Conceptual "Stretching" Revisited: Adapting Categories in Comparative Analysis

Author(s): David Collier, James E. Mahon and Jr.

Source: The American Political Science Review, Vol. 87, No. 4 (Dec., 1993), pp. 845-855
Published by: American Political Science Association
Stable URL: https://www.jstor.org/stable/2938818
Accessed: 26-08-2020 00:07 UTC

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

American Political Science Association is collaborating with JSTOR to digitize, preserve and
extend access to The American Political Science Review

This content downloaded from 163.178.101.95 on Wed, 26 Aug 2020 00:07:40 UTC
All use subject to https://about.jstor.org/terms
American Political Science Review Vol. 87, No. 4 December 1993

CONCEPTUAL "STRETCHING" REVISITED:

ADAPTING CATEGORIES IN COMPARATIVE ANALYSIS
DAVID COLLIER University of California, Berkeley
JAMES E. MAHON, JR. Williams College

I A Then scholars extend their models and hypotheses to encompass additional cases, they
/fj\f commonly need to adapt their analytic categories to fit the new contexts. Giovanni
T v Sartori's work on conceptual "traveling" and conceptual "stretching" provides helpful
guidance in addressing this fundamental task of comparative analysis. Yet Sartori's framework draws
upon what may be called classical categorization, which views the relation among categories in terms
of a taxonomic hierarchy, with each category having clear boundaries and defining properties shared
by all members. We examine the challenge to this framework presented by two types of nonclassical
categories: family resemblances and radial categories. With such categories, the overly strict
application of a classical framework can lead to abandoning to category prematurely or to modifying
it inappropriately. We discuss solutions to these problems, using examples of how scholars have
adapted their categories in comparative research on democracy and authoritarianism.

table concepts and a shared understanding of historical analysis,2 as well as by the comparative
categories are routinely viewed as a foundation
politics literature on authoritarianism and corporat-
of any research community. Yet ambiguity, con- ism in the 1970s and on democratization in the 1980s
fusion, and disputes about categories are common in and 1990s.3 It is evident from these bodies of schol-
the social sciences. A major source of this difficulty is arship that broad comparison requires a use of cate-
the perpetual quest for generalization. As scholars gories that is sensitive to context. Further, the histor-
seek to apply their models and hypotheses to more ical depth in many of these studies offers a useful
cases in the effort to achieve broader knowledge, they reminder that the problem of conceptual stretching
must often adapt their categories to fit new contexts. can arise not only from movement across cases but
One of the most incisive treatments of this problem of also from change over time within cases. Hence, the
adapting categories is Giovanni Sartori's (1970, 1984) challenge of achieving the virtue of conceptual trav-
work on conceptual traveling (the application of con- eling without committing the vice of conceptual
cepts to new cases) and conceptual stretching (the stretching remains very much with us today.
distortion that occurs when a concept does not fit the We shall examine how categories change-or
new cases). should change-as they are applied to new cases.
This is an old debate, and it might appear that this Sartori's original framework is based on the assump-
problem of categorization has been superseded by tions of what is sometimes called classical categori-
new analytic and statistical approaches. However, zation, in which the relation among categories is
this is not the case. Scholars accustomed to the understood in terms of a taxonomic hierarchy of
language of "variables" will recognize that issues successively more general categories (1970, 1038).
raised here are closely related to problems of estab- Each category possesses clear boundaries and defin-
lishing the validity of observation and measurement ing properties that are shared by all members and
across cases. For example, analysts who have care- that serve to locate it in the hierarchy. Yet twentieth-
fully derived and tested a set of hypotheses about century linguistic philosophy and contemporary cog-
political participation in one set of cases will com- nitive science have presented a fundamental chal-
monly wish to probe the generality of their findings lenge to this understanding of categories by claiming
by examining the same hypotheses in additional that many types of categories do not possess these
cases. To do so, they must first establish that political attributes (Lakoff 1987). This challenge might seem to
participation has a sufficiently similar meaning in the undermine Sartori's approach. However, we show
new cases. An excessive concern with the difficulties that these alternative types of categories can be
of establishing equivalence among contexts of analy- treated in a way that is distinct from, yet complemen-
sis could, of course, lead to the abandonment of the tary to, Sartori's perspective.
comparative enterprise altogether. The merit of Sar- To provide a base line against which alternative
tori's approach is that it encourages the scholar to be perspectives on categories can be evaluated, we first
attentive to context, but without abandoning broad review Sartori's procedure for modifying categories.
comparison. We then explore the distinctive problems that arise in
In recent years, new interest in the problem of dealing with types of categories that do not fit the
applying categories across diverse contexts has been classical pattern, which is the basis of Sartori's ap-
generated by the rise of a school of comparative- proach. First, we examine the issues that arise with

845

This content downloaded from 163.178.101.95 on Wed, 26 Aug 2020 00:07:40 UTC
All use subject to https://about.jstor.org/terms
 Conceptual "Stretching" Revisited December 1993

"family resemblance" categories. This discussion

suggests that Sartori's procedure can be applied too
strictly, causing analysts to abandon a category pre- The Ladder of Generality
maturely when it initially does not appear to fit High
additional cases. We then consider what cognitive
scientists call the "radial" category,4 providing a
rationale for why different types of categories (e.g., 0

democracy, as opposed to authoritarianism) are mod-

ified in distinct ways as they are adapted to fit new
'U
cases. We conclude by suggesting new guidelines for
adapting categories in the process of comparative Low A
analysis.
Low High

INTENSION 2
AVOIDING CONCEPTUAL
STRETCHING: SARTORI'S A = Initial category
B = Category adapted to more cases
FRAMEWORK
1 Range of cases
2 Number of defining attributes
A central element in the classical view of categories,
which provides the underpinning for Sartori's frame-
work, is the understanding of extension and intension
(Sartori 1970, 1041; idem. 1984, 24). The extension of they reduce the intension to the degree necessary to
a category is the set of entities in the world to which fit the new contexts. For example, scholars engaged
it refers. The intension is the set of meanings or in a comparative study of patrimonial authority might
attributes that define the category and determine add cases to their analysis that only marginally fit this
membership. category. To avoid conceptual stretching, they might
Two complementary patterns in the relation be- move up the ladder of generality and refer to the
tween extension and intension are of concern here, larger set of cases as instances of traditional author-
namely, the occurrence of (1) more specific categories ity. This interplay between extension and intension
with more limited extension and greater intension on the ladder of generality is illustrated in Figure 1.
and (2) more general categories with greater exten- With the categories obeying the law of inverse varia-
sion and more limited intension. Some philosophers tion, the ladder of generality appears as a line of
have held that these patterns reflect a "law of inverse negative slope.6
variation" (Angeles 1981, 141). In a taxonomic hier- In short, this framework helps researchers proceed
archy, these more specific and more general catego- with greater care when addressing one of the basic
ries occupy subordinate and superordinate positions, tasks of comparative research: the effort to achieve
with the extension of the subordinate categories broader knowledge through analyzing a wider range
contained inside the superordinate ones. The hierar- of cases. The value of this framework merits empha-
chy represented by these sets of terms can be called, sis particularly in light of skepticism, on the part of
adapting Sartori's label, a "ladder of generality."5 scholars committed to an "interpretive" perspective,
An example will serve to illustrate these patterns. about the possibility of achieving a viable compara-
In Max Weber's famous typology, patrimonial au- tive social science (Geertz 1973, 1983; MacIntyre 1971;
thority is a type of traditional authority, which is one Rabinow and Sullivan 1987; Taylor 1971; Winch 1959).
of his three overall types of authority or legitimate Sartori's framework addresses some of the important
domination, which, in turn, is one type within the concerns raised by this perspective, namely, that
broader category of domination (1978, 212-15, 226, broad comparison is difficult, that political and social
231). Within each successive pair of categories, the reality is heterogeneous, that applying a category in a
first is subordinate, the second, superordinate. In given context requires detailed knowledge of that
relation to each subordinate category, the corre- context, and that it is easy to misapply categories.
sponding superordinate category contains a less spe- The ladder of generality offers a specific procedure to
cific meaning and covers more cases; thus, it has address these issues. This procedure has deservedly
greater extension and less intension. served as a benchmark for analysts who wrestle with
This classical understanding of categories helps the problem of extending categories to new cases.
address the problem of conceptual stretching, When
scholars take a category developed for one set of
cases and extend it to additional cases, the new cases FAMILY RESEMBLANCE CATEGORIES
may be sufficiently different that the category is no
longer appropriate in its original form. If this problem The application of the ladder of generality assumes
arises, they may adapt the category by climbing the the clear boundaries and defining attributes of classi-
ladder of generality, thereby obeying the law of cal categories. An exploration of family resemblance
inverse variation. As they increase the extension, categories shows that at times, this assumption

846

This content downloaded from 163.178.101.95 on Wed, 26 Aug 2020 00:07:40 UTC
All use subject to https://about.jstor.org/terms
 American Political Science Review Vol. 87, No. 4

An Iterated Generalization: The Case of Family Resemblance

True Cases Included Attributes Included
Distribution of in Generalization in Generalization
Cases Attributes (Extension) (Intension)

A 1 2 3 4 5 A 1 2 3 4 5

B 1 2 3 4 6 A B 1 2 3 4

C 1 2 3 5 6 A B C 1 2 3

D 1 2 4 5 6 A B C D 1 2

E 1 3 4 5 6 A B C D E 1

F 2 3 4 5 6 A B C D E F

* All cases have five of the six attribu

should be relaxed. Ludwig Wittgenstein's idea of sesses a different combination of only five of these.
family resemblance entails a principle of category No attribute is shared by all cases.
membership different from that of classical catego- Using this example (see Figure 2), we will examine
ries, in that there may be no single attribute that the consequences if the analyst were to rigidly apply
category members all share. The label for this type of the ladder of generality. If the original case-study
category derives from the fact that we can recognize research were done on Case A, the intension of the
the members of a human genetic family by observing initial category would encompass Attributes 1-5.
attributes that they share to varying degrees, as Upon adding Case B to the analysis, other analysts
contrasted to nonfamily members who may share few might note that Attribute 5 was lacking. They could
of them. The commonalities are quite evident, even seek to avoid conceptual stretching by climbing the
though there may be no trait that all family members, ladder of generality to a category that encompassed
as family members, have in common (Wittgenstein both cases (A and B) and whose intension was
1968, nos. 65-75; see also Canfield 1986 and Hallett reduced to Attributes 1-4. Adding Case C could lead
1977, 140-41, 147-48).7 to a further step up the ladder to a still more general
A similar pattern often appears in the social sci- category that encompassed only Attributes 1-3. As
ences. A category, defined in a particular way, may can be seen in the figure, when this iterated process
fit a number of cases reasonably well, but on close finally reached Case F, the final step up the ladder of
examination it can become clear that for most cases generality would bring the elimination of the final
the fit is not perfect. Nonetheless, the category cap- trait, leaving a category with no attributes. Thus, the
tures a set of commonalities considered by the re- analyst might abandon the category prematurely.
searcher to be analytically important. This pattern is The example in Figure 2 serves as a warning that, in
found, for example, in the literature on corporatism, the course of applying a category to additional cases,
which generally presents a series of defining at- it can be counterproductive to insist on eliminating
tributes, usually without the expectation that the full those attributes not held in common by all the cases
set of attributes would be found in every instance under consideration.
(Malloy 1977; Schmitter 1974). Thus, over many dec- One way of avoiding this problem is to look at the
ades during the twentieth century, it was reasonable larger set of cases simultaneously, so that the com-
to characterize labor relations in Argentina, Brazil, monalities evident in Figure 2 would be recognized.
Chile, and Mexico as corporative, despite variation in Yet because every case is missing at least one at-
the features of corporative structuring, subsidy, and tribute, a researcher accustomed to thinking in terms
control of groups found in the four cases (Collier and of classical categories might still conclude that this is
Collier 1991). a weak category that should be abandoned. A possi-
What would happen if we applied Sartori's method ble response would be to emphasize that the category
to a family resemblance category? Let us consider a is an analytic construct which the researcher should
hypothetical exercise in comparative analysis. Sup- not expect to be a perfect description of each case. A
pose that (1) the analyst begins with a case study well-known example of this kind of construct is the
yielding a new category of theoretical interest initially ideal type, of which each specific case is expected to
appearing to have five defining attributes, (2) the be only a partial approximation.8
initial case is one of six cases that share a family Some of the creative approaches to the refinement
resemblance, (3) this family resemblance turns out to of categories in the field of comparative politics can be
entail six shared attributes, and (4) each case pos- seen as attempts to deal with family resemblance.

847

This content downloaded from 163.178.101.95 on Wed, 26 Aug 2020 00:07:40 UTC
All use subject to https://about.jstor.org/terms
Conceptual "Stretching" Revisited December 1993

Przeworski and Teune argue that in comparative understood together, and most quickly recognized
research, conceptualization and measurement at when found together. "Noncentral subcategories"
times require a "system-specific" approach. They are variants of the central one. They do not necessar-
suggest, in effect, that in diverse contexts different ily share defining attributes with each other but only
attributes can be used as defining properties of the with the central subcategory-hence the term radial,
same category.9 Nie, Powell, and Prewitt employ a which refers to this internal structure.
similar perspective in comparing political participa- One of Lakoff's ordinary-language examples of a
tion in the United States and four other countries radial category is "mother" (1987, 83-84). Here the
(1969, 377). For all the countries, their analysis fo- central subcategory corresponds to an individual
cuses on four relatively standard attributes of partic- who, in the context of conventional gender relations
ipation. However, in analyzing a fifth attribute- in the United States, is often considered a "true"
membership in a political party-they observe that mother-that is, one who (1) is a woman, (2) contrib-
whereas in four of the countries it has a roughly utes half the child's genetic makeup, (3) bears the
equivalent meaning, party membership in the United child, (4) is the wife of the father, and (5) nurtures the
States has a significantly different form and meaning. child. The noncentral subcategories arise when the
The authors conclude that in the United States, component elements are taken singly, or in sets of
involvement in electoral campaigns reflects an equiv- two or more. In this example, familiar types emerge if
alent form of political participation. Hence, for that these roles are taken singly: "genetic mother," "birth
country, they analyze campaign participation instead mother," "stepmother," and "nurturing mother."
of party membership. Radial categories merit attention here because they
As in this last example, it is evident that family play an important role in the language of social
resemblance can sometimes be assessed by identify- science. For example, following Ostiguy (1993), one
ing attributes that are present to varying degrees in can view "democracy" as a radial category. Obvi-
particular cases, rather than being simply present or ously, the problem of identifying the components of
absent. This can be accomplished by applying some democracy has long been a matter of debate. For
form of multidimensional scaling that specifies un- present illustrative purposes, the following partial
derlying dimensions for comparing cases. Yet it is definition will suffice.'2 We might say that the central
important to remember that multidimensional scaling subcategory "democracy" is constituted by elements
does not eliminate the original problem of forming such as (1) broad and effective participation in the
the concept. In the spirit of Sartori's dictum "concept process of rule, (2) limitation of state power and
formation stands prior to quantification" (1970, 1038), protection of individual rights, and (3) according to
one must recognize that a prerequisite for such scal- some accounts, egalitarian (or at least relatively more
ing is to establish what it is that is being scaled.'0egalitarian) economic and social relationships. The
When the analyst encounters a family resemblance first component taken alone might be seen as consti-
pattern, two priorities must be addressed. First, in tuting the noncentral subcategory "participatory de-
assessing the attributes empirically, one must avoid mocracy," the first and second combined as consti-
an application of the ladder of generality that is so tuting "liberal democracy," and the first and third
strict as to result in the inappropriate rejection of a combined as constituting what may be called "popu-
potentially useful category. Second, it is essential to lar democracy."
explore the underlying analytic relationship among
the attributes that constitute the family resemblance,
Comparing Radial and Classical Categories
thereby establishing the justification for retaining the
category. A concern with this analytic relationship is The internal form of radial categories differs from that
central to the discussion of radial categories, to which of classical categories. The variants that branch out
we now turn. within a radial structure such as "mother" or "de-
mocracy" might be viewed as subsets of the overall
category. Yet they do not share the full complement
RADIAL CATEGORIES of attributes by which we would recognize the overall
category, as they do with classical categories. Rather,
Another type of category that does not fit the classical they divide them. This difference has important impli-
pattern is the radial category, analyzed by cognitive cations for how these two types of categories are used
scientists such as Lakoff (1987, chap. 6). As with in comparative analysis.
family resemblance, with radial categories it is possi- Before radial and classical categories are compared
ble that two members of the category will not share further, an issue of labeling should be clarified. We
all of what may be seen as the defining attributes. In have referred to the component elements of classical
contrast to the family resemblance pattern, with categories as superordinate and subordinate,
radial categories the overall meaning of a category is whereas for radial categories we have referred to
anchored in a "central subcategory," which corre- central and noncentral subcategories. For the sake of
sponds to the "best" case, or prototype, of the comparison, we can apply more generic labels (see
category." In the process of cognition, the central Figure 3). The term primary category will be used to
subcategory functions as a gestalt, in that it is consti- refer to the overall category, whereas secondary cate-
tuted by a bundle of traits that are learned together, gory will be used to refer to the category whose

848

This content downloaded from 163.178.101.95 on Wed, 26 Aug 2020 00:07:40 UTC
All use subject to https://about.jstor.org/terms
American Political Science Review Vol. 87, No. 4

may now be examined, first using Lakoff's examples

of "dog," a classical category in the scheme of tradi-
Generic Labels for Comparing Classical and Radial tional taxonomy, and "mother," a radial category
Categories (1987, 46, 74-76). As can be seen in Figure 4, in the
Classical Radial Generic Label case of the classical category of "dog" the differenti-
ating attributes of the secondary categories occur in
Superordinate Category -f- Primary addition to those of the primary category. By contrast,
Central Subcategory C[ategory with the radial category of "mother" the differentiat-
ing attributes of the secondary categories are contained
Subordinate Category S secondary
within the primary category.
Noncentral Subcategory CF .ategory
The examples of "authoritarianism," understood
as a classical category,'3 and "democracy," under-
stood as a radial category, yield the same contrast
meaning is derived from the primary category. Thus, (Figure 5). In relation to authoritarianism, the differ-
"mother" and "democracy" are primary categories, entiating attributes of the secondary categories of
and "birth mother" and "liberal democracy" are the '"populist" and "bureaucratic" authoritarianism occur
corresponding secondary categories. In the realm of in addition to those of the primary category. In the case
classical categorization, "authoritarianism" is a pri- of "democracy," the differentiating attributes associ-
mary category, and 'bureaucratic authoritarianism" ated with "participatory," "liberal," and "popular"
is the corresponding secondary category. democracy are contained within the primary category.
The contrast between classical and radial categories This contrast between the two types of categories

Differentiation of Classical and Radial Categories

Classical Category: Dog

Category Components

Primary Category Dog A B C.

Secondary Categories Retriever A B C D

Sheepdog A B C E

Spaniel A B C F

Note: Differentiating characteristic

A, B, and C = Hypothetical set of general attributes of dogs

D, E, and F = Hypothetical attributes that differentiate specific types of dogs

Radial Category: Mother

Category Components

Primary Categorv Mother A B C D E

Secondary Categories Genetic mother A B

Birth mother A C

Nurturing mother A D

Stepmother A. E.
.............

Note: Differentia

A = Female
B = Provides 50% of genetic makeup
C = Gives birth to child
D = Provides nurturance
E = Married to father

849

This content downloaded from 163.178.101.95 on Wed, 26 Aug 2020 00:07:40 UTC
All use subject to https://about.jstor.org/terms
 Conceptual "Stretching" Revisited December 1993

Differentiation of Classical and Radial Categories: Exampl

Classical Category: Authoritarianism

Category Components

Primary Category Authoritarianism A B

Secondary Categories Populist Authoritarianism A B C

Bureaucratic Authoritarianism A B D

Note: Differentiating characteristics of secondary categories are in addition to those of the p

A Limited pluralism
B 5 Distinctive mentalities, not guiding ideology*
C = Substantial mobilization of working class and/or middle class
D Alliance of military, technocrats, and transnational capital against previously mobilized popular classes

Radial Category: Democracy

Category Components

Primary Category Democracy A B C

Secondary Categories Participatory democracy A

Liberal democracy A B

Popular democracy A C

Note: Differentiating characteristics of secondary categories are contain

A = Effective political participation

B = Limitation of state power
C - Social and economic outcomes of relative equity

* The definition of authoritarianism employs two elements used by Linz (1975).

has a major practical consequence in terms of how Authoritarianism Versus Democracy: Contrasting
we go about addressing the problem of conceptual Patterns of Category Change
stretching: the extension of the secondary category
in radial categorization may exceed that of the pri- Let us apply these ideas to two examples of concep-
mary category. Consider an example from common tual traveling. During an earlier period of wide inter-
usage: a woman who is a birth mother might not be est in bureaucratic authoritarianism, that category
seen as fitting the overall category of what is under- was at times extended to cases that only marginally fit
stood as a "true" mother.'4 All mothers that fit the the original meaning (Collier 1979, 1993). Using the
primary category (i.e., all "true" mothers) are birth ladder of generality, scholars sometimes avoided this
mothers, but the converse is not the case. Hence, problem of conceptual stretching by shifting to the
there are more birth mothers in the world than "true" broader category of authoritarianism.
mothers. A parallel problem has arisen with recent efforts to
The same pattern appears with democracy. If only apply the category "democracy" to new regimes in
the extensive political participation associated with Central America, Eastern Europe, and the former
democracy is present in a given country, without Soviet Union. In some of these cases, where leaders
protection of the rights of those who at any given are selected in competitive elections but where many
time may be in a minority, many observers will of the institutions and practices often associated with
conclude that it is not what they consider to be a democracy are absent, the problem of conceptual
"true" democracy. Here again, the extension of the stretching may be addressed by making the more
secondary category will exceed that of the primary modest claim that these are, for example, "electoral
category, involving the same inverse relationship democracies," thereby abandoning the implication
between extension and intension already discussed. that they are "true" democracies.

850

This content downloaded from 163.178.101.95 on Wed, 26 Aug 2020 00:07:40 UTC
All use subject to https://about.jstor.org/terms
 American Political Science Review Vol. 87, No. 4

Second, in the case of radial categories, the pos-

sibility of encompassing more cases through the
Avoiding Conceptual Stretching with Classical and elaboration of secondary categories can allow for
Radial Categories considerable flexibility regarding the meaning and
application of the category. Although this flexibility is
often desirable, it can be the source of major scholarly
High P
debates. For example, as scholars seek to identify
new subtypes of democracy, disputes can easily arise
z Classical as to whether it is appropriate to consider the cases
0 S
that fit these subtypes to be "truly" democratic.'5 By
U.1 contrast, in analyses of a classical category such as
?I.- Radial S "bureaucratic authoritarianism," no parallel debate
emerged about whether the cases of bureaucratic
Low P authoritarianism were instances of "true" authoritar-
ianism.16
Low High

INTENSION 2
P = Primary category
S - Secondary category Further Illustrations of Radial Structure in Recent
1 Range of cases
Discussions of Democracy
2 Number of defining attributes

Recent analyses of democracy by Terry Karl and

Philippe Schmitter illustrate some of the concomi-
tants of the radial structure of this category (Karl
1990, 2; Schmitter and Karl 1991, 76-82; idem 1992,
The contrast that emerges here must be kept in
52). Of the three attributes of democracy we have
mind as scholars seek to avoid conceptual stretching
discussed, Karl and Schmitter deliberately set aside
in comparative research. In the case of bureaucratic
the questions of equity raised above and focus on
authoritarianism, this undesirable outcome is
issues associated with participatory democracy and
avoided by moving up a ladder of generality from a
liberal democracy. Summarizing schematically, we
secondary category to the primary category. In the
case of democracy, it is avoided by moving away from may say that they are concerned with four elements:
the primary category to employ a secondary category. "(1) contestation over policy and political competition
Figure 6 summarizes this contrast between classical for office; (2) participation of the citizenry through
and radial categories in the framework of a diagram partisan, associational, and other forms of collective
of extension and intension like Figure 1. As can be action; (3) accountability of rulers to the ruled through
seen in Figure 6, with classical categories, to avoid mechanisms of representation and the rule of law"
conceptual stretching one moves from the secondary (Karl 1990, 2; emphasis added); and (4) protection of
category, S, to the primary category, P, by rising up a rights essential to meaningful contestation, participa-
line of negative slope like that of Figure 1. With radial tion, and accountability.'
categories, by contrast, to avoid conceptual stretching Karl explicitly notes what we see as an essential
one moves up a parallel line, but from the primary component of the radial structure of this category. In
category, P, to the secondary category, S. a discussion of subtypes of democracy (which we call
Two further contrasts between radial and classical the secondary categories), she observes that they "are
categories may be noted. First, an important differ- characterized by different mixes and varying degrees
ence is evident in how the formal label is modified as of the chief dimensions of democracy: contestation,
one moves from one level of generality to another. participation, [and] accountability" (1990, 2). Thus,
With both types of categories, it often (though not she recognizes the essential point evident in Figures 4
always) occurs that primary categories are made into and 5: secondary categories tend to divide up the
secondary categories by adding an adjective. Thus, component elements of the primary category, and
"bureaucratic authoritarianism" is a secondary cate- they may vary considerably in how closely they
gory in relation to "authoritarianism," and "electoral resemble the central subcategory.
democracy" is a secondary category in relation to This pattern also appears in the subtypes devel-
"democracy." This similarity helps to underline a oped jointly by Schmitter and Karl (1992, 56-58).
crucial contrast in how we move to a broader set of They identify "corporatist" democracy and "popu-
cases with classical, as opposed to radial, categories. list" democracy in part by the shared attribute that
In the example of the classical category "bureaucratic the dominant center of power is located in the state.
authoritarianism," this is done by dropping an adjec- Clearly, this attribute mitigates the weight of other
tive. By contrast, with the radial category "democra- components of their understanding of democracy,
cy," it is done by adding an adjective. Thus, the such as citizen participation and the accountability
analyst seeking to avoid conceptual stretching will of rulers. Thus, in their framework these subtypes
use adjectives in opposite ways, depending on the are less democratic than what might be deemed
type of category in question. to be "true" democracies. The fact that these sub-

851

This content downloaded from 163.178.101.95 on Wed, 26 Aug 2020 00:07:40 UTC
All use subject to https://about.jstor.org/terms
 Conceptual "Stretching" Revisited December 1993

types are seen as less democratic comes out clearly CONCLUSION

in Schmitter and Karl's empirical analysis of 24 recent
cases of democratization. Of the eight countries they The goal of this discussion has been to offer new
assign to the subtypes of "populist" and "electoral- guidelines for comparative analysts who are con-
ist" democracy, they treat six as marginal cases, cerned with the problems of conceptual traveling and
either because they have "not yet crossed the mini- conceptual stretching. We conclude that Sartori's
mal democratic threshold" or because they "are not framework for addressing these problems deservedly
yet consolidated into a recognizable type of democ- remains a benchmark for scholars of comparative
racy" (p. 68). politics. Yet some caution and refinement are in order.
The examination of family resemblances reminds
Schmitter and Karl's analysis brings us back to our
us that an overly strict application of classical princi-
argument about conceptual traveling. It would ap-
ples of categorization can lead to the premature
pear that one of their goals is to introduce a broad
abandonment of potentially useful categories. This
range of empirical cases into the debate on democra-
problem can be avoided by self-consciously thinking
tization, yet without stretching the concept. Along
in terms of ideal types, by using a system-specific
the lines of our discussion, they attempt to do this by
approach to applying categories in particular con-
creating secondary categories (e.g., "corporatist" and texts, or by adopting other techniques that do not
"populist" democracy) referring to cases that observ- depend on the assumption that members of a cate-
ers might hesitate to call "true" democracies. These gory share a full set of defining attributes.
secondary categories serve to increase the extension The effort to avoid conceptual stretching must
of the overall category, without distorting it. In this likewise take a somewhat distinct form when one is
way, the authors bring these cases into the frame- dealing with radial categories. This is the case be-
work of a general discussion of democracy, without cause, with such categories, what we have called the
having to claim that they are all truly democratic. secondary category (e.g., "electoral democracy")
This elaboration of the category by Schmitter and tends to divide up the constituent elements of the
Karl may be placed in perspective through compari- primary category ("democracy"). By contrast, with
son with an innovation proposed some time ago by classical categories, the secondary category (e.g.,
Robert Dahl (1956, 1963, 1971). Dahl argues that for "bureaucratic authoritarianism") tends to contain ad-
the analysis of concrete cases, it is more productive ditional elements beyond those of the primary cate-
to employ the term polyarchy, rather than democracy. gory ("authoritarianism"). As a consequence, with
He uses democracy to refer to "an unattained and radial categories, the secondary category may have
perhaps unattainable ideal," whereas polyarchy re- greater extension, whereas with classical categories,
fers to existing political systems that could be seen as the primary category has greater extension. Relat-
edly, with classical categories one may often avoid
"relatively (but incompletely) democratized" (Dahl
conceptual stretching by removing an adjective,
1963, 73; idem 1973; 8). To avoid conceptual stretch-
whereas with radial categories one may often avoid
ing, Dahl uses distinct labels for the idealized version
conceptual stretching by adding an adjective.
of the category and for the version that refers to
We also argue that because the secondary catego-
actual cases. Schmitter and Karl's treatment differs
ries tend to divide up elements of a radial category
from Dahl's in two ways: (1) in their usage, the term
such as "democracy," the formation of secondary
democracy refers to at least some cases, rather than
categories creates both an opportunity and a prob-
to a hypothetical ideal, and (2) instead of using
lem. It creates an opportunity for broader and more
separate label to extend the category to more cases,
flexible application by increasing the category's ex-
they avoid conceptual stretching by adding adjectives
tension. Yet this very flexibility can lead to major
to the existing label. However, the treatments are
scholarly disputes about whether the category fits the
similar in that Dahl, like Schmitter and Karl, creates a
cases under study.
secondary category (i.e., a noncentral subcategory), A final observation may be made about this central
following a radial pattern. Dahl's term polyarchy issue of the fit between categories and cases. Insights
might be thought of as a "catch-all" secondary cate- into the structure of categories do not tell us every-
gory in relation to the primary category democracy; thing we need to know about how to apply them in
that is to say, using polyarchy to refer to relatively research. Rather, this application depends on sub-
democratized systems is the functional equivalent of stantive expertise regarding the cases under analysis.
adding an adjective to create the secondary category We have suggested the example of a debate on
"partial" democracy or "incomplete" democracy in whether a particular case should be called an instance
order to capture a larger number of partial cases. of patrimonial, as opposed to traditional, authority.
To summarize, the radial category "democracy" Though our methodological understanding of catego-
has a structure that, through the elaboration of sec- ries can frame such a debate, its resolution requires
ondary categories, allows for wide variation in mean- knowledge of the cases. In this sense, the arguments
ing and application within a generally agreed-upon about categories that have been our focus here play
area of discussion. Yet whether these variations in the useful role of bringing us back to our own
meaning and application are accepted or contested detailed understanding of the political settings we
within the scholarly community is an abiding issue. study.

852

This content downloaded from 163.178.101.95 on Wed, 26 Aug 2020 00:07:40 UTC
All use subject to https://about.jstor.org/terms
 American Political Science Review Vol. 87, No. 4

Looking beyond these guidelines, we recognize 1991; Goldstone 1991; Luebbert 1991; Paige 1975; Ragin 1987;
that various issues raised here require further exam- Rokkan 1970; Rueschemeyer, Stephens, and Stephens 1992;
Skocpol 1979, 1984; Tilly 1975, 1984; and Trimberger 1978.
ination. More analysis is needed of the relation be-
3. For authoritarianism and corporatism, see Anderson
tween classical and radial categories. Whereas some 1970; Berger 1981; Collier 1979; Lembruch and Schmitter 1982;
categories unambiguously correspond to one of these Linz 1975; Linz and Stepan 1978; Malloy 1977; O'Donnell
types, others may contain elements of both. Further, 1973; Schmitter 1971, 1974; Schmitter and Lembruch 1979;
in pursuit of particular analytic goals, social scientists Stepan 1973, 1978. For democratization, see Di Palma 1990;
Huntington 1991; O'Donnell, Schmitter, and Whitehead 1986;
deliberately modify categories, often attempting to Pastor 1989; and the new Journal of Democracy.
impose a classical structure on radial categories. 4. The task of developing an adequate typology of different
These attempts to modify categories raise the larger forms of categories and of situating these three types (classi-
issue of the relation between ordinary and technical cal, family resemblance, and radial) within it goes well be-
language. When scholars create a technical language, yond our goal here, which is simply to explore certain
contrasts among these three types in order to illustrate some
they may well succeed in achieving greater clarity
dilemmas that arise as researchers extend their categories to
and consistency or in highlighting what they view as
more cases. For a broad overview of different forms of
important aspects of the phenomena they study. On categorization, see Lakoff 1987. For a brief discussion of the
the other hand, it is possible that this new language relation of ideal types to family resemblance and radial
will not be anchored in the familiar linguistic proto- categories, see n. 8.

types that play such an important role in making 5. Sartori refers to a ladder of "abstraction" (1970, 1040;
1984, 44 46). However, because the term abstract is often
categories interesting and vivid. The modified cate-
understood in contrast to concrete, this label can be confusing.
gories might fail to gain currency, perhaps being We therefore find that it expresses our meaning more clearly
displaced by more familiar usage. to refer to a ladder of generality.
This tension between the advantages and pitfalls of 6. We are aware that a more complete formulation of the
modifying categories raises the question of the proper law of inverse variation entails the expectation that an in-
crease in the intension implies either decreasing, or simply
task of methodology. To what extent should an
nonincreasing, extension (Angeles 1981, 141). In this formu-
understanding of how we tend to use categories lation, the slope would be either negative or zero (i.e.,
inform our judgment about how we ought to use horizontal). However, we follow Sartori's usage, which for
them? Should the methodological analysis of catego- present purposes adequately captures the pattern of varia-
tion.
ries emphasize description, which might encourage
7. For the purpose of the present discussion, we do not
realism about the constraints imposed by ordinary consider the additional characteristic that Wittgenstein at-
language on technical usage, or prescription, which tributes to this type of category, namely, that cases may have
might recommend means to overcome these con- differing degrees of centrality within the category.
straints?'9 We have tried here to give reasons for 8. The commonalities between family resemblance (and
also radial categories) and Weberian ideal types are made very
attending to both.
clear in Burger's insightful interpretation of Weber's theory of
concept formation (1976, 115-16, 156-57). Burger emphasizes
that Weber's use of ideal types grew out of his recognition
that the most interesting concepts are not based on defining
Notes
properties shared by all cases to which the concept refers (and
thus do not follow the pattern of classical categorization).
This article has benefited from many conversations with Weber, therefore, embraced the use of ideal types in which
George Lakoff, as well as from discussions at meetings key attributes associated with the concept were expected to be
sponsored by the Committee on Conceptual and Terminolog- present to varying degrees. Procedures we discuss for making
ical Analysis, of the International Political Science Association generalizations with family resemblances and radial catego-
and the International Sociological Association. We acknowl- ries are thus of considerable relevance to ideal types as well.
edge helpful comments from Benedicte Callan, Ruth Berins 9. Their presentation is couched in the language of mea-
Collier, Stephen Collier, Judith Gillespie, Louis Goodman, surement; and they refer to indicators, rather than attributes.
Andrew Gould, Tomek Grabowski, Karen Kampwirth, Mar- However, they have a broad understanding of measurement
cus Kurtz, Robert Kaufman, James Mahoney, James M. as entailing an "ordered language" that serves "for the
McGuire, Deborah L. Norden, Richard Snyder, Arun Swamy, expression of empirical observations" (Przeworski and Teune
and David Woodruff. Carol A. Medlin and Elizabeth Busbee 1970, 11), which is essentially what is understood here as
provided research assistance. Support for the research came concept formation.
from the Institute of Governmental Studies and the Mac- 10. For example, forming the concept is essential when one
Arthur Interdisciplinary Group for International Security seeks to differentiate, within a set of highly intercorrelated
Studies of the Institute of International Studies, both at the attributes, between those attributes which are components of
University of California, Berkeley. David Collier's work on the concept and those which are causes or consequences of the
the paper was supported by a Guggenheim Fellowship. concept.
1. We treat concepts and categories as similar. Here we use 11. See note 8 concerning the relationship between radial
the term category because it seems to point more directly to the categories and ideal types.
issue of boundaries (a central concern of this analysis) and 12. We do not attempt to take account here of the vast
because it follows the usage of Lakoff (1987), whose work we literature that has analyzed democracy and its components
build upon. Sartori, whom we cite extensively, refers to and dimensions.
"concepts," instead of "categories." Our usage is not in- 13. A fuller analysis of authoritarianism could reveal cer-
tended in a spirit of disagreement with his: it is simply more tain respects in which it departs from the classical model.
helpful for present purposes. But we retain his usage in such However, its usage appears in major respects to have a
special expressions of his as conceptual traveling and conceptual classical form.
stretching. 14. We recognize that this usage is contested (see Gallie
2. The founders of this school include Bendix (1956, 1964), 1956). With changes in gender relations and in the legal
Lipset and Rokkan (1967), and Moore (1966). Subsequent framework within which they are situated, the usage may
work includes Bendix 1978; Bergquist 1986; Collier and Collier change. The point here is that regardless of the outcome of

853

This content downloaded from 163.178.101.95 on Wed, 26 Aug 2020 00:07:40 UTC
All use subject to https://about.jstor.org/terms
Conceptual "Stretching" Revisited December 1993

such disputes, the secondary category will often have greater Dahl, Robert A. 1963. Modern Political Analysis. Englewood
extension than the primary category. Cliffs, NJ: Prentice-Hall.
15. For a discussion of democracy as a "contested con- Dahl, Robert A. 1971. Polyarchy: Participation and Opposition.
cept," see Gallie 1956. New Haven: Yale University Press.
16. In his assessment of post-1964 Brazil, Linz drew atten- Di Palma, Giuseppe. 1990. To Craft Democracies: An Essay on
tion to the poorly institutionalized character of the prevailing Democratic Transitions. Berkeley: University of California
political institutions by referring to them as constituting an Press.
"authoritarian situation" rather than "authoritarian regime" Gallie, W. B. 1956. "Essentially Contested Concepts." Proceed-
(1973, 235). However, their authoritarian character was not at ings of the Aristotelian Society (London) 56:167-98.
issue. Geertz, Clifford. 1973. The Interpretation of Cultures. New York:
17. We find it helpful to view Karl's fourth element, civilian Basic Books.
control over the military, as an aspect of accountability. We Geertz, Clifford. 1983. Local Knowledge: Further Essays in Inter-
have added a different fourth element, the protection of
pretive Anthropology. New York: Basic Books.
rights, to capture related issues discussed by Schmitter and
Goldman, Alvin. 1987. "Cognitive Science and Metaphysics."
Karl (1991, 1992).
Journal of Philosophy 74:537-44.
18. Recent efforts to develop "minimal" or "minimalist"
Goldstone, Jack A. 1991. Revolution and Rebellion in the Early
definitions of democracy which build on the earlier work of
Modern World. Berkeley: University of California Press.
Schumpeter (1950), represent an effort to shift the category of
Hallett, Garth. 1977. A Companion to Wittgenstein's "Philosoph-
democracy toward a classical pattern (Di Palma 1990, 28;
ical Investigations." Ithaca: Cornell University Press.
Huntington 1991, 9). On the other hand, Ostiguy (1993) is
Huntington, Samuel P. 1991. The Third Wave: Democratization
attempting to push scholars toward recognizing its radial
in the Late Twentieth Century. Norman: University of Okla-
structure. With reference to the category of authoritarianism,
homa Press.
we have treated it as classical, thereby following what seems
Karl, Terry Lynn. 1990. "Dilemmas of Democratization in
to be the generally accepted usage. Yet Linz's (1964) pioneer-
ing article bases the discussion on the analysis of a prototype, Latin America." Comparative Politics 23:1-21.
Franco's Spain. To the extent that scholarly understanding of Lakoff, George. 1987. Women, Fire, and Dangerous Things: What
the category is strongly influenced by such a prototype, an Categories Reveal about the Mind. Chicago: University of
element more characteristic of radial categories is thereby Chicago Press.
introduced. Lembruch, Gerhard, and Philippe C. Schmitter, eds. 1982.
19. Among philosophers, research in cognitive science has Patterns of Corporatist Policy-making. Beverly Hills: Sage.
sparked a parallel debate about the proper role of epistemol- Linz, Juan J. 1964. "An Authoritarian Regime: Spain." In
ogy. See Bealer 1987; Churchland 1987; and Goldman 1987. Cleavages, Ideologies and Party Systems, ed. Erik Allardt and
Yrjo Littunen. Helsinki: Academic Bookstore.
Linz, Juan J. 1973. "The Future of an Authoritarian Situation
or the Institutionalization of an Authoritarian Regime: The
Case of Brazil." In Authoritarian Brazil, ed. Alfred Stepan.
References New Haven: Yale University Press.
Linz, Juan J. 1975. "Totalitarian and Authoritarian Regimes."
Anderson, Charles W. 1970. The Political Economy of Modern In Handbook of Political Science, ed. Fred Greenstein and
Spain: Policy-Making in an Authoritarian System. Madison: Nelson Polsby. Reading, MA: Addison-Wesley.
University of Wisconsin Press. Linz, Juan J., and Alfred Stepan, eds. 1978. The Breakdown of
Angeles, Peter A. 1981. Dictionary of Philosophy. New York: Democratic Regimes. Baltimore: Johns Hopkins University
Barnes & Noble. Press.
Bealer, George. 1987. "The Boundary between Philosophy Lipset, Seymour Martin, and Stein Rokkan. 1967. "Cleavages,
and Cognitive Science." Journal of Philosophy 74:553-58. Structures, Party Systems, and Voter Alignments: An In-
Bendix, Reinhard. 1956. Work and Authority in Industry: Ideol- troduction." In Party Systems and Voter Alignments, ed.
ogies of Management in the Course of Industrialization. New Seymour Martin Lipset and Stein Rokkan. New York: Free
York: Wiley. Press.
Bendix, Reinhard. 1964. Nation-building and Citizenship: Studies Luebbert, Gregory M. 1991. Liberalism, Fascism, or Social De-
of Our Changing Social Order. New York: Wiley. mocracy: Social Classes and the Political Origins of Regimes in
Bendix, Reinhard. 1978. Kings or People: Power and the Mandate Interwar Europe. New York: Oxford University Press.
To Rule. Berkeley: University of California Press. MacIntyre, Alasdair. 1971. Against the Self-Images of the Age:
Berger, Suzanne B., ed. 1981. Organizing Interests in Western Essays on Ideology and Philosophy. New York: Schocken.
Europe: Pluralism, Corporatism, and the Transformation of Pol- Malloy, James M., ed. 1977. Authoritarianism and Corporatism
itics. Cambridge: Cambridge University Press. in Latin America. Pittsburgh: University of Pittsburgh Press.
Bergquist, Charles. 1986. Labor in Latin America: Comparative Moore, Barrington, Jr. 1966. Social Origins of Dictatorship and
Essays on Chile, Argentina, Venezuela, and Columbia. Stanford: Democracy: Lord and Peasant in the Making of the Modern
Stanford University Press. World. Boston: Beacon.
Burger, Thomas. 1976. Max Weber's Theory of Concept Forma- Nie, Norman H., G. Bingham Powell, Jr., and Kenneth
tion: History, Laws, and Ideal Types. Durham: Duke Univer- Prewitt. 1969. "Social Structure and Political Participation:
sity Press. Developmental Relationships I." American Political Science
Canfield, John V., ed. 1986. Method and Essence. Vol. 5 of The Review 63:361-78.
Philosophy of Wittgenstein. New York: Garland. O'Donnell, Guillermo A. 1973. Modernization and Bureaucratic-
Churchland, Patricia Smith. 1987. "Epistemology in the Age Authoritarianism: Studies in South American Politics. Politics of
of Neuroscience." Journal of Philosophy 74:544-53. Modernization Series, no. 9. Berkeley: University of Cali-
Collier, David, ed. 1979. The New Authoritarianism in Latin fornia, Institute of International Studies.
America. Princeton: Princeton University Press. O'Donnell, Guillermo A., Philippe C. Schmitter, and Lau-
Collier, David. 1993. "Bureaucratic Authoritarianism." In The rence Whitehead, eds. 1986. Transitions from Authoritarian
Oxford Companion to World Politics, ed. Joel Krieger. New Rule. Baltimore: Johns Hopkins University Press.
York: Oxford University Press. Ostiguy, Pierre. 1993. "Democracy: A 'Radial Concept'?"
Collier, Ruth Berins, and David Collier. 1991. Shaping the University of California; Berkeley. Typescript.
Political Arena: Critical Junctures, the Labor Movement, and Paige, Jeffery M. 1975. Agrarian Revolution: Social Movements
Regime Dynamics in Latin America. Princeton: Princeton Uni- and Export Agriculture in the Underdeveloped World. New
versity Press. York: Free Press.
Dahl, Robert A. 1956. A Preface to Democratic Theory. Chicago: Pastor, Robert A., ed. 1989. Democracy in the Americas: Stopping
University of Chicago Press. the Pendulum. New York: Holmes & Meier.

854

This content downloaded from 163.178.101.95 on Wed, 26 Aug 2020 00:07:40 UTC
All use subject to https://about.jstor.org/terms
American Political Science Review Vol. 87, No. 4

Przeworski, Adam, and Henry Teune. 1970. The Logic of Schmitter, Philippe C., and Gerhard Lembruch, eds. 1979.
Comparative Social Inquiry. New York: Wiley & Sons. Trends toward Corporatist Intermediation. Beverly Hills: Sage.
Rabinow, Paul, and William M. Sullivan, eds. 1987. Interpre- Schumpeter, Joseph A. 1950. Capitalism, Socialism, and Democ-
tive Social Science: A Second Look. Berkeley: University of racy. 3d ed. New York: Harper & Row.
California Press. Skocpol, Theda. 1979. States and Social Revolutions: A Compar-
Ragin, Charles C. 1987. The Comparative Method: Moving beyond ative Analysis of France, Russia, and China. Cambridge: Cam-
Qualitative and Quantitative Strategies. Berkeley: University bridge University Press.
of California Press. Skocpol, Theda, ed. 1984. Vision and Method in Historical
Rokkan, Stein. 1970. Citizens, Elections, Parties: Approaches to Sociology. Cambridge: Cambridge University Press.
the Comparative Study of Processes of Development. New York: Stepan, Alfred, ed. 1973. Authoritarian Brazil: Origins, Policies,
David McKay. and Future. New Haven: Yale University Press.
Rueschemeyer, Dietrich, Evelyne Huber Stephens, and John Stepan, Alfred. 1978. The State and Society: Peru in Comparative
D. Stephens. 1992. Capitalist Development and Democracy. Perspective. Princeton: Princeton University Press.
Chicago: University of Chicago Press. Taylor, Charles. 1971. "Interpretation and the Sciences of
Sartori, Giovanni. 1970. "Concept Misformation in Compar- Man." Review of Metaphysics 25:3-51.
ative Politics." American Political Science Review 64:1033-53. Tilly, Charles, ed. 1975. The Formation of National States in
Sartori, Giovanni. 1984. "Guidelines for Concept Analysis." Western Europe. Princeton: Princeton University Press.
In Social Science Concepts: A Systematic Analysis, ed. Giovanni Tilly, Charles. 1984. Big Structures, Large Processes, Huge Com-
Sartori. Beverly Hills: Sage. parisons. New York: Russell Sage Foundation.
Schmitter, Philippe C. 1971. Interest Conflict and Political Trimberger, Ellen Kay. 1978. Revolution from Above: Military
Change in Brazil. Stanford: Stanford University Press. Bureaucrats and Development in Japan, Turkey, Egypt, and Peru.
Schmitter, Philippe C. 1974. "Still the Century of Corporat- New Brunswick: Transaction Books.
ism?" Review of Politics 36:85-131. Weber, Max. 1978. Economy and Society: An Outline of Interpre-
Schmitter, Philippe C., and Terry Lynn Karl. 1991. "What tive Sociology. Ed. Guenther Roth and Claus Wittich. Trans.
Democracy Is ... and Is Not." Journal of Democracy 2:75-88. Ephraim Fischoff et al. Berkeley: University of California
Schmitter, Philippe C., and Terry Lynn Karl. 1992. "The Press.
Types of Democracy Emerging in Southern and Eastern Winch, Peter. 1958. The Idea of a Social Science. London:
Europe and South and Central America." In Bound to Change: Routledge & Kegan Paul.
Consolidating Democracy in East Central Europe, ed. Peter M. E. Wittgenstein, Ludwig. 1968. Philosophical Investigations. 3d ed.
Volten. New York: Institute for EastWest Studies. London: Basil Blackwell.

David Collier is Professor of Political Science, University of California at Berkeley,

Berkeley, CA 94720.

James E. Mahon, Jr., is Assistant Professor of Political Science, Williams College,

Williamstown, MA 01267.

855

This content downloaded from 163.178.101.95 on Wed, 26 Aug 2020 00:07:40 UTC
All use subject to https://about.jstor.org/terms