Lewis White Beck*
The “Natural Science Ideal” in the Social Sciences
Scientific Monthly,68 (6
June 1949, 386-394
IMAGINE a man who builds a house like the Joneses, at considerable inconvenience to himself because actually he needs something quite different. As soon as he gets the foundations laid, the Joneses begin tearing down parts of their house, adding new wings, and overhauling its foundations. Our poor social climber has committed himself; he has to continue to build according to his plans whether he likes them or not. To comfort himself, he says he has the kind of house the Joneses have and ignores the fact that they are doing a big job of renovating.
This little fable of keeping up with the Joneses fits the relations existing until a short time ago between the social and the natural sciences. The climbers, the social scientists, have tried to imitate the Joneses, the natural scientists. Now the social sciences have an immense house much of which is not very useful; it lacks many of the modern conveniences; but it seems to be scientific, just the same, and that often seems to be enough. But the social scientist might be far happier in his house, or he might be more successful in renovating it to meet modern needs, if he gave up pursuing the past glory of the great edifice of nineteenth-century physics.
When splitting off from philosophy in order to become scientific, the social studies took a bad moment to imitate the natural sciences. They did so just before the natural sciences themselves began to undergo major changes. The result is that many social scientists pride themselves on being natural scientists or regret that they cannot be, whereas the science they emulate or would like to emulate became obsolescent fifty years ago.
In imitating the natural sciences, the social sciences attempted to follow both the methods and the metaphysics of the former. The social studies tried to attend only to observable and measurable entities and to connect these by simple causal or functional laws. If the social scientists thought that they were like the natural scientists in studying “reality,” they became mechanists or materialists. If they feared equating their verified hypotheses with “reality,” as many natural scientists did, they became positivists. In either case they took over ready-made philosophies of the nature of scientific objects. But there was no unanimity on the philosophical foundations current among the natural scientists, and the “unity of the natural sciences,” by virtue of which they might have served as an unequivocal model, was an illusion even before the death of Comte.
The social sciences, therefore, neither emerged from, nor could they later merge into, a homogeneous body of natural science doctrine. The natural science ideal, which many social scientists wished to pursue but which was vehemently rejected by others, was much more ambiguous than it appeared to be in the work of Comte and Spencer. By the time of Dilthey, with his emphasis on the function of sympathetic imagination in social studies, the opposition of the natural and the social sciences was predicated upon an almost complete misunderstanding of the methodological foundations and metaphysical implications of the natural sciences. It would have been much more to the point to have compared the status of the new social sciences with that reached by physics in the time of Galileo than to compare these nascent sciences with a physics already showing signs of passing through the change of life. The contrasts between explanation and description, between nomothetic and ideographic procedures, and between the ideals of a beschreibende and a verstehende psychology were not so much contrasts between the natural science ideal and the ideals perhaps more germane to the social studies, as they were signs of problems which every science, whether it be natural or social, must face in the early stages of its development.
It is consequently beside the point to contrast the natural and the social sciences in the language used in the early part of this century. Neither the
* Dr. Beck (Ph.D., Duke, 1937) who has taught philosophy at Emory University, the University of Delaware, and at George Washington University, is now professor of philosophy at Lehigh. His article is based on a lecture he gave last year before the Delaware Psychological Association.
natural nor the social sciences were homogeneous bodies of doctrine in simple conflict with each other. No clear-cut decision could have been intelligently made between the alternative of following or rejecting the natural science ideal. There were analogous conflicts within both bodies of knowledge between opposing strategies. In each case these conflicts have been resolved in analogous ways during the present century. There is now a continuity of method and philosophy in the two branches of science that could not have been dreamed of even by the most naturalistic of social scientists of the time of Spencer, because this continuity is a consequence of a rapprochement in which both sciences have actively participated. We shall see this in detail throughout this essay; at the moment let it suffice to mention the vocabularies of the two. It would not be possible, upon looking into the index of a scientific book, to tell whether it was a book on natural or social science if it contained only the following entries: constitution, dimension, experiment, field, migration, population, prediction, probability, space, statistics, vector. And the list of common terms is growing year by year.
It would be going too far to say that there is no difference between the two groups of sciences, but we should not overemphasize their differences, as was frequently done early in the century, or underestimate them, as has been fashionable since then. It is sound scientific procedure to substitute differences of degree for differences in kind whenever apparent differences in kind can be interpreted as consequences of variation of some common factor. The common variable that I believe will account for both the unmistakable differences and the current rapprochements between the natural and the social sciences is “complexity of subject matter.” It is my belief that the major differences between them are due to the greater complexity of the subject matter of the social sciences, and that differences of method and interpretation of results are due primarily to awareness of this difference.
If this is correct, we should be able to test it empirically, by seeing whether the social sciences, when they deal with simple subject matters, are able to approach the natural science methods, and whether the natural sciences, when, they deal with complex subject matters, appropriate social science methods. Let us, then, turn to an examination of their respective subject matters in order to answer the question: Will differences in complexity account for differences in their observational, experimental, and conceptual techniques?
When we think of the social sciences as only the “poor relations” of the natural sciences, we forget that an insight into the order of society was prior to that into nature. Every primitive people sees nature by an analogy with its social organization. Science began when laws, like those given by governments and tribunals, were projected into nature.
The great Greek philosophers approached nature with the anticipation that it would conform to simple principles, some aspects of their society providing them a model for the interpretation of nature. Anaximander (ca. 550 B. C.), in an epoch-making analogy, held that changes in nature are regulated by justice, anticipating the function later ascribed to laws. Henceforth nature was to be seen as a cosmos.
But in searching for regularity and simplicity and lawfulness, the philosophers and early scientists found that they had to work with abstractions from observations and not with complex observations themselves. From the time of Galileo, at the latest, we feel that the “right abstractions” were made, because he chose to report those aspects of his observations which could be related to each other by simple mathematical laws.
Since Galileo, the subject matter of the natural sciences has been relatively simple and repetitive series of simple events. Such series are repetitive because they can be isolated from many other events and understood without reference to them. The natural sciences deal with isolated systems, because the variables they choose to observe are controllable by means of varying other chosen variables. Solely for this reason are simple experiments possible. Only a small number of variables have to be known for us to give functional laws relating one to another.
Certainly every event in nature is related to untold number of others, perhaps even to every-
thing else in nature. But by abstraction and material isolation, we are able to reduce the effects of most of the others to negligible quantities, and to attend only to the functional relations of certain chosen events. In the natural sciences, lack of repetitiveness in a series of events, as this occurs when an experiment “turns out wrong,” is always taken as evidence that the systems were not isolated, and we thereupon carry through a process of successive approximations toward complete isolation.
Nature not only has serial orders which can be studied in relative isolation; the things of nature also come in “vertical” arrangements, or wholes with contemporaneous parts. Field concepts, rather than merely serial concepts, apply to this aspect of nature. Because we can isolate systems, we can determine the boundaries of these fields, and eliminate the factors which would make our study of a given whole unmanageably complex. By moving the “isolation partitions,” we can determine experimentally the effects of parts on wholes and wholes on parts, even though we never deal with an entity which is not a part of some whole.
The subject matter of the social sciences, on the other hand, consists of highly complex constellations of complex events in systems that are only poorly isolated. Instead of indistinguishable atoms, as the chemist considers his subject matter, the social scientist must deal with societies of individuals of almost infinite internal complexity and variability. No one has yet made the fortunate discovery comparable to that of Galileo in physics: though we know that science cannot deal with an unlimited number of variables, no social scientist has yet shown us precisely which ones to choose to interrelate and which ones may be safely neglected.
When we try to isolate systems in the social sciences, we therefore do not know what to include in them and what to try to eliminate. We cannot move our partition boundaries at will, because the contexts within which we find human beings are not variable to such an extent that we can try out many different wholes for a single part. We cannot isolate a child from all social environments to see where the partition between eliminable and noneliminable environmental factors should be drawn. Until we do so, however, we have no generally acceptable rule by which we can decide what factors to include in our descriptions of the relevant environment or social field. We have parts always within wholes; and, though the social sciences have advanced on the basis of this recognition, which has often in the past not been given sufficient weight, it is hard to specify the relevant part-whole relation because it always obtains.
Let us not overlook the fact that these differences are differences of degree, and that as the social sciences approach the stage where they may be able to decide which few variables may be most profitably observed, the natural sciences are undergoing developments of techniques for taking more and more variables into account. It is now recognized that the high regularities of the physical sciences are only statistically simple; as the physical scientist gets closer to the individual object, as it were, the complexities that had been neglected before reappear. Instead of attending only to serial collocations of simple events, the physicist is now finding it necessary, in spite of all his efforts, to deal with field concepts and probabilities as ineluctable parts of his conceptual system. We should not forget that “statistics” is originally a concept and technique of social science, and its use in physical science signifies an often overlooked appropriation of social science methodology.
Observation in the natural sciences differs widely from that of everyday life. Most observations in natural science are instrumental results, usually observations of pointer readings. The major part of natural science work is not the taking of observations, but deciding what to observe and constructing instruments to make the observation. The observations of the natural scientist, therefore, are never the raw data or brute facts of common sense; they come to him already conceptually transformed and instrumentally abstracted from irrelevancies. They are what Loewenberg has aptly called “postanalytical data.” In getting these postanalytical data, the scientific instrument reduces the subjective contribution of the observer almost to zero and “narrows the field of vision” to a specific observable event uniquely correlated with some unobservable we are interested in measuring.
Until about a century ago, observations in the social sciences hardly differed at all from those of everyday life. The student of social phenomena observed the phenomena of society as a physician would observe a patient if he had no thermometer or laboratory reports. The data of the social studies were “preanalytical.” Where the contribution from the object ended and that from the observer began, no one could tell. Because there was no standardized instrument to narrow the field of vision to specific and relevant phenomena, the facts of so-
cial science might vary from common-sense observations to the narrow observations of a man with an idée fixe. The facts of the social studies were about as objective as journalistic observation, and no science could be based on such unstable and disputable facts.
As the social studies became scientific, they did so in part by the use of instruments. Usually these were not “brass instruments,” but conceptual devices that served comparable purposes - reducing the subjective contribution to observations, and abstracting the desired observable from irrelevant data normally given along with it. But these conceptual devices served the same purpose as physical instruments: they gave an indisputable post-analytical datum which seemed to be uniquely correlated with some vague preanalytic observation or with some wholly unobservable entity in which the scientist was interested.
Consider an intelligence test, perhaps the most nearly perfect of all social science instruments. For a quality not directly observable but the object of many common-sense judgments, the test substitutes a postanalytical datum, a ratio between two observed quantities, namely, age and a set of marks on a paper. The set of marks and this ratio are obtained by standardized and conventional procedures. “Intelligence” is not only measured by this instrument; it is operationally defined by the methods used to measure it. Until the test is devised, “intelligence” is not a part of scientific vocabulary at all.
Even with this instrument, the results still differ widely in scientific standing from those obtained with, say, a galvanometer. The galvanometer substitutes a postanalytical datum, a number, for a preanalytical datum, the shock we all feel when we hold a wire under some conditions. The galvanometer standardizes the conditions, eliminates subjective differences between observers, disregards irrelevancies such as the “appearance” of the circuit, and gives us a “hard” and indisputable datum. With a galvanometer, we can forget all about the original shock we felt. But with an intelligence test, we still think that it is measuring something that we already knew about, and if its results conflict too widely with those of our common sense, we decide the instrument must be changed. The social scientist simply does not trust his instruments as much as the natural scientist trusts his. The social scientist rightly reserves some insight against the reduction that his instrument would effect. However much the instruments of social science localize and control the subjective contribution to observations, the design, choice, and evaluation of instruments still depend upon the same kind of insight that social philosophers have always possessed or claimed; otherwise the results of instrumental observation may be very neat and elegant, but they have no noticeable relevance to the prescientific problems which led to the development of these, rather than other instruments.
Hence the social scientist, equipped with the finest batteries of tests, is still in the position of the legendary people who wished to weigh a pig very accurately. They planed the board to which the pig was to be tied until it was of identical thickness, measured in “milli-micro-mulahs,” throughout its length; they used as counterweights stones whose sphericity had been established within limits of one “milli-micro-mulah” they carefully balanced the pig and board against the stones - then they asked the first stranger who came along to estimate the weight of the stones.
Because the operational definitions of the objects of natural science are applied to terms of no great emotional significance, and are definitions of which there are no counterparts in everyday language, we tend to forget that the way in which the natural scientist has obtained them is logically not unlike that of the social scientist or the legendary pig-weighers. The natural scientist’s objects themselves do not determine what aspects shall be observed. The instruments he uses are extensions or projections of the questions he asks. With other questions, there would be other instruments and other data. The choice of his instruments is not ultimately determined by the object, but by the kind of answers he wants. In this respect he is exactly like the social scientist.
But here, again, the natural scientist knows better what he is looking for. As he is interested in correlating his data in simple functional laws, he is interested only in an instrument whose reading will be a variable in an equation by which he can predict what the reading on another instrument will be. He uses only those types of instruments which will give him such results; even further, he uses only those specific instruments which will give him those results, and sends the others to the shop. The social scientist, however, is lucky
if he possesses even a single instrument for getting data. Outside a few fields, such as that of factor analysis in psychometrics, he must correlate his instrumental results with his vague common-sense preanalytical observations; he therefore has little or no check on the accuracy of his instruments. In consequence, although the introduction of instruments into the armory of social sciences has given intersubjectively valid data which the social scientist did not formerly have, it has not permitted him to state categorically what is the conceptual significance of his results. He must still “estimate the weight of the stones.”
Hence observation in both the natural and the social sciences necessarily involves a subjective element of choice of observable variables. But, whereas in the natural sciences this choice is constantly modifiable by reference to other chosen observations, in the social sciences the choice is usually corrigible only by reference to the “enlightened common sense” of the observer, which tells one social scientist (but unfortunately often him alone) what weight is to be attached to the results, which observations are worth getting, and which ones are not. We can see the reason for this in the differences in complexity of the sciences: the instrumental “sieving” of the facts of nature is very precise and fine-grained, whereas the facts of society are large-grained and recalcitrant to narrow abstractive procedures, whether instrumental or conceptual.
Often the contrast between the natural and the social sciences is as succinctly drawn as that between experimental and observational sciences. But there are nonexpenimental natural sciences, and there are experiments in the social sciences. This contrast, therefore, is not perfect; but it throws light on another consequence of the different complexity of the two kinds of science.
The natural sciences, as we have said, can establish physically isolated systems in which only a small number of variables play a significant role; therefore, an experimental determination of their correlation is possible. The social sciences cannot physically or temporally isolate their subjects. Though experiments may be performed under conditions of imperfect isolation, neither in physics nor in sociology would we know how much of the object we were experimenting with. The physicist can meet this objection by moving his partition boundaries; the sociologist cannot. An experiment on children puts the boundary, let us say, at 9: 00 AM, in a classroom; but the previous history of the child, the home conditions, the hereditary conditions, and so on are uncontrolled variables from which the subjects of the experiment are by no means isolated. The social scientist, therefore, has to perform the same experiment over and over again with the idea that the uncontrolled variables will be randomly distributed in the series and thus cancel each other out. Hence in the experiments of social science there is a large inductive element lacking in the interpretation of good experiments in the natural sciences.
In recent years the social scientists, especially Lewin, have developed techniques for deriving results from only one or a very small number of experimental situations. This is possible when there are a large number of variables within the “field,” so that some interconnection between them can be found and little or no recourse has to be made to relatively unknown variables outside the field. Work of this kind, in which the conceptual apparatus is adequate to the complexity of the subject matter, is one of the most encouraging signs of a further affiliation between the natural and the social science techniques. In contrast, when the external trappings of a natural science experiment are imitated so that only a few highly abstract data are obtained, the lack of isolation of the variables being measured really prevents the experiment from being comparable to those of the natural sciences.
There is another difference between the experiments of the two branches of science which is dependent upon differences in their complexity. Isolation from the operator is difficult to achieve in the social sciences; the adventitious circumstances of the experimental setup, the isolation partitions themselves, function as significant causal variables. From the experimental results we can extrapolate to “normal situations” only with a wide margin of error, since these variables may be very important in the experiment and wholly absent in the situation we wish to make predictions about.
In experiments in natural sciences, the experimental situation is comparable to the normal, or at least the effects of the experimental situation can usually be estimated and conceptually eliminated. Certainly putting a new meter into a functioning circuit affects the circuit as a whole, but this effect can be measured in other experiments on the meter itself and we can eliminate the interference.
As the physical sciences come to deal more and more with the “individual physical object” - e.g.,
a single particle - it is found that the experimental conditions may play a more disturbing role which cannot be eliminated. The Heisenberg principle of uncertainty is illustrative of this comparatively unusual situation in the natural sciences, but one very common in the social. The study of individual members of a population of electrons may suffer from many of the same disabilities as the study of human individuals in society. As physics turns its attention to the complexities of the individual case, and sociology finds itself able to deal with large numbers of cases, their operational conditions and results become more nearly comparable.
Each science begins with “middle-sized” facts, those which are within range of convenient observation. The middle-sized facts of physics have a specious simplicity because individual differences have been statistically canceled out; if physics, like the human sciences, had begun with the individual case, it is likely that it would have made no more rapid progress than sociology. The middle-sized fact of sociology is the small community, and this is more complex and variable than the individual particle in physics. As sociology approaches statistically evened-out states of affairs, it may approach the simplicity of classical physics, which dealt only with its evened-out, middle-sized facts.
The differences between the theoretical structures of the natural and the social sciences are even more obviously contingent: upon differences of complexity in their subject matter. We shall see this in two respects: the parsimony of the two systems, and the modes of explanation in the two systems.
First, a word about the theoretical structure of any science. In scientific research there are three types of hypotheses functioning. First, there is the substantive hypothesis, the hypothesis being tested. Second, there is an operational hypothesis, stating that if such and such things are done, such and such observable results should be attained, provided the substantive hypothesis is true. The operational hypothesis is always formulated as a basis for experiment or observation, and it is chosen in the light of the substantive hypothesis we wish to test. Finally, there are collateral hypotheses, which are not being tested at the moment, but which provide the route by which the mind moves from the substantive to the choice of the operational hypothesis.
To illustrate these hypotheses, let us take an exceedingly simple example. We have the substantive hypothesis “Salt is soluble in water.” We test it by performing an experiment based on an operational hypothesis: “If I put the crystals from this bottle into water, they will disappear.” How do we move from the former to the latter hypothesis, by which it is to be tested? We do so by means of certain collateral hypotheses, viz., “These crystals are salt,” “This liquid is water.”
A given hypothesis is not inherently substantive while another is always collateral. We may subject any one of them to test. In our previous example, for instance, we could test the hypothesis “These crystals are salt,” using the other hypothesis, “Salt is soluble in water,” as collateral.
When an observational result differs from the prediction from a set of hypotheses, it is always possible to choose whether we shall consider (a) the operational hypothesis to have failed (experimental error); (b) the substantive hypothesis, the one we intended to test, to be wrong; or (c) some collateral hypothesis, by virtue of which we choose this experiment, to be in error (systematic error).
If we decide on the first alternative, we are in effect “testing a fact by a theory.” This is sometimes necessary in even the best-organized sciences in order to avoid renegade instances and to give credit to the obvious fact that not all observations are equally trustworthy. But science becomes dogmatic if this procedure is always followed, because then there can be no occasion to modify a theory once adopted. We have already seen the difficulty of eliminating experimental error in the social sciences, and consequently in them frequent recourse is had to this expedient; if the result is not as predicted, we can always say that there were disturbing and uncontrolled factors, or the observer was inaccurate, or the like.
Assuming that the experiment has been done well, we then have a choice as to which of the other hypotheses is to be modified or rejected. In the natural sciences this choice can be made by performing still other experiments involving different collateral hypotheses (in our example, we could use crystals from another bottle), or by undertaking other experiments in which the collateral hypothesis is tested without reference to the hy-
pothesis in which we were originally interested. (In our example, we could undertake a chemical analysis of the crystals in the bottle to see if they are sodium chloride.) The result of this multiplicity of approach is that in the natural sciences there need be no untestable hypotheses, and every well-performed experiment is crucial for some hypothesis in the body of the science.
Because of the complexity of each hypothesis in the social sciences, testing seriatim is rarely possible. For instance, we wish to determine the existence or nonexistence of racial differences in intelligence. We give a test to a group of children of different races. Their marks differ significantly. Does that prove the hypothesis that there are significant differences? Not unless we assume the collateral hypothesis, namely, that the test is independent of cultural differences. Can we test that experimentally? Only by devising a test in which the different cultural groups make approximately the same marks. But usually we cannot independently control the racial and cultural components; therefore, we do not know which hypothesis - a hypothesis about our particular intelligence test, or a hypothesis about the intelligence of different races- -must be rejected.
Because some assumptions are untested in our experiment, there will be disagreement about them. The result is that we have “schools” of psychology and sociology (e.g., “racial theories” and “cultural theories”) that are distinguished by disagreement about collateral hypotheses which function as “presuppositions.” Crucial experiments which might resolve controversies between schools are thus almost unknown in the social sciences. The route by which we move from a substantive hypothesis to an observation or experiment is so circuitous, and involves so many assumptions, that experiments can usually be cited equally well by both sides in a controversy.
The hypotheses of the natural sciences are so simple that they can be tested seriatim; those of the social sciences are so complex and interpenetrating that we have to take them in families. Nevertheless, here again the difference is one of degree, and recent science is narrowing the distance between the two theoretical structures. The natural scientist now realizes that no hypothesis can be tested without assuming others, and ultimately a circle in testing is completed. There now exist in physics several alternative families of hypotheses in which the circle has been completed. All the observations of one are translatable into results of the other, though the two sets are not logically equivalent, and future observations may lead to decision between them. At present, the choice must be made in terms of their relative parsimony. Yet the estimation of the degree of parsimony involves aesthetic, procedural, and subjective considerations of elegance, ease of inference, and the like. The philosophy of science during this century has largely emphasized subjective elements in even the most objective sciences, and we find a prominent physicist speaking of science as “nature refracted through human nature.” If the subject matter of physics were as complex as that of the social sciences, this human refractivity and selectivity would be more obvious than it is. If the subject matter of physics permitted the same variety of abstractions to be parsimoniously organized, it is likely that the conceptual structure of the natural sciences might appear as arbitrary as that of sociology or political science.
Finally, we come to the general strategy of explanation in the two branches of science. I do not refer to the age-old problem of mechanical vs. teleological explanation, for this metaphysical controversy appears in both types of science. I refer rather to the logic of explanation. In the natural sciences, the chief mode of explanation is description of the more pervasive and abstract features of the situation, whereby prima-facie different states of affairs are described in the same terms. For instance, a freely falling body and the moon are special cases falling under Newton’s laws. Explanation in the natural sciences is therefore analytic or reductive, through discovery of common and simple conditions of diverse effects whose prima-facie description would involve a very large vocabulary. Hence a phenomenon in chemistry is explained when it is described in the simpler terms of physics; the motions of the planets and of bodies rolling down an inclined plane are explained when a common set of variables is discovered in the description of each phenomenon.
Certainly this relation between explanation and
description is met with also in the social sciences.
We would, for instance, describe war
and migration in quite diverse terms; but we might explain them in terms of a
condition not obvious in either but underlying both, e.g., “population
pressure.” We shall, in the following
section, deal with the limits of this type of explanation as one of the
unsolved problems in the logic of the social sciences.
Still, it must be admitted that at
least at present the common mode of explanation in the social sciences is not
reductive and analytic, but synthetic. That
is, we predict some event in terms of psychology alone; but for more complex
events we have to add to the psychological causes suffi -
cient factors to get to the effect we actually find. Thus we say that we must attend not only to the psychological conditions, but draw in also the sociological, the economic, and the like. What would be called explanation in the natural sciences is all too often seen as “oversimplification” in the social sciences.
The extent to which reductive techniques should be universally employed, especially in psychology in its relation to physiology, is one of the crucial problems in the philosophy of social science. Just as physiological description is translated into physiological explanation, it is often held that the logically simpler is everywhere the explanation of the more complex, and psychology must be “reduced” to physiology. If this is the case, then of course there is no autonomous social science; it is simply a division of labor to be tolerated only until the natural sciences are able to effect a reduction. Such reducibility, if it exists, strengthens the thesis that the difference between the two branches of science lies in their differing complexity. The argument of the reductionist is that in the future the natural sciences will become better able to deal with states of affairs of high complexity, and the social sciences will have succeeded in conceptually diminishing the complexity of societal facts, so that the transition can be made. At present it cannot be done, perhaps because of the great disparity in degrees of complexity. Whether it can and should be done is one of the unsolved problems to which we now turn.
There are two problems we have lightly touched on, but which deserve more than passing notice even in a brief discussion. The first is strictly metaphysical: Are there any indigenous and irreducible categories of societal nature (e.g., culture, personality) that will successfully resist all attempts at reduction? Is the only difference between societal and natural reality a difference in complexity?
I have called this question metaphysical rather than scientific, for, whatever answer we give to it, the effect on scientific procedure will be the same. Different answers to this question will affect only the philosophical evaluation of the findings and procedures of the social sciences. Admitting irreducible categories would not in the least exempt the social scientist from reducing all that he can in order to increase the likelihood that the remaining ones are irreducible and not simply nonreduced. He would still do everything in his power to diminish the scope and importance of the not-yet-reduced concepts. Following the principle of parsimony, he would and should try to account for as much as possible by means of reductive explanation.
Analogous questions are met with in the natural sciences. The world of nature is not prima facie homogeneous, but has manifold discontinuities, levels of organization, and emergent properties. Acceptance of these with “natural piety” would have arrested the development of science. Yet the natural scientist does not have to explain them away in order to be scientific; he has only to attempt to explain them by showing the conditions under which they occur. Inasmuch as the necessary, but not the sufficient, conditions of life have been found in chemical studies, there still remains a task for the biologist - the description of his own phenomena, and the interrelation of them under unreduced biological categories. There can thus be purely biological explanation if he is successful in elaborating a general system of biological categories.
Similarly, in the social sciences it may be argued that, though it is important to know the natural conditions of societal phenomena, the social sciences have an indigenous subject matter, their own categories for its elaboration (e.g., “meaning” or “value”), and their own techniques for dealing with them (e.g., “understanding”). Much can be said for this point of view so long as it is not allowed to arrest the reductive procedures by which societal phenomena can be related to those of non-human nature. We have to deal here with two basic principles of method.
In the logic of science there is a principle as important as that of parsimony: it is that of sufficient reason. The former directs us to look for simplest causes; the latter cautions us not to simplify so far that the explanation is inadequate to the facts to be explained. Opposition to the hegemony of reductionism, insisting on the autonomy of social science categories, emphasizes the importance of the maxim that the adduced reasons shall be sufficient, rather than that they shall be parsimonious. Parsimony is not itself a simple criterion of a good methodology; we cannot simply count the factors of explanation and say that the theory containing
the smallest number is the best, The ideal of parsimony cannot be expressed without the proviso that the conditions for which it is a norm shall themselves be adequate. But if simplicity is difficult to define adequately, how far from simple it is to define adequacy!
Whether an explanatory system is adequate depends in the final analysis on what we want of an explanation. No one holds a brief -for an “autonomous chemistry” and for the “indigenous and irreducible facts of chemistry” and thus fights physicalistic reduction. But for practical, even if not for metaphysical, reasons, a comparable reduction of social science concepts to those of natural science may be quite legitimately resisted. Even if we overlook the possibility of nnetaphysical discontinuities between nature and man, the social sciences, if they are to be of use either practically or for the sake of an insight into social problems, are inextricably tied to enlightened common sense with its terminology. Operational simplification of sociological terms may be oversimplification in the sense that the problems as solved in the reduced vocabulary of natural science are not practically or intuitively equivalent or germane to the problems that originally led to the undertaking of the study. It may be that we ask for the bread of social insight and are given stones of natural science. It may be that an explanation of societal events in natural terms will have to be translated back into original language before anyone will admit that the explanation is adequate to the problem at hand,
Translations of this kind seem trivial and inadequate to the understanding of the initial problem. At the present stage of social sciences, then, it is quite defensible to hold that the explanatory concepts shall be germane to the motivating problem and not simply statements of correlations between societal and nonsocietal phenomena. This being the case, the tasks of the social sciences are the determination of adequate germane categories, such as culture, meaning, function, and value; their rigorous definition within the context of social science phenomena; their theoretical elaboration into parsimonious explanatory systems; and the establishment of rigorous procedural rules for their empirical application. For these tasks, the history of the more highly developed sciences may provide useful cues, but no more,
The second problem, though closely related to the former one, is logically independent of any answer we give to that question. Assuming that reduction will be practiced as far as possible, we still have to decide the proper procedures with respect to the concepts and hypotheses which have not yet been reduced. It is a question of the strategy of theorizing in the social sciences themselves. To be more specific: The social scientists now debate the question as to whether the chief desideratum is a general overarching theory or a series of particularistic hypotheses of relatively low degrees of generality.
The history of the natural sciences provides a valuable guide to the answer to this strategic problem. The physicists kept their hypotheses as close to observations as possible; their theories were integrations of hypotheses, not highly abstract summaries of concrete facts all on the same level. Galileo and Kepler had to do their theoretical as well as observational work before there could he Newton’s. But, we may be told, the social scientists now have almost as many hypotheses as facts; more unkindly, they may be said to be long on hypotheses and short on facts,
The mind of man, however, is not so prodigal of imaginative hypotheses that it can generate an infinite variety of them. Hypotheses show an inner kinship of common parentage in a given milieu and in the inventiveness of the social scientists - the demonstration of this being one of the great accomplishments of the sociology of knowledge. Hypotheses are increasing in number, but the variety of their types may be diminishing. General theory is not to be built by addition of hypotheses, except indirectly; it arises from their analysis and reduction,
Hence it may be expected that when a plethora of facts is elaborated in hypotheses of low generality, the broad outlines of an overarching theory may be subtly adumbrated. But in view of the complexity of subject matter, the looseness of theoretical structure, and the uncontrolled character of many of the observations of society, it is too soon to expect - indeed, it is too soon to be impatient for - a Newton of the social sciences.