Conceptualization in Comparative Politics

Abridged from Sartori Giovanni, "Concept Misformation in Comparative Politics", American Political Science Review, Vol. 64, No. 4, 1970, pp. 1033-1053.

Especially during the last decade comparative politics as a substantive field has been rapidly expanding. The scale, if not the scope, of this expansion raises thorny and unprecedented problems of method. But we seem to embark more and more in comparative endeavors without comparative method, i.e., with inadequate methodological awareness and less than adequate logical skills. That is to say, we seem to be particularly naive vis-a-vis the logical requirements of a world-wide comparative treatment of political science issues.

My focus is conceptual- about concepts - under the assumption that concepts are not only elements of a theoretical system, but equally tools for fact-gathering, data containers. The empirical problem is that we badly need information, which is sufficiently precise to be meaningfully comparable. Hence we need a filing system provided by discriminating, i.e., taxonomic, conceptual containers. If these are not provided, data misgathering is inevitable; and statistical, computerized sophistication is no remedy for misinformation. The theoretical problem can be stated, in turn, as follows: we grievously lack a disciplined use of terms and procedures of comparison. This discipline can be provided, I suggest, by awareness of the ladder of abstraction, of the logical properties that are implied, and of the rules of composition and decomposition thus resulting. If no such discipline is obtained, conceptual mishandling and, ultimately, conceptual misformation is inevitable (and joins forces with data misgathering).

The view presented in this article is that the political science profession as a whole is grievously impaired by methodological unawareness. The more we advance technically, the more we leave a vast, uncharted territory behind our backs. And my underlying complain is that political scientists eminently lack (with exceptions) a training in logic - indeed in elementary logic. I stress "elementary" because I do not wish to encourage in the least the over-conscious thinker, the man who refuses to discuss heat unless he is given a thermometer. My sympathy goes, instead, to the "conscious thinker", the man who realizes the limitations of not having a thermometer and still manage to say a great deal simply by saying hot and cold, warmer and cooler. Indeed I call upon the conscious thinker to steer a middle course between crude logical mishandling on the one hand, and logical perfectionism (and paralysis) on the other hand.

The lower the discriminating power of a conceptual container, the more the facts are misgathered, i.e., the greater the misinformation. Conversely, the higher the discriminating power of a category, the better the information. Admittedly, in and by itself this reply is not very illuminating, for it-only conveys the suggestion that for fact-finding purposes it is more profitable to exaggerate in over-differentiation than in over-assimilation. The point is, however, that what establishes, or helps establish, the discriminating power of a category is the taxonomical infolding. Since the logical requirement of a classification is that its classes should be mutually exclusive and jointly exhaustive, it follows from this that the taxonomical exercise supplies an orderly series of well sharpened categories, and thereby the basis for collecting adequately precise information. And this is indeed how we know whether, and to what extent, a concept has a fact-gathering validity.

Once again, then, it appears that we have started to run before having learned how to walk. Numbers must be attached - for our purposes - to "things," to facts. How are these things, or facts, identified and collected? Our ultimate ambition may well be to pass from a science "of species" to a science of "functional corelations". The question is whether we are not repudiating a science of species in exchange for nothing. And it seems to me that premature haste combined with the abuse of a quantitative idiom is largely responsible not only for the fact that much of our theorizing is muddled, but also for the fact that much of our research is trivial and wasteful.

All in all, and regardless of whether we rely on quantitative data or on more qualitative information, in any case the problem is the same, namely, to construct fact-finding categories that own sufficient discriminating power. If our data containers are blurred, we never know to what extent and on what grounds the "unlike" is made "alike." If so, quantitative analysis may well provide more misinformation than qualitative analysis, especially on account the aggravating circumstance that quantitative misinformation can be used without any substantive knowledge of the phenomena under consideration.

To recapitulate and conclude, I have argued that the logic of either-or cannot be replaced by the logic of more-and-less. Actually the two logics are complementary, and each has a legitimate field of application. Correlatively, polar oppositions and dichotomous confrontations cannot be dismissed: they are a necessary step in the process of concept formation. Equally, impatience with classification is totally unjustified. Rather, we often confuse a mere enumeration (or checklist) with a classification, and many so-called classifications fail to meet the minimal requirements for what they claim to be.

If quantification cannot solve our problems, in that we cannot measure before conceptualizing, and if, on the other hand, "conceptual stretching" is dangerously conducive to the Hegelian night in which all the cows look black (and eventually the milkman is taken for a cow) then the issue must be joined from its very beginning, that is, on the grounds of concept formation.

A few preliminary cautions should be entered. Things conceived or meaningfully perceived, i.e., concepts, are the central elements of propositions, and - depending on how they are named provide, in and by themselves, guidelines of interpretation and observation. It should be understood, therefore, that I shall implicitly refer to the conceptual element problems which in a more extended treatment actually and properly belong to the rubric "propositions." By saying concept formation I implicitly point to a proposition-forming and problem-solving activity. It should also be understood, in the second place, that my focus will be on those concepts which are crucial to the discipline, that is, the concepts which Bendix describes as "generalizations in disguise". In the third place, I propose to concentrate on the vertical members of a conceptual structure, that is, on 1) observational terms, and 2) the vertical disposition of such terms along a - ladder of abstraction.

While the notion of abstraction ladder is related to the problem of the levels of analysis, the two things do not coincide. A highly abstract level of analysis may not result from "ladder climbing". Indeed a number of universal conceptualizations are not abstracted from observables: they are "theoretical terms" defined by their systemic meaning. For instance the meaning of isomorphism, homeostasis, feedback, entrophy, etc., is basically defined by the part that each concept plays in the whole theory. In other instances, however, we deal with "observational terms," that is, we arrive at highly abstracts levels of conceptualization via ladder climbing, via abstractive inferences from observables. For instance, terms such as group, communication, conflict, and decision can either be used in a very abstract or in a very concrete meaning, either in some very distant relation to observables or with reference to direct observations. In this case, we have, then, "empirical concepts", which can be located at, and moved along, very different points of a ladder of abstraction. If so, we have the problem of assessing the level of abstraction at which observational or (in this sense) empirical concepts are located, and the rules of transformation thus resulting.

High level categorizations obtain universal conceptualizations: whatever connotation is sacrificed to the requirement of global denotation either in space, time, or even both. HL concepts can also be visualized as the ultimate genus, which cancels all its species. Descending a step, medium level categorizations fall short of universality and thus can be said to obtain general classes: at this level not all differentiate are sacrificed to extensional requirements. Nonetheless, ML concepts are intended to stress similarities at the expense of uniqueness, for at this level of abstraction we are typically dealing with generalizations. Finally, low level categories obtain specific, indeed configurative conceptualizations: here denotation is sacrificed to accuracy of connotation. One may equally say that with categories the differences of individual settings are stressed above their similarities: so much so that at this level definitions are often contextual.

A couple of examples may be usefully entered. In a perceptive essay which runs parallel to my line of thinking Neil J. Smelser makes the point that, for purposes of comparability, "staff is more satisfactory than administration, and administration is more satifactory than civil service. This is so, according to Smelser, because the concept of civil service "is literally useless in connection with societies without a formal state or governmental apparatus." In this respect "the concept of administration is somewhat superior...but even this term is quite culture-bound." Hence the more helpful term is "Weber's concept of staff... since it can encompass without embarassment various political arrangements..." In my own terms the argument would be rephrased as follows. In the field of so-called comparative public administration, "staff" is the high level universal category. "Administration" is still a good travelling category, but falls short of universal applicability in that it retains some of the attributesassociated with the more specific notion of "bureaucracy." Descending the ladder of abstraction further we then find "civil service," which is qualified by ifs associations with the modern state. Finally, and to pursue the argument all the way-down to the low level of abstraction, a comparative study of, say, French and English state employees will discover their unique and distinguishing traits and would thus provide contextual definitions.

The example suggested by Smelser is fortunate in that we are offered a choice of terms, so that (whatever the choice) a different level of abstraction can be identified by a different denomination. The next example is illustrative, instead, of the far less fortunate situation in which we may have to perform across the whole ladder of abstraction with one and same term. In illustrating his caution that many concepts are "generalizations in disguise", Bendix comes across such a simple concept as "village". Yet he notes that the term village may be misleading when applied to Indian society, where "the minimum degree of cohesion commonly associated with this term is absent. Even in such a simple case, then, a scholar is required to place the various associations of "village" along an abstraction ladder in accord with the travelling extension afforded by each connotation.

In principle the extension of a concept should not be broadened beyond the point at which at least one relatively precise connotation (property or attribute) is retained. In practice, however, the requirement of positive identification may be too exacting. But even if no minimal positive identification can be afforded, I do not see how we can renounce the requirement of negative identification. The crucial distinction would thus be between 1) concepts defined by negation or ex adverse), i.e., by saying what they are not, and 2) concepts without negation, i.e., no-opposite concepts, conceptions without specified termination or boundaries. The logical principle involved in this distinction is that any determination involves a negation. According to this principle the former concepts are, no matter how broad, determinate; whereas the latter are indeterminate, literally without termination.

If this principle is applied to the climbing process along a ladder of abstraction - and precisely to the point at which ML categories are turned into HL universals - in the first instance we obtain empirical, universals. In the second instance, by contrast, we obtain universals, which lack empirical value (pseudo-universals for an empirical science). The reason for this is that a concept qualified by a negation may, or may not, be found to apply to the real world. Whereas a non-bounded concept always applies by definition: having no specified termination, there is no way of ascertaining whether it applies to the real world or not. An empirical universal is such because it still points to something. A non-empirical universal indiscriminately points to everything (as any researcher on the field soon discovers).

It should be noted that it is in the middle level where we are required to perform the whole set of operations that some authors call "definition by analysis". This operation include the process of defining a term by finding the genus to which the object designated by the word belongs, and then specifying the attributes which distinguish such object from all the other species of the same genus. David Apter complains that our "analytical categories are too general when they are theoretical, and too descriptive where they are not". I understand his complaint to apply to our disorderly leaps from observational, findings all the way up to universal categories and vice versa - by-passing the middle-range as it were the stage of definition by analysis.

The low level of abstraction may appear uninteresting to the comparative scholar. He would be wrong, however, on two counts. First, when the comparative scholar is engaged in field work, the more his fact-finding categories are brought down to this level, the better his research. Second, it is the evidence obtained nation by nation, or region-by-region (or whatever the unit of analysis may be) that helps us decide which classification works, or which new criterion of classification should be developed.

While classifying must abide by logical rules, logic has nothing to do with the usefulness of a classificatory system. Botanists, mineralogists and zoologists have not created their texonomical trees as a matter of mere logical unfolding; that is, they have not imposed their "classes" upon their animals, any more than their animals flowers or minerals) have imposed themselves upon their classifiers. Let it be added that the information requirements of such an unsettled science as a science of politics can hardly be satisfied by single-purpose classifications (not to mention single-purpose checklists). As I have stressed, we desperately need standard fact-finding and fact-storing containers (concepts). But this standardization is only possible and fruitful on the basis of "multi-purpose" and, at the limit, all-purpose classifications. Now, whether a classification may serve multiple purposes, and which classification fits this requirement best, this is something we discover inductively, that is, starting from the bottom of the ladder of abstraction.

The overall discussion is recapitulated in Table 1 with respect to its bearing on the problems of comparative politics. A few additional comments are in order. In the first place, reference to three levels of abstraction brings out the in "broad" and "narrow" meanings of a term. For this does not clarify, whenever this is necessary, whether we distinguish, 1) between HL universal and ML general conceptualizations, or 2) between ML genuses and species or, 3) between ML and LL categories, or even 4) between HL universal and LL configurative conceptualizations.

In the second place, and more important, reference to (he ladder of abstraction forcibly highlights the drastic loss of logical articulation, indeed the gigantic leap, implied by the argument that all differences are "a matter of degree". This cannot be conceded, to begin with, at the level of universal categories. But all differences cannot be considered a matter of more-or-less at the medium level either. At the top we inevitably begin with opposite pairs, with polar opposites, and this is tantamount to saying that the top ML categories definitely and only establish differences in kind. From here downwards definitions are obtained via the logic of classification, and this implies that a logic of gradation cannot be applied as long as we establish differences between species. Differences in degree obtain only after having established that two or more objects have the same attributes or properties, i.e., belong to the same species. Indeed, it is only within the same class that we are entitled - and indeed required - to ask which object has more or less of an attribute or property.

In principle, then, it is a fallacy to apply the logic of gradation whenever ladder climbing (or descending) is involved. If we are reminded that along the ladder we augment the extension by diminishing the denotation (and vice versa), what is at stake here is the presence or absence of a given property; and this is not a matter of degree, but a matter of establishing the level of abstraction. Hence it is only after having settled at a given level of abstraction that considerations of more-and-less correctly apply. And the rule of thumb seeing to be that the higher the level of abstraction, the less a degree language applies (as anything but a metaphor); whereas the lower level of abstraction, the more a degree optics correctly and necessarily applies, and the more we profit from graduation concepts.

In the third place, and equally important, reference to the ladder of abstraction casts many doubts on the optimistic view - largely shared by the methodological literature - that "The more universal a proposition, i.e., the greater the number of events a proposition accounts for, the more potential falsifiers can be found, and the more informative is the proposition. The sentence suggests a simultaneous and somewhat natural progression of universality, falsifiers and informative content. It seems to me, instead, that reference to the correct technique of ladder climbing (and descending) confronts us at all points with choosing between range of explanation (thereby including the explanation of the relationships among the items under investigation), and accuracy of description (or informative accuracy). By saying that the "informative content" of a proposition grows by climbing the abstraction ladder, we should not be misled into understanding that we are supplying more descriptive information. Hence it is dubious whether we are really supplying more potential falsifiers (let alone the danger of "overly universal" propositions of no informative value for which falsifiers cannot be found).

Before concluding it should not pass unnoticed that in this section I have never used

word "variable," nor mentioned operational definitions, nor invoked indicators. Equally, reference to gradation concepts and to considerations of more-or-less has been, so far, entirely pre-quantitative. What is noteworthy, then the length that has been traveled before entering the problems which seem to monopolize methodological awareness. There is nothing wrong, to be sure, in taking up an argument at whichever point we feel that we have something to say - except that the tail of the methodological cal argument should not be mistaken for its beginning. Since I have taken up the issue at an early stage, I cannot possibly carry it through to its end. It behooves me, nonetheless, to indicate how I would plug what I have said into what shall have to remain unsaid.

For one thing, it should be understood that considering conceptualize - the genus - have not eluded the consideration of variables, which a species. That is, a variable is still a concept but a concept is not necessarily a variable. If all concepts could be turned into variables, the difference could be considered provisional. Unfortunately, as a scholar well versed in quantitative analysis puts it, "all the most interesting variables are nominal."" Which is the same as saying that all the most interesting concepts are not variables in the proper, strict sense of implying "the possibility of measurement in the most exact sense of the word".

A closely linked and similar argument applies to the operationist requirement. Just as concepts are not necessarily variables, definitions are not necessarily operational. The definitional requirement for a concept is that its meaning is declared, while operational definitions are required to state the conditions, indeed the operations, by means of which a concept can be verified and, ultimately, measured. Accordingly we may use fully distinguish between definition of meaning and operational definition. And while it is obvious that. an operational definition still is a declaration of meaning, the reverse is not true.

The contention often is that definition of moaning represents a pre-scientific age of definition, which should be superseded in scientific discourse by operational definitions. However, this contention can hardly meet the problems of concept formation, and indeed appears to ignore them. As the ladder of abstraction scheme helps to underline, among the many possible ways and procedures of defining the ex adverso definitions and taxonomic unfoldings (or definition by analysis) some correspond to different levels of analysis and play, at each level, a non-replaceable role. Moreover operational definitions generally entail a drastic curtailment of meaning for they can only maintain those meanings that comply with the operationist requirement. Now, we are surely required to reduce ambiguity by cutting down the range of meanings of concepts. But the operational criterion of reducing ambiguity entails drastic losses in conceptual richness and in explanatory power. Take, for instance, the suggestion that "social class" should be dismissed and replaced by a set of operational statements relating to income, occupation, educational level, etc. If the suggestion were adopted wholesale, the loss of conceptual substance would be not only considerable, but unjustified. The same applies, to cite another instance, to "power." To be concerned with the measurement of power does not imply that the meaning of the concept should be reduced to what can be measured about power - the latter view would make human behavior in whatever collective sphere almost inexplicable.

It should be understood, therefore, that operational definitions implement, but do not replace, definitions of meaning. Indeed there must be a conceptualization before we engage in operationalization. As Hempel recommends, operational definitions should not be "emphasized to the neglect of the requirement of systematic import. This is also to say that definitions of meaning of theoretical import, hardly operational definitions, account for the dynamics of intellectual discovery and stimulation. Finally, it should be understood that empirical testing occurs before, and also without, operational definitions. Testing is any method of checking correspondence with reality by the use opertinent observations; hence the decisive difference brought about by operationalization is verification, or falsification, by measurement.