selection has become even more imperative. To understand and describe in full detail the political process even in a single country may well take the work of a lifetime. To do the same for several countries means to multiply the amount of possible data to be looked for and of possible questions to be asked.

Clearly, selection is of the essence. What are the data most worth getting? What are the questions most worth asking? What are the propositions most worth verifying or disproving? Our answers to these questions will determine the shape of any investigation and may do much to determine the value of the outcome. But to say that we must choose our questions is another way of saying that we must choose our concepts or models.

We obtain the answers to these questions by our preliminary "understanding" of the situation we propose to study. But we could perhaps obtain better answers if we had a better critical grasp of the vague process of "understanding" to which we commit so much of our professional fortunes.

This process of understanding, from its early stages to the final theory or strategy of inquiry, is carried on by means of symbolic models that all of us use in our thinking. It may be worth our while to gain a clearer picture of this process of choosing models and of using them, and we may end with more and sharper intellectual tools than were available to past generations. To be sure, a master sometimes may accomplish more with crude tools than could a novice with better technical equipment. Even so, there may be some usefulness in a brief survey of the nature of conceptual models, and a discussion of a few more recently developed models of this kind in terms of their possible usefulness to comparative political studies.

To discuss some of the recent models in the field of political science, we must recall briefly the role of any model in the pursuit of knowledge. In order to "know" a process, we must use symbols that we match in some way against the distribution of some aspects of the process we study, much as we match the distributions of symbols on a map against the distributions of coastlines, rivers, or roads in the landscape to be pictured. To know thus always means to omit and to select. In this sense, no knowledge is completely "objective."

But to know also means to match our standards of selection, explicit or implied, against the practical requirements of the action for which this knowledge is to be used. If we want to drive a car, we can omit offshore sandbanks from our map, but we must not omit roads. (If we want to sail a boat, some of the sandbanks may have to be included.) In this sense, no knowledge can be completely "nonobjective," if it is to be applied.

Knowledge depends on four things:

1. the selective interests of the knower;

2. the actual characteristics of the situation to be known;

3. the selective operations by which these characteristics can in fact be experienced or measured; and

4. the system of symbols and physical facilities by which the data selected in stages 1 and 3 are recorded and used for later application.

Knowledge is thus a process in which subjective and objective elements inevitably meet. Its first stage is subjective; the interests and needs of the knower. Its second stage-the existing characteristics of the situation-is objective to the extent that these characteristics are not significantly changed by the effects of observation. The third and fourth stages involve both elements: an objectively existing repertory of available measuring operations, and of encoding and recording facilities; and a set of subjective choices of items from this available repertory for use in actual operations. Moreover, the selection, in terms of interest or the asking of questions (stage 1), and in terms of measurability or the getting of relevant data for answers (stage 3), need not coincide: we may be interested in things for which we may at present have no operations of measurement. We may be interested, as it were, in color, but may find ourselves limited by our equipment to line drawings or black­and­white photography.

Pictures, photographs, and maps are simple models of situations in space. Various geopolitical maps and crude diagrams of political or social structures have been used by social scientists for a number of years.

Sometimes we are interested in mapping the performance of some thing or process over time. Here, again, the success of such mapping or diagramming over time will depend on the four stages of the process of knowing: the criteria of interest selected, the actual characteristics of the process to be studied, the operations of measurement employed, and the symbols and symbol systems used for recording and using the results. Curves of the growth of population, of production in particular industries or of votes for a particular party, or of membership in labor unions, are examples of such mapping over time.

Maps, as well as time diagrams, can do more than summarize existing knowledge. They can suggest ways of looking for new knowledge, and help to predict regularities that may or may not be confirmed by later experience or measurement. We can do these things through the operation of prediction. This operation consists in noting the pattern of the distribution of a set of known data, and extending tentatively a similar pattern into some area of space, or some period of time, from which we have as yet no firsthand data. In this manner we may guess at least something of the features of an unexplored country by noting the distribution of rivers and mountains leading to the edges of the "white patch" on our map. Nineteenth­century explorers did use such reasoning to guide their search for the sources and tributaries of the river Nile.

In making predictions over time, we must similarly collect series of selected data for the past, abstract from them some pattern, and extend or "extrapolate" that pattern tentatively into the future. As in the case of the map, this procedure may yield two kinds of predictions: (1) general predictions of interest (for example, does the future seem likely to contain data of interest to us, in terms of our criteria of interest assumed at the outset?); and (2) specific predictions of distribution (for example, what relevant events seem likely to occur in the future, and when do they seem likely to occur?). General predictions of interest are related to the heuristic functions of models; they tell us where to go to look for something interesting. Specific predictions of distributions are predictive in the narrower sense of the word; they tell us just what we should expect to find. Heuristic as well as predictive forecasts-for example, of future population growth, of anticipated market changes or needs for raw materials, of business cycles, or of changes in military potential or political stability-are all well known to social scientists.

By extending several time series tentatively into the future, side by side, we may make a guess as to what might happen if the peaks or valleys of several such series, let us say, of industrial strikes and agrarian unrest, or exports and domestic credits, should happen to coincide at some date in the future, even if they did not do so in the past. Natural scientists can predict in this manner the likelihood of rip tides, when the time of flood, the phase and position of the moon, and a strong onshore wind may combine to maximum effect. Students of social and political science might similarly become able to appraise the likelihood of rip tides of social change, when several normally separate processes making for social stress might coincide so as to exercise their greatest force. Thus, if in each of, say, one hundred countries there were at work three mutually independent stress­producing processes-such as agrarian revolts, industrial unrest, and foreign military conflict-and if each of these processes should tend to become acute, or to reach a peak about once every ten years, then the chances would be better than ever that these three peaks would coincide, and the "rip­tide effect" would shake or even overthrow the government in at least one of these countries within the next ten years.

Our discussion of the nature of knowledge has clear implications for the functions of models. We may think of models as serving, more or less imperfectly, four distinct functions: the organizing, the heuristic, the predictive, and the measuring (or mensurative).

By the organizing function is meant the ability of a model to order and relate disjointed data, and to show similarities or connections between them that had previously remained unperceived. To make isolated pieces of information fall suddenly into a meaningful pattern is to furnish an esthetic experience; Professor Paul Lazarsfeld once described it as the "Aha!­experience" familiar to psychologists. Such organization may facilitate its storage in memory, and perhaps even more its recall.

If the new model organizes information about unfamiliar processes in terms of images borrowed from familiar events, we call it an explanation. The operational function of an explanation is that of a training or teaching device that facilitates the transfer of learned habits from a familiar to an unfamiliar environment. If it actually does help us to transfer some familiar behavior pattern to a new problem, we may feel that the explanation is "satisfactory," or even that it "satisfies our curiosity," at least for a time. Such an explanation might be subjectively satisfying without being predictive; it would satisfy some persons but not others, depending on each person's memories and habits. Since it yields no predictions that can be tested by physical operations, it would be rejected by some scientists as a "mere explanation" that would be operationally meaningless.

Certainly, such "mere explanations" are models of a very low order. It seems, however, that explanations almost invariably imply some predictions. Moreover, even if these predictions cannot be verified by techniques practicable at the present time, they may yet serve as heuristic devices leading to the discovery of new facts and new methods. This heuristic function of making discoveries more probable has already been discussed above. Here, it is mainly important for us to remember that the heuristic function of a model may be independent to a considerable degree from its orderliness or organizing power, as well as from its predictive and mensurative performance.

Little need be added to our earlier discussion of the predictive function of a model, beyond the well­known requirement of verifiability by physical operations. There are different kinds of prediction, however, that form something of a spectrum. At one extreme we find simple yes­or­no predictions; at higher degrees of specificity we get qualitative predictions of similarity or matching, where the result is predicted to be of this kind or of that kind, or of this particular delicate shade; and at the other extreme we find completely quantitative predictions that may give us elaborate time series that may answer the questions of "When?" and "How much?"

At this extreme, models become related to measurement. If the model is related to the thing modeled by laws that are not clearly understood, the data it yields may serve as indicators. If it is connected to the thing modeled by processes clearly understood, we may call the data obtained with its help a measure ­- and measures again may range all the way from simple rank orderings to full­fledged ratio scales.

The effectiveness of our predictions will depend in all cases upon the four elements of the process of knowledge we listed earlier. One of these elements is never completely under our control: the actual structure of the process that we are trying to know in the present and to predict for the future. If this structure happens to have sufficiently large discontinuities in its performance, all our guesses and extrapolations may remain mistaken.

The other three elements of knowledge, however, are under our control to a greater degree. Our selective interests develop with our increasing experience and knowledge, and with our expanding range of needs and of things we are able to do. Our operations of observation and measurement develop with the growth of our technology, and with the introduction of new instruments or methods of inquiry in the social as well as in the natural sciences. Finally, the range and effectiveness of our symbols and symbol systems increase, and accompany increases in our power to select and abstract, to store and recall, to analyze and recombine the sets of data we obtain, to extend them for prediction, to transfer them easily for communication, to submit them to operational tests for verification, and to apply them to behavior. Progress in the effectiveness of symbols and symbol systems is thus basic progress in the technology of thinking and in the development of human powers of insight and action.

A symbol is an order to recall from memory a particular thing or event, or a particular set of things or events. Any physical work or event that functions repeatedly as such a command can thus function as a symbol. If we use several symbols, so as to be able to recall several different things, we must connect our symbols with some operating rules. Together, the set of symbols and the set of operating rules form a symbol system or a model.

Any language is such a symbol system. Roughly speaking, it is a set of socially standardized words, or shorter sound patterns or "phonemes" with a set of rules of grammar and syntax, which specify their combinations. A system of geometry or arithmetic, a logical calculus, a game like chess or poker, or a model, physical or abstract, of some process, or a "conceptual scheme" in a natural or social science are all symbol systems of this kind.

If the system has been chosen for purposes of play, as in choosing a game, the symbols or rules adopted need only be such as to give the player such thrills or challenges for his talents as he desires. If the system has been chosen for purposes of knowledge of the existing world, as in scientific models and conceptual schemes, and in applied mathematics, then it is desirable that the symbols and rules should match as well as possible the distributions and sequences of events in the process of which knowledge is desired. If the model actually matches the reality, then the outcome of operations on the model may be used to predict the outcome of operations in reality, where such operations might be difficult or costly.

Models can be formal or material. In the case of formal models, such as mathematical or geometric models, both symbols and rules are themselves abstract, and are recorded by means of signs that can be set down on paper. Some models of this kind may be quite unsuited to visual representation. The "consumption function" and the "production function" in John Maynard Keynes' system of economics give precise quantitative predictions that can be represented on a graph, but the mathematical model from which this graph is derived is almost as difficult to imagine visually as are some of the mathematical models used in quantitative physics.

Other formal models may seem more familiar to us, since they are at least loosely connected with some familiar pictures from everyday experience­even though these pictures may fail to give the content of the models much precision. In this manner, the lonian philosophers transferred the familiar models of "law" and "cause" from social life into the world of nature, but the precise formal content of the concept of "strict causality" in either nature or society remained a subject of discussion for many centuries.

In the case of material models, symbols may be tangible objects (as in the parts of a model airplane) or unseen processes (as electric currents in a network analyzer). The operating rules are then given by the physical properties of the resulting system.

In all cases, models must be tested for their relevance: do they match those aspects of the empirical process in which we are interested to a degree of accuracy sufficient for our purposes? Whether or not a model matches reality must be established by some critical process, that is, some physical process, simple or complex, that has one kind of outcome if the matching is close enough, and another outcome if it is not.

It seems clear, from what has been said thus far, that we all use models in our thinking all the time, even though we may not stop to notice it. When we say that we "understand" a situation, political or otherwise, we say, in effect, that we have in our mind an abstract model, vague or specific, that permits us to parallel or predict such changes in that situation of interest to us.

When we say that we "understand" a person, we may mean one of two things. Either we mean that we understand his situation, and can "put ourselves into his place," that is, that we have a model of the conditions under which he is acting that permits us to conclude that we, with our memories and values, would act very similarly in that situation as he does with his memories and his values. Or we mean that we understand his outlook; that is, we can imagine a model of his mind, with his memories and values, that is sufficiently accurate for us to predict-and perhaps to experience emotionally by empathy-how he would act with his mind and his personality under conditions in which we ourselves might act quite differently.

The first type of understanding, which built models of different situations but treated human nature as essentially uniform, was prominent in the political science of Hobbes and Locke. The search for the second type of understanding, which seeks for models of different personality, culture, and value patterns so as to retrace or predict their choice of goals and goal­oriented actions, has become prominent in the "understanding sociology" (verstehende Soziologie) of Max Weber and in the work of modern anthropologists.

This kind of understanding of individuals and groups "from the inside," as it were, can again be visualized in two broad ways: as a rational reconstruction of the personality, culture, or cognitive map of the actors concerned, or an act of empathy or role­playing, that is an emotional simulation of their feelings by an imaginative manipulation of our own minds. This type of understanding by empathy has been stressed and elaborated by Wilhelm Dilthey and his followers, but its basic idea is simple: can we in our imagination feel as the other person feels, value what he values, experience his inner tensions as if they were our own-regardless of whether we approve or disapprove of his purposes? This is the understanding of what the sociologist Talcott Parsons would call the "evaluative" and the "cathectic" aspects of the other person's actions.

If we suggest that understanding of impersonal situations, as well as personal actions, is possible by mean of models, and perhaps only by means of models, we are apt to meet with two kinds of objections.

The first objection is based on the fact of uncertainty. Since many events in politics and social life are uncertain until the moment they occur, would not a prediction based on models introduce an unwarranted bias in favor of some assumed strict causality or determinism? This objection, where it still persists, is based on ignorance or, more charitably put, on a preoccupation with obsolete models. There is no need to put more "causality" or "determinism" into our symbolic models than we have reason to expect to find in the situations we intend to investigate with their aid. Models can be set up in terms of probability, and they can be revised in line with the probability distributions found in the empirical data. Our entire discussion of prediction was in terms of a repeatable operation, and not in terms of any construct of "causality." The pitfalls of the notion of causality have been pointed out for the natural sciences by P. W. Bridgman and for the social sciences by R. M. MacIver. Political scientists can very well seek out and test possible regularities and probabilities without becoming entangled in the metaphysics of any absolute causality concept.

The second objection comes from a seemingly opposite viewpoint. The most important events, it claims, are not merely uncertain; they are unique. They can, therefore, be indicated by symbols, but not described by them, regardless of their arrangement in any system, language, or model whatever. Such events are thus ineffable. At most (and only if we exempt the nerve cells of the human brain from the limitations of all other symbol­carrying structures), they can be understood by solitary individuals through incommunicable intuition. In a less extreme version of this argument, comparability is not denied, but is limited to the unimportant aspects of each situation, leaving intact the "essential uniqueness" of each historical event.

This theory of uniqueness rests on unexamined assumptions regarding the nature of knowledge. As we have seen, no knowable object can be completely unique: if it were radically unique it could be neither observed nor recorded, nor could it be known. Any object or event that can interact with others sufficiently to make a relevant difference to their outcome must have sufficient structural similarities to permit such interaction. Anything that can interact with events important for us must have some structural similarities with them, and to a lesser extent with us; and once it has structure, there seems to be no a priori reason why it could not be matched by suitable symbols. Of course, our current models of many particular events may be too crude to permit the effective mapping of the probabilities involved, or the effective prediction of any probable results that would be important for us. But to conclude from this that these events cannot be effectively paralleled for such purposes by any symbol system requires either metaphysical convictions or a sweeping prediction of the entire future course of social science.

In current social science, our problems are more practical. All political processes and institutions we observe contain combinations of similarities and differences, and thus become accessible to our knowledge. Indeed, it is only against a background of similarities that differences can be recognized. It is only later, as a second step, that new symbols can be assigned to those groups of aspects that remain different from those previously familiar, and different from each other, and that these new data become part of our experience. In the course of this process, political scientists-like other men-must use comparisons of the relatively simple and familiar as stepping­stones to the gradual conquest of the relatively complex and unique.

This is in fact what they have done. From the comparative study of universal traits, anthropologists have gone on to the first steps in the study of particular configurations of culture. Psychologists have used general schemes, such as those proposed by Sigmund Freud, Abraham Kardiner, and others, as the background against which they could try to evaluate the particular personality problems of individual patients. Economists began their work with the search for uniform laws governing the relations between supply and demand, or the changes in the wealth of nations, and are now gradually progressing to the study of the "propensity to consume" (J. M. Keynes) or "propensity to innovate" (W. W. Rostow) in particular periods and countries; and of the performance, stability, and growth of particular national or regional economies.

From Chapter I, The Nerves of Government: Models of Political Communication and Control, Karl W. Deutsch, New York: The Free Press (A Division of the Macmillan Company), 1966.