1.

WORKING WITH CASE STUDIES IN THE SOCIAL SCIENCES: THE

ISSUES AHEAD

1.1 INTRODUCTION: CASES AND CASE STUDIES

Despite fads and fashions in the academic culture, case-based reasoning has proved to be
a persistent form of analysis in the social sciences, in the humanities, and even in moral
thinking. Broadly understood, case-based reasoning locates the ultimate source of our
epistemic and moral intuitions in the concreteness and idiosyncrasy of particulars. Even
though they can be traced back to a common root, different traditions of reasoning with
cases and of using case studies coexist in the academic landscape. To some extent, the
divide between traditions maps onto the distinction between practice-oriented and theory-
oriented disciplines, the former being concerned with the solution of concrete problems
and the latter mostly aiming at the formulation of abstract conceptualizations and general
principles.
One can discern three main contexts of use: the case study as pedagogical tool, as method
of treatment or intervention, and as an epistemic strategy. Even though, as a matter of
fact, contexts of use to some extent match scientific fields, they do so loosely, in the sense
that a given field can be hospitable to multiple uses of case studies.
In practice-oriented fields, which to a large extent aim at producing and transmitting
practical knowledge that is knowledge of how to practice a profession, the study of cases
became an essential pedagogical tool. Ever since case studies were first introduced in the
Harvard Law School in 1870 by C.C. Langdell, and soon after were adopted by the medical
and business schools in Harvard, 1 the case-based form of reasoning is being used to
familiarize the students with the principles of the discipline and the types of problems they
will confront eventually as professional practitioners (Forrester 1996, Creager et al. 2007).
The case, in fact, instantiates the basic principles, illustrates both recurrent and atypical
problems, and offers practical solutions to those problems. Knowledge of cases is

1
For a discussion of the use of case studies for teaching purpose in the business school see McNair
(1954).
expected to guide the future practitioner through the specificity of the situations she will
eventually confront.
In fields such as psychology and psychiatry the case-study is the clinical study. The
treatment of the patient creates case-based knowledge in the first place, which is in turn
used on subsequent occasions as raw evidence to diagnose, explain, and actually treat other
patients. In these domains, case-based reasoning and knowledge, though used to some
extent as a pedagogical tool, informs primarily the therapeutic practice.2 One might look
at the art of casuistry as a similar context of use. Casuistry was directed at adjudicating
moral issues and based its moral judgments on detailed acquaintance with a huge
repertoire of cases, rather than on the principled knowledge of moral theories. It was the
method of settling moral disputes and dilemmas used by the priest and the savior, regarded
as fully legitimate until Blaise Pascal Provincial Letters contributed to its demise (Jonsen
and Toulmin 1988). Pascal attacked fiercely the Jesuits of Paris for using case-based moral
reasoning to placate wealthy Church donors while punishing poor penitents. It is still
within the casuistic
tradition, as a rigorist objecting to the laxism
rejecting the whole tradition as deceptive or compromising (15). Be that as it may, as a
matter of fact since disrepute for a very long
time.
In the social sciences, case studies are regarded primarily as an epistemic strategy.3 It
is plausible to distinguish between weaker and stronger epistemic usage of case studies. At
the weaker end of the spectrum, case studies occupy an ancillary role in the scientific
investigation. That is, they are subordinate to other research strategies for illustrative
purposes: in this context the case study is the illustration of a concept, a claim, or even a
theory, as instantiated in a concrete case. Alternatively, case studies are the final output of
a long and complex research procedure of inquiry, case study research (CSR), aimed at

2
For a distinction between clinical study and case study see Eckstein (2000 [1975]). For a criticism of the
use of case studies in psychiatry
(2002).
3
In the present thesis, (research) strategy and method are used as synonyms.
acquiring familiarity with, interpreting, characterizing, explaining, and theorizing
phenomena of scientific interest; and, in turn, at using these descriptions, explanations,
theorizations to predict and control social outcomes. When used as epistemic strategies in
this stronger sense, case studies posit hypotheses about phenomena. These hypotheses are
subject to methodological scrutiny for what concerns their credibility and their capacity to
fulfill the research goals.
Despite being widespread, the use of case studies to establish hypotheses in the social
sciences has typically been regarded with suspicion, or with lenience, by some part of the
scientific community (Gerring 2007a, Flyvbjerg 2006). Some scholars simply dismissed
case studies as merely providing anecdotal knowledge and likened them to mere story-telling
(Campbell 1975, Eysenck 1976)4. Other scholars adopted a more charitable view and tried
to rehabilitate what they call the case study method in the arena of scientific methods
(Lijphart 1971, Eckstein 2000 [1975]
experimental and statistical methods. Similar to the latter in purpose and logic, it would
mainly differ from them as to the strength of its findings. Typically, it would be understood
as a research strategy of limited reliability, to which one resorts when the other designs,
regarded as stronger in these respects, cannot be employed.
More recently, scholars started looking at case studies with renewed attention and it is
reasonable to talk of a new trend in the field. Testimony of this turn is the increasing
number of methodologically informed publications in the social sciences and in
philosophy.5 Several circumstances have contributed to this
the growing dissatisfaction with the old view that saw merely the method of case studies
as a last resort, to be employed in the absence of a better alternative; the growing belief
that skepticism about case studies was probably due to a lack of understanding the
method; finally, the realization that case studies have a potential to offer that is still

4
Both Campbell and Eysenck have been initially fierce critics of CSR who have later softened and
modified their original views (Flyvberg 2006).
5
Important works in this areas are Becker and Ragin (1992), Bennett and Elman (2006), Brady and Collier
(2004), George and Bennett (2005), Gerring (2007a), Ragin (2000).
unexplored. This new perspective emphasizes the specificity of case studies and aims at
assessing how they can best fulfill their potential.
This new perspective informs my thesis. Its main focus is the use of case studies in the
social sciences as an epistemic strategy to formulate, establish, and generalize causal
hypotheses. A secondary focus is an investigation into the use of causal findings generated
in case studies to inform policy making in the social realm. The case study, as I understand
it, is the thick analysis of social phenomena as they occurred in circumscribed contexts. It
is thus a situated analysis, rich in detail and directed to represent faithfully the complexity
of the object of interest. As such, it constitutes a reservoir of local knowledge that becomes
evidentially useful when understanding and intervening in the social world in order to fix
it, change it, and improve it. One way in which case studies can perform this evidential
role is by formulating causal hypotheses in a rigorous and conscientious way.
In this chapter, I characterize what can be regarded as two alternative views of case
studies and the understanding of science in which they are embedded.
The first approach flourished in the 70s and looked at case studies as a special, and
typically weaker, form of the experimental, statistical, or comparative methods. Since this
approach tends to evaluate case studies by criteria belonging to other methodological
traditions, it can be said to present a heteronomous paradigm. This view prevented the full
grasp of the specificity and potential of case studies. The second, alternative view, which
developed during the last decades, is taking shape gradually and is still far from being fully
articulated. This approach strives for an understanding of case studies liberated from the
narrow mindset that caricatures case studies as the method of last resort. In particular, it
sees case studies as an autonomous epistemic genre (Morgan 2012). It thus carries the promise
to come to terms with the idiosyncrasy that renders this research strategy so peculiar and
challenging.
The two approaches are rarely found in their ideal typical form and there is no neat
dividing line between authors and historical periods. Recent contributors to the debate are
not always fully consistent in their adherence to the new paradigm and are still often
conditioned by the old one. It is nevertheless useful to keep the two views distinct in order
to state more clearly the reasons why the former approach should be abandoned and the
latter is where we need to move instead.

1.2 THE HETERONOMOUS PARADIGM

The traditional way to understand Case Study Research (CSR)

the experimental one (Morgan 2012). This view emphasizes the affinity in logic and
purposes between CSR and the other research strategies and reduces their differences to
matters of degree. CSR would, in fact, differ from the other methods essentially because
it focuses on a single case rather than many (Lijphart 1971). From the homogeneity of
purpose among the methods, it follows that CSR is characterized as stronger, or weaker,
than the other strategies in some relevant respects, such as its capacity to discover
phenomena, formulate new hypotheses, and suggest new concepts (heuristic power), or
its capacity to provide rational grounds for phenomena, theories, and concepts already in
use (testing power).
Within the traditional view, positions differ as to the comparative strengths and
weakness of CSR. Some scholars, for instance, see it as stronger than the other methods
as to heuristic power while others recommend the use of statistical methods for
formulating new theories and concepts (Eckstein 2000 [1975]). On the other hand,
scholars disagree on whether the testing power constitutes a comparative advantage or
disadvantage of CSR. Eckstein, for example, defends the admittedly unconventional view
that regards CSR as comparatively powerful as a testing tool, while the majority in this
tradition defends the superiority of the alternative methods for theory justification
(Flyvbjerg 2006).
Despite disagreement as to the comparative advantages or disadvantages of CSR, the
traditional view emphasizes the homogeneity of purpose and rationale between CSR and
the other research methods. In particular, adherents to this view typically maintain the
following tenets:6
a) Science centers on the formulation of very broad, or universal,
generalizations;
b) It primarily aims to provide explanations of scientific phenomena;
c) Scientific explanations employ general propositions (see a above)
that relate two or more variables, or regularities.7
d) The scientific practice is structured around:
1. The context of discovery: the process of finding new
scientific hypotheses.
2. The context of justification: the process by means of
which scientific hypotheses are judged.8
According to this view the comparative strength of a method depends on the extent
to which it contributes either to the context of discovery or to the context of justification.
Scientific hypotheses are seen in this context as general propositions that describe
empirical regularities. When assessed in this framework, the heuristic and testing power of
or which we report

6
These four tenets are by no means exhaustive of the traditional view discussed here. The adherents to
this view of CSR also share to a large extent a similar explanatory model, hold to a sharp distinction
between facts and theory, and so on. My purpose here is not to fully articulate the underlying conception
of science; rather, it is presenting the underlying tenets that are more directly relevant to their assessment
of CSR.
7
I shall discuss below that regularities are but a form of scientific generalizations. Mechanisms are
generalizations of a different form, at least according to some scholars (see section 1.3).
8
Paul Hoyningen-Huene notices how the distinction between context of discovery and justification (the
DJ distinction), as used in the 1960s and 1970s, is not just one distinction but a set of intermingled
distinctions (2006: 119). He thus distinguishes five versions of the DJ distinction (six including the one
he proposes): two of them concern the object-level that is discovery and justification as things in the
world; the other four distinguish between discovery and justification at the meta-level by appealing to the
methodological and disciplinary differences in the analysis of discovery and justification. In my
understanding, adherents to the traditional view of CSR most probably conflate the five versions and certainly
endorse the first version which sees discovery and justification as distinct temporal processes. This version is
regarded as untenable by Hoyningen-Huene because it is often not possible in the history of science to
identify processes of discovery as opposed to processes of justification (2006: 121; see also section 1.3
below).
only a single measure on an 2000 [1975]: 124), contributes
either to the formulation of new regularities or to their corroboration.
This view as applied to CSR is epitomized in some influential contributions such as
Arend Lijphart (1971), Harry Eckstein (2000 [1975]), Joseph Campbell (1975), and Stanley
Lieberson (1991). However, it is not exhausted there. Even though certainly dominant
among social scientists back in the 1970s, traces of it still survive nowadays in some major
contributions to the literature on CSR (Vennesson 2008).

1.2.1 DEGREES OF FREEDOM AND GENERALIZATIONS

In Comparative Politics and the Comparative Method (1971) Lijphart articulates an approach to
CSR that can be regarded as paradigmatic of the traditional view:9 he characterizes science
as an explanatory endeavor to which the various research strategies contribute by
establishing empirical regularities among variables of interest. Establishing regularities in
a reliable way demands, however, controlling the other relevant variables, which the
various research strategies attain to a different extent (1971: 683). Lijphart thus articulates
a hierarchy of methods with (1) the ideal controlled experiment on top, and, in descending
order, (2) the randomized experiment, (3) the statistical method, (4) the comparative
method, and (5) CSR respectively. The methods in the hierarchy share the same logic but
pursue control by different means and to a different extent. These means, in fact, are not
all equally effective to the end of control. In particular, the lower a given strategy is in the
hierarchy, the more imperfect is the control it achieves on the relevant variables.
The ideal controlled experiment creates two identical situations in which perfect
control of the relevant variables is achieved and the relationship between the variables of
interest is therefore established with certainty. In most actual scientific practice this ideal
situation can only be approximated and the confidence in the conclusion weakened
accordingly.

his hierarchical view on methods that sees the ideal experiment as the gold standard (see below).
 (2) When the ideal controlled experiment is not available, randomized experiments can
(sometimes) be used that aim to create two equivalent groups in which causal factors are
identically distributed. To this end subjects are assigned to either group by way of a
randomizing procedure. One of the groups is exposed to a treatment, the experimental
group, and the other is given a control, the control group. A positive experimental result
is evidence for the causal claim that the difference in the outcome observed in the two
groups is due to the stimulus provided. Insofar as the randomization is performed
successfully, the assumption that the causal factors are identically distributed in the two
groups is satisfied and the ensuing conclusion is safely established.10
(3) At the next level below, the statistical method relies on the same logic of control
but uses observational data that cannot be manipulated experimentally. The manipulation
of the data is in this case conceptual and performed by way of partial correlation. The data
sample is divided into different sub-groups within which the relevant variables are held
fixed. Each sub-group is examined to see whether the regularity of interest holds when
the other relevant variables are controlled for. Unlike the experimental design where all
relevant variables are taken care of by way of a successful randomization,11 the statistical
method attains control only for those variables that are known to be relevant. For this
reason Lijphart regards it as an approximation to the controlled experiment.

10
This formulation is somewhat imprecise in two respects. First, other assumptions besides the causal

bility and
causation and the inference of probabilities from frequencies (see also Chapter 4). Second, more than
just successful randomization is required for causes to be identically distributed. In fact, randomization
when properly implemented avoids systematic differences between the treatment and control group that

between the two groups in the long run or for high numbers that are not frequently available in the social
expectations
means of relevant variables can differ between groups in the short run due to sampling error (Shadish,
Cook and Campbell 2002). Furthermore, differences that are neither due to sampling error or selection
bias might arise between the treatment and control group. They might be caused, for example, by failure
of blinding, failure of compliance, and so on.
11
See fn. 9 above.
 (4) The comparative method is, regarded as an approximation of the ideal experiment,

number of cases is too small to permit systematic control by means of partial correlation
(1971:
one, in fact, has to be able to observe whether the regularity in question persists when the
other relevant variables are held fixed. For this, one needs to examine a large number of
cases. One should thus resort to the comparative method only when the number of
available cases is too small for finding reliably partial correlations. The degree of control
one achieves by means of this method is therefore very limited.
There are ways, however, to alleviate the problem that the comparative method
encounters of too many variables to control for and too few cases to examine. Lijphart
proposes a few strategies to this end. They are either directed at increasing the number of
cases subject to analysis, for instance by using variables that are more widely applicable
and thus comparable across a larger number of cases in such a way that the sample is
enlarged accordingly. Alternatively, they reduce the number of variables that need to be
controlled for, for example by a general commitment to theoretical parsimony and by
judiciously restricting the analysis to the really key variables while omitting those of only
marginal importance (686-690).
(5) The lack of control becomes a much more troubling concern in CSR where the
number of cases reduces to one. If one examines a single case, in fact, one only obtains a
single observation for each variable of interest.12 Due to lack of variation in the putative
cause and lack of control on the other relevant variables, it seems thus impossible to
establish a reliable conclusion by means of case studies. Multiple explanations are in fact
compatible with the available evidence and the observations one has are insufficient to
rule them out. The lack of control in CSR has been famously reframed as the problem of
degrees of freedom (Campbell 1975, George and Bennett 2005).

12
This i
1.2).
 This notion, widely used in statistics,13 was then carried over to the case-study context.
In general, the number of degrees of freedom is always equal to the number of
observations minus the number of necessary relations among those observations (Walker
1940). If applied to a statistical sample, it is the difference between the sample size (the
number of independent observations) and the parameters estimated on those
observations, such as sample means and variance. In statistics it is a crucial concept
because the sampling distribution depends from it, and if the number of degrees of
freedom are miscalculated the level of significance of the test is also misinterpreted (1940:
260). In general, a high number of degrees of freedom is needed in order to have a stronger
design. Intuitively, the higher the number of observations, the better is the evidence
against which to test alternative hypotheses. It follows that the chance that one hypothesis
fits the whole data is lower and that alternative hypotheses are ruled out is higher. Hence,
if a hypothesis fits the data, the confidence in it is also higher.
Transferred to CSR, what one would observe in this case is a total lack of degrees of
freedom; actually, negative degrees of freedom (George and Bennett 2005: 28). If one treats
the case study as the study of a single case in the statistical sense, in fact, this would be
understood as a situation in which a single independent observation is obtained (N=1)
and too many necessary relations are imposed on that one observation (the relevant
variables). One thus faces here a situation of under-determination of theory by the data
where multiple explanations that fit the same evidence equally well are available, and one
does not have the means to rule all but one out. We would therefore have a very weak
design where the chance of each alternative hypothesis of fitting the data is very high, and
the confidence in it is correspondingly lower.
It is important to stress that this conclusion depends on treating the case study as the
study of one case where a single observation is collected on each relevant variable. This
view was already challenged by Campbell (1975: 179) and since then by many others
(among which Gerring 2004, 2007a, George and Bennett 2005). Even though the type of

13
The notion was introduced by physician James Clerk Maxwell in the 19 th century.
answers provided by each author slightly differs, the general response has been that the
number of observations (and, thus, the evidence against which the hypotheses are tested)
dramatically increases if one considers that observations are collected not just for the
single variables of interest, but also for the various, observable, implications generated by
the theory. In some sense, says Campbell, the researcher has tested the theory with degrees
of freedom coming from the multiple implications of any one theory; and she does not
retain the one theory unless most of these are also confirmed (1975: 182).
There is a distinct but related problem arising from the limited number of cases. As
mentioned above, the view of CSR that is described here depends on a very specific view
of science broadly understood. Science is regarded by these scholars, first and foremost,
as an activity orientated to the production of general knowledge (Eckstein 2000 [1975])
that takes the form of empirical regularities. When the empirical material is constituted by
a single case, the evidential basis seems too narrow to extract general knowledge that is
valid in circumstances different than the ones actually examined. From this prima facie
plausible observation, these scholars concluded that, if the central goal of the scientific
practice consists in establishing generalizations, then it is hard to understand what place
CSR can occupy in such an endeavor. If learning about the general is rendered difficult in
the analysis of a single case, the worthiness of the whole enterprise becomes disputable.
In this respect Lijphart comments (691):

The scientific status of the case method is somewhat ambiguous because science is a
generalizing activity. A single case can constitute neither the basis for a valid
generalization nor the ground for disproving an established generalization.

This conclusion, however, highlights a tension in the view described by Lijphart. Even
if one agrees with Lijphart that generalizing is problematic in case studies and that science
primarily aims at broad generalizations, he would still have to explain how experiments
solve this problem successfully so as to deserve to be ranked first in his hierarchy of
methods while case studies are confined to the bottom of the ranking. In fact, experiments
and case studies seem to face a very similar challenge. Well renowned experimentalists say
in this respect:

Most experiments are highly localized and particularistic. They are almost always
conducted in a restricted range of settings, often just one, with a particular version of
one type of treatment rather than, say, a sample of all possible versions [...] A conflict
seems to exist between the localized nature of the causal knowledge that individual
experiments provide and the more generalized causal goals that research aspires to
attain (Shadish, Cook and Campbell 2002: 18-19).

If one pays heed to what these authors say, one is led to think that also experiments
establish very narrow causal conclusions. The tension in Lijphart might be then
explained by the (implicit) assumption that results obtained via experimental methods
automatically generalize. This is arguably a strong assumption for which no justification is
provided. Lijphart seems to entertain the idea that the high level of control achieved via
experiments entails generalizability. This would be, however, an invalid argument. Control
and generalizability are two distinct issues and, whatever their relation, certainly the former
does not entail the latter. I will discuss these issues below under the label of internal and
external validity.
Given the lack of control that affects CSR and its limited capacity for producing
general knowledge, Lijphart places the CSR at the very bottom of its hierarchy of methods.
If the CSR deserves a place among the scientific methods at all, however, it remains to be
specified what its positive contribution to scientific progress is. Lijphart thinks that,
despite the serious problems that affect them, case studies can still contribute to theory
development even though in a less direct way (1971: 691).

1.2.2 CASE STUDIES: WHAT ARE THEY GOOD FOR, THEN?

Aware of the challenges that CSR confronts due to its lack of control and generalizing
power, Lijphart outlines a taxonomy of case studies where the possible alternative uses of
the method are articulated. The way in which case studies can be useful depends on the
distinctive relation that each type has to theoretical generalizations. Lijphart distinguishes
the following types (1971: 691):
1. A-theoretical case studies;
2. Interpretive case studies;
3. Hypothesis-generating case studies;
4. Hypothesis-confirming case studies;
5. Hypothesis-infirming case studies;
6. Deviant case studies.
Similar taxonomies are outlined by scholars such as Eckstein (2000 [1975]), Flyvbjerg
(2006), Vennesson (2008), where the classification, and the principle behind it, remains
essentially the same even though some change is introduced in the characterization of each
type and the terminology in use. In what follows I will provide a brief description of the
xonomy, and refine it when useful with details provided by the other
contributors. The general purpose of this description is not to establish which taxonomy
is the most accurate or useful, but rather to illustrate the logic they share.
1. Atheoretical case studies
Atheoretical case studies move in a theoretical vacuum (1971: 691). They are neither
guided by theory nor aim at the formulation of theoretical generalizations: they are
characterized by Lijphart as purely descriptive endeavors. 14 As such their usefulness
resides in providing empirical material for further enquiry; in other words, they are data-
gathering tools that contribute to theory formation at best indirectly. Eckstein calls this
type of case- - 2000 [1975]: 132). They aim to present
exhaustive depictions of the overall configuration of the individual so as to capture its
uniqueness (the configurative element). Furthermore, they bring about the significance of
the facts collected by largely intuitive interpretation claiming validity on the ground that

14
Lijphart is categorizing ideal-types. He clarifies that in the practice case studies seldom fit a specific

because almost any analysis of a single case is guided at least by some vague theoretical notions and some
anecdotal knowledge of some other cases, and usually results in some vague hypotheses or conclusions
intensive study and empathetic feel for cases provide authoritative insights into them (the
idiographic element).
Even though Eckstein agrees with Lijphart that these studies essentially amount to the
collection of facts without theoretical import, he also emphasizes that they are often based
on philosophical assumptions less innocent than Lijphart seems to suggest. In the end,
e
sees them as irremediably affected by theoretical poverty. Even though Eckstein does not
deny that these studies can be persuasive and subtle in the intuitions they offer, their
incapacity (actually, the explicit refusal) to collect facts in a systematic way renders them
useless for theory building.
2. Interpretive case studies
Similar to atheoretical case studies, interpretive case studies are selected because of an
interest in the case rather than in the formulation of a general theory (Lijphart 1971: 692).
They have a stronger link to theory than atheoretical case studies but the link is still weak.
In fact, they do not contribute to theory development in any meaningful way. Rather, they
rely on previously established generalizations that are found applicable to the context of
interest and in this way help shed light on the case at hand.
- 2000 [1975]: 134). The
case is explained by subsuming it under well-established propositions: the outcome
observed is in fact inferred by deduction from the extant theory and a set of specified
antecedent conditions. According to Eckstein, the explanation of the case (in his words,
the interpretation) is successful if it is logically compelled by the theory: one should be able
to demonstrate that, given the regularity and the characteristics of the case, the outcome
must have occurred or had a high probability of doing so (136).
3. Hypothesis-generating case studies
Lijphart asserts that hypotheses-generating case studies are selected for purpose of
theory-building. They rely to a very limited extent on previously developed but vague
generalizations. The purpose of these studies is to contribute to the formulation of new
generalizations in areas where no theory, or very unsatisfactory theory, exists. The
generalizations thus formulated ought then to be tested subsequently by further empirical
inquiry.
, hypothesis- 2000
[1975]: 137). He remarks that there is a good track of records of case studies used for
heuristic purposes which typically consist in discerning important general problems and
possible theoretical solutions. One reason why this might be the case is that case studies
consist in the intensive analysis of a subject, do not restrict the inquiry to a limited set of
variables, and in so doing increase the probability that new variables and critical
relationships among them be discovered (138).
4. Theory-confirming case studies
Theory-confirming case studies are developed within the framework provided by well-
developed theory. The general purpose of this type of studies is testing theoretical
generalizations by providing evidence in support of the extant theory. Lijphart maintains
that the theoretical value of these studies is limited if the theory has been confirmed
already by a large number of cases. In other words, if the theory obtained already
substantial evidence in its support, the confidence in the hypothesis is only marginally
increased by one additional positive instance. The theoretical value of these studies is
however enhanced when the studied cases turn out to be crucial (1971: 692).

constitute a
definition and does not explain why crucial cases are so important for testing. In general,
crucial cases are such that one would expect from them either a strong refutation or a
-
-likely case, in fact, is such
that, given the characteristics of the case, one would expect a strong rejection of the
theory. If the study of a least-likely case eventually confirms the theory, the confidence in
the theory is remarkably increased, or at least it increases more than it would if the case
was not crucial.
5. Theory-infirming case studies
 Similar considerations as above extend to theory-infirming case studies, which are also
developed within the framework of a well-developed theory. They test theoretical
generalizations and provide negative evidence for it that is, evidence that refutes the
theory. Also in this case, if the theoretical proposition is solidly based on a large number
of cases, the theoretical value of one additional infirming study is limited as it decreases
the confidence in the theory only marginally.
Furthermore, the infirming power of the case study is higher if it uses a crucial case. A
crucial case is useful for a theory-infirming study if one would expect from it a strong
confirmation of the theory. This type of crucial cases is called by Eckstein and Flyvbjerg
-l -likely cases are such that, given the characteristics of the case, one
would expect the theory to be strongly supported in the circumstances. If the study of a
- refutes the theory, the corresponding decline of confidence in
the theory is particularly severe or, at least, more severe than the decline one would have
if the case was not crucial.
6. Deviant case studies
Finally, deviant case studies select cases that are known to deviate from well-
established generalizations; they are known as outliers (Flyvbjerg 2006). They are used to
develop theory further either including new variables or refining the variables in use by
disclosing why the case at hand is deviant. They thus weaken the original proposition, but
replace it with a modified version of it, to be tested against new cases.
Even though the scholars mentioned above share the same organizing principle for
their taxonomy, they see the comparative strength of the CSR in different types of case
studies. Lijphart seems to emphasize their heuristic power while acknowledging that they
are weak in terms of testing power when contrasted with the other research strategies in
the hierarchy. In this position, he is not alone. It used to be conventional wisdom among
social scientists that case studies are most valuable in the context of discovery rather than
justification, and this view is still endorsed in more recent contributions (Gerring 2004,
2007a).
 Eckstein holds instead a rather unconventional view. He defends the idea that the
comparative strength of the CSR resides in the testing power of crucial cases. In his view
testing a theory is an effort to falsify, and such an effort is most profitable when most-likely
and least-likely cases are identified (Eckstein 2000 [1975]: 146).15 Despite the differences
in emphasis, these scholars can be understood as adopting a similar view on CSR, and on
science more in general: science is seen as a generalizing endeavor aimed at explaining
phenomena of interest by formulating empirical regularities. In this perspective, they share
=
of study contribute to science so understood in any meaningful way?

1.3 THE NEW PROJECT: CSR AS AN AUTONOMOUS EPISTEMIC

GENRE
An alternative view of CSR is, however, on offer. It started taking shape recently in the
methodological debate on qualitative research and its relationship with the quantitative
tradition that followed the publication of Designing Social Inquiry by King, Keohane, and
Verba in 1994 (Brady and Collier 2004, George and Bennett 2005). It is also traceable in
isolated contributions that either preceded or lie at the margin of that debate (Creager et
al. 2007, Flyvbjerg 2006, Forrester 1996, Geertz 1973, Morgan 2012, Vennesson 2008).
These authors do not belong to the same philosophical tradition. Nevertheless, they
contributed maybe unintentionally, and to different extent, to what can be seen as the
same project, that is, liberating CSR from the statistical, comparative, and experimental
mindset. They tend to emphasize the distinctiveness of CSR and the difference in logic
and purpose, rather than the similarity, with respect to the other research strategies. Thus,
at its core this view can be regarded as different from the approach sketched above in the
recognition of CSR as an autonomous style of reasoning which, as such, needs to be evaluated
on its own terms.

15
See Flyvbjerg (2006) for a contemporary criticism of the conventional wisdom on the lack of testing
power of case studies.
which historically became a permanent component of science (1992). Hacking adopts the
definition by historian of science Alistair Crombie who explains ostensibly the notion by
pointing to six styles. In the formulation by Hacking they are:

a) The simple method of postulation exemplified by the Greek mathematical

sciences.
b) The deployment of experiment both to control postulation and to explore by
observation and measurement.
c) Hypothetical construction of analogical models.
d) Ordering of variety by comparison and taxonomy.
e) Statistical analysis of regularities of populations, and the calculus of probabilities.
f) The historical derivation of genetic development (1992: 4).

Hacking is not interested in the specific way in which every style came into being.
Every style, in fact, became independent of its own history to be what we now regard as

10). Rather, he is concerned with what

makes of a certain method a style of reasoning. The necessary condition for being a style
of reasoning is the introduction of novelties in the form of objects, evidence, sentences,
modalities, and possibilities.
novelties of most or all of the
listed types, and should do so in an open-textured, ongoing and creative way. He gives the

numbers and shapes, and then just stop. The typ -ended once we
12). In addition to the ostensive explanation and
necessary condition, Hacking suggests that what constitutes a style of reasoning is a set of
techniques of self-stabilization: it is by way of these techniques that each style persists in its
peculiar and individual way (1992: 16).
a seventh style on top of the six listed by Hacking (Forrester 1996: 2). Forrester motivates
his proposal with the purpose of characterizing the dominant style of reasoning in
psychoanalysis and the related sciences since the early Twentieth century. His project is,
however, more ambitious than that. Reasoning in cases is not only typical of fields such
as psychiatry, psychology, and criminology; rather, there is a methodological continuum
from the psychoanalytic case, to the case used as a pedagogical tools in certain academic
traditions, to the style of reasoning that is typical of the man of practice.
Thus, Forrester launches a unifying philosophical project that he leaves, however,
unfulfilled. To some extent his line of thought is followed up more recently by Mary
of styles of reasoning, which she re-
labels epistemic genres. She writes in defense of case studies as an epistemic genre, and

each epistemic genre developed its own generic way of finding and validating knowledge,
so that work within that genre came to be judged within that epistemic genre and by its

(2012: 671). It then follows that if CSR constitutes indeed an epistemic genre similar
considerations should be extended to it as well.
This view of CSR sits comfortably in a conception of science rather different from the
traditional view presented above. This alternative approach is not as coherent and uniform
as the one to which it reacts is; it hosts in fact a range of positions which, however various,
challenge the old paradigm in similar respects. Generally speaking, the new approach sees
science as a much less homogeneous practice than the old view would have it. In particular,
it challenges to varying extents the four tenets presented above (see sec. 1.2). The first
tenet to be disputed is:
a. The (indiscriminate) characterization of science as a generalizing
epistemic practice.
Objections to the first tenet come in different forms and are not all equally radical.
Some scholars understand this statement as referring to universal generalizations and point
 more limited scope.
This is the milder objection to (a). Other scholars understand the first tenet as having
normative import to the effect that general theoretical knowledge is the only (scientific)
knowledge there is; or, less radically, is more valuable than concrete context-dependent
knowledge. These scholars reject this normative stance, argue that concrete context-
dependent knowledge is as valuable, if not more, than theoretical knowledge, and that its
value is proved by the crucial role this type of knowledge plays in human learning
(Flyvbjerg 2006). Still other authors entertain a radically different view of science, or at
least some fields in it. They see some part of science as a case-based epistemic activity
where specific instances of scientific practice in the form of case studies, models, exemplars,
occupy a central role in learning and research strategies (Creager et al. 2007). Related to
this is the challenge to the second tenet:
b. The primary aim of science is the explanation of scientific
phenomena.
The new view tends to de-emphasize the prominence of explanation as the unique,
and/or most important, goal in science. On the one hand, it rehabilitates description as
an important goal on its own decoupled from explanation and not just preliminary to it,

faithful to the complexity of the studied subject. The ultimate goal is not to explain the
specific by subsuming it under a general law; it is rather providing understanding of the
specific case by revealing its complexity. Description so understood is pursued for its own

explanation. Furthermore other purposes of science are more explicitly contemplated.

Some authors see in policy making one of the main goals of scholarly research. The idea
behind this view is that scientific inquiry should lead to the production of knowledge that
is useful ultimately for the policy maker.
c. The only generalizations relevant to science are empirical regularities.
 Several authors object to tenet (c) by pointing out that general knowledge of scientific
interest does not only come in the form of empirical regularities but of causal processes
and mechanisms (Coleman 1986, Stinchcombe 1991). The debate on the concept and use
of causal mechanisms in the social sciences is too vast to do justice to it in a few lines.
This debate has grown massively in the last years, with contributions from many social
scientists and philosophers of science.16 The point of interest here is that these scholars
defend the importance of mechanisms vis-à-vis regularities in the philosophical and
methodological discourse about science. They defend this claim on the following
grounds:17 mechanisms are distinct from regularities; the epistemic strategies conducive to
the discovery of mechanisms and regularities are different; mechanisms play an equally, if
not more, important role than regularities in the explanation of social phenomena.
d. Discovery and justification are distinct temporal processes.
Even though, as noticed above, strong objections have been raised against the version
of the DJ distinction that sees justification and discovery as distinct temporal processes
(see footnote 5), traces of it still survive nowadays in several contributions. The general
idea behind the rejection is that in the practice one cannot separate the moment and
process by means of which hypotheses are constructed from the moment and process by
means of which they are tested. If this is indeed the case, CSR cannot be said to contribute
either to the context of discovery or to the context of justification (Vennesson 2008).

1.3.1 CASE, CASE STUDIES, AND CASE STUDY RESEARCH

The new perspective leads to re-conceptualizing the notions of case, case study, and case
study research. Definitions that were typical of the traditional view are dropped. In key
contributions to the old view, case, case study, and case study research are defined as
follows:

16
Prominent contributors to the debate on the role of mechanisms in the social sciences are Bunge 1997,
2004, Elster 1989, 1998, Hedström and Swedberg 1998, Hedström and Ylikoski 2010 Little 1991, 1998,
Mayntz 2004.
17
This is a gross generalization. The philosophical positions in this debate are almost as numerous as the
contributors. My purpose here is just conveying the sense of what are the issues at stake.
 CASE: A phenomenon for which we report and interpret only a single
measure on any pertinent variable (Eckstein 2000 [1975]: 123-4).
CASE STUDY: the study of a single case.
CASE STUDY METHOD:18 is the method applied to the study of a single
case, rather than many as in the statistical method, or just a few as in the
comparative method (Lijphart 1971: 691).
These definitions seem not only strongly influenced by a statistical viewpoint but also
hardly informative, and inaccurate. Certainly, there is much more to a case study than just
reporting and interpreting, whatever that means in this context, a single measurement on any
pertinent variables; furthermore, it is also disputable that the case study only reports a single
measurement on these variables, rather than many (George and Bennett 2005). Even
though the old definitions are rejected and replaced by new ones, there is not agreement
yet on a definition of case, case study, and case study method within the individual
disciplines, let alone across them. The following definition by political scientist Pascal
Vennesson are probably quite representative of the average feeling. Not all features they
include, however, would be regarded as defining case study by the scholars in this new
trend.19 Disagreement on basic concepts shows that the new trend has not developed yet
into a fully-fledged view.
CASE*: A phenomenon, or an event, chosen, conceptualized and
analyzed empirically as a manifestation of a broader class of events.
CASE STUDY*: is a research strategy based on the in-depth investigation
of one, or a small number, of phenomena in order to: (i) explore the
configuration of each case, and (ii) elucidate features of a larger class of
(similar) phenomena, (iii) by developing and evaluating theoretical
explanations.

18

19
I will discuss these features below.
 One can find similar definitions in George and Bennett (2005), Gerring (2004, 2007a)
and Ragin (2000). These definitions deserve some comments, and some refinements.
In the first place, case is a theoretical construct: it is not the specific phenomenon or
event, but the conceptualization of it. Something such as a phenomenon or event in
political science, an individual in psychology and psychiatry, a cultural practice in
anthropology is turned into a case by the study of it. Forrester, citing Foucault talking about
the development of the clinical sciences, says that the examination, surrounded by all its
12).
The above definition is, however, too restrictive when it defines the case as the
manifestation of a broader class of events. The theoretical work behind the selection and
analysis of cases is often much more fluid. The class of reference is often not established
at the beginning of the research, can change along the process, or emerge at its conclusion.
Sometimes it might even remain indeterminate.
Furthermore, the purpose of the study does not need to shed light on a larger class of
elements, but might simply be the exploration of the single case. This does not imply that
generalizations from a case study are not possible or even undesirable. Rather, the other
cases to which the results can be generalized are often identified once the case study has
been accomplished. In fact, results from previous case studies may turn out to be
applicable to new cases in an unexpected or unplanned way. In a similar vein, case studies
do not always have such a strong and explicit theoretical motivation. Thus, the purpose
of the study does not need to be the development and evaluation of a theoretical
explanation. It might just be the characterization, or description, of the case at hand. In this
sense, features (ii) and (iii) above should not be regarded as a defining feature of case
studies but as additional epistemic benefits of this research strategy.20
Finally, case study and case study research (or case study method) should be treated as
distinct notions. The definition of case study given above refers in fact to the research

20
In particular, that feature (ii) is indeed an epistemic benefit of CSR will be argued in the next section
and in Chapter 3.
strategy, hence to case study research. The case study is the (written) output of this research
strategy when employed to study one or a few cases.

literature but at the same time emphasizes some of the aspects that to some extent fail to
be present in their definitions. She identifies the following as characteristic features of this
research strategy (Morgan 2012: 668):

I. Boundedness: a case study investigates a bounded whole object of analysis;

II. Open-endedness: the boundary between subject of analysis and context is not
clear at the start of research and may remain fluid during the study;
III. Depth: the case study creates a considerable depth of engagement with the
subject and dense evidential materials across a range of aspects of the topic;
IV. Multi-method: many potential research methods may be used within the case
study.
V. Complexity: the outcome is a complex, often narrated, account that typically
contains some of the raw evidence as well as its analysis.

The pressing issues that CSR confronts can be understood in terms of validity and
relevance.21 I define and discuss validity and relevance in the sections that follow. First I
distinguish between internal and external validity. Rather than assessing CSR in terms of its
heuristic or testing power, and try to establish whether CSR contributes to the context of
discovery or justification, what matters here are the conditions under which CSR can
obtain internal validity, and what criteria are adequate to assess whether a given case study
is in fact internally valid. Furthermore, rather than assessing whether and to what extent
case studies contribute to the formulation of empirical regularities, the broader question
of interest here is how one generalizes from a given case study. As I shall argue below, there
are different strategies to generalize in CSR, one such a strategy is by producing results
that are externally valid. Finally, once one acknowledges that the social sciences can aim at

21
Whereas the notion of validity has been used quite extensively in the literate on CSR (see, for instance,
Gerring 2007a, Morgan 2012), to the best of my knowledge the notion of relevance was never explicitly
adopted and discussed.
a variety of purposes all equally worthy, the pressing issue is under what conditions a case
study provides results that are relevant vis-à-vis the purposes at stake.
If CSR is to be treated as an epistemic genre, as the experimental and statistical
methods are, then it has its own way of finding, validating, and, I would add, generalizing
knowledge. The challenge that lies ahead for those scholars that see in it an autonomous
style of reasoning is to understand which way this is.

1.3.2 INTERNAL VALIDITY AND EXTERNAL VALIDITY

The concept of validity refers to the correctness of scientific claims. The methodological
literature typically draws a distinction between internal and external validity. Scientific results
are internally valid if, and only if, the inference from the evidence to hypothesis within a
given study is correct (Reiss 2008). The internal validity of a study, thus, is compromised
when throughout the inferential process from evidence to hypothesis mistakes are made.
Mistakes might be due to fallacious reasoning, to the violation of the assumptions, or to
an incorrect measurement procedure. Internal validity is a scientific ideal since, as a matter
of fact, it is very hard to control for all sources of error in the inferential procedure, and
have adequate knowledge that the relevant errors have been in fact controlled. Within a
given study, one approaches, rather than attains, internal validity. The extent to which
internal validity can be achieved varies from method to method. Certain methods have
higher chance to deliver conclusions that are internally valid because of the type of
inference they rely on (deductive rather than inductive) or because they have strategies to
ensure that the assumptions are satisfied.
When discussing internal validity in relation to CSR one can tackle the issue from two
points of view. On the one hand, one can ask how specifically CSR produces results that are
internally valid. This strategy amounts to establishing what conclusions can be reached
vis-à-vis the evidence at hand, how, and with what degree of confidence. It thus examines
the techniques by means of which conclusions are inferred and aims at identifying under
which conditions these techniques are correctly employed. These are conditions of internal validity
and pertain to the inferential procedure from evidence to hypotheses. On the other hand,
one can formulate an assessment of the case study based on criteria other than inferential
correctness. Rather than examining the techniques by means of which the evidence is
analyzed and the conclusion derived, one would focus on features of the narrative itself.
This way typically abstains from judgments of correctness; rather, it imposes criteria, or
standards, on the narrative as various as coherence, credibility, persuasiveness, or
significance, and use those as indicative of the validity of the study. The two approaches
coexist in the literature on CSR. Proponents of either approach regard their strategy as
conducive to ascertain the internal validity of CSR. Both, however, are subject to specific
challenges.
In the former case, the difficulty arises from the fact that CSR relies on the use of
several techniques that are often employed jointly. Upon examining the assumptions,
inferential procedure, and type of evidence proper of each technique, one thus needs to
formulate the conditions that are conducive to correct conclusions, and the degree of
confidence one can have in conclusions thus reached. Furthermore, one needs to specify
how the techniques are used correctly when employed jointly, and how the joint use of
these techniques affects the confidence in the final conclusion. The techniques most
commonly used for analyzing evidence in CSR are process-tracing and the method of
comparison.
The method of comparison shares the same logic that underlies the comparative
method discussed in section 1.2.1.22 The method of comparison refers to a technique of data
analysis that is employed in case study research for within-case and across-case analysis
(CSR can in fact focus on the study of one or a few cases). In CSR the results obtained by
the method of comparison are always integrated with other types of evidence to develop
a complex narrative which the method of comparison could not possibly develop on its
own. The comparative method is a research strategy that consists in the comparison of a
(typically higher) number of cases to establish causal relations between relevant variables.
We can thus regard CSR and the comparative method as distinct study designs which can

22
This logic was first articulated by J.S. Mill in his description of the method of difference and method
of agreement (1858).
both be directed to causal analysis. The former differs from the latter in that it draws on
a variety of causal insights and techniques, among which the method of comparison, the
results of which are eventually integrated into the narrative.
Whereas the method of comparison is well mastered and understood, process-tracing
is a technique whose inner workings, despite broad usage, are not fully understood yet.
Even though its conditions of validity are not fully articulated, it is characterized as
directed to the identification of causal processes and the circumstances in which these
processes unfold. The literature typically refers to the former as mechanisms, and to the
latter as the systems in which the mechanisms are embedded. When process-tracing is used
jointly with the method of comparison, the description of the mechanism, and of the
conditions in which it is triggered, complements, enriches, and refines the causal relation
that the method of comparison identifies. In Chapter 2, I will try to clarify what conditions
of validity do apply to process-tracing and how, and to what extent, the method of
comparison and process-tracing when employed jointly increase the plausibility of the
general conclusion.
An alternative, and perhaps more controversial, way to assessing validity in CSR is by
imposing criteria on the final output (Morgan 2012). Typically, the final output is a
narrated account which has a very complex form and contains some of the raw evidence
as well as its analysis and ties together the many different bits of evidence (2012: 668).
Standards can therefore be established on how the different bits of evidence should tie

work on legal cases, Morgan suggests that valid case study accounts exhibit the following
criteria: consistency with all the evidence found, coherence within the account (the bits of
evidence fit together), and credibility of the explanation in social scientific terms (2012:
674). This approach to internal validity does not find full consensus in the scientific
community. While standards are acknowledged as leading to valuable case studies, they
are also criticized for being either too loose or simply inadequate to ensure rigor (Geertz
1973, Morse et al. 2002). Whether they are in fact conducive to internal validity is therefore
still open to discussion.
 Scholars belonging to the old tradition worried about the scientific usefulness of case
studies because of their lack of generalizing power. Since this tradition sees science
primarily as aiming at finding regularities that explain phenomena of interest, it is hard to
appreciate how the study of a single instance can help formulate broad generalizations of
this sort. This is, however, a very specific way to characterize what is a much more
complex issue: generalizability pertains to scientific findings whose scope is more
encompassing than the subject of the study. Addressing generalizability requires one to
specify (i) what should be generalized from a given study, (ii) to what such a result should
be generalized, (iii) and how. Different answers can be given to these questions. I
distinguish three approaches that differ in the answers given to each of the three questions
above. In the first two cases, generalizability can be understood in terms of external validity
and what is actually generalized is an empirical finding; in the third case, the generalization
is theoretical.
Scientific conclusions are externally valid if they are correct in the studied context and
in other contexts yet unstudied. Thus, whereas internal validity is obtained by way of
correct inference from evidence to hypothesis within a given study, external validity is
obtained by way of correct inference from the results obtained in a context studied directly
(the original context) to a hypothesis about a new context yet unstudied (the target
context). As such, external validity is not only achieved if one formulates general
conclusions that are universally, or almost universally, valid. It is obtained any time
scientific conclusions about a target context are inferred correctly from results that were
established in the original context; that is, they are extrapolated correctly.
Under the broader umbrella of external validity, the scholars of the old school
provided a possible answer, very specific indeed to the three questions posed above. Any
study should establish an empirical regularity to be generalized, if not universally, then to
a broader class of cases of which the studied case constitutes a representative instance.
This view on generalizability by no means belongs only to the traditional view. It is still
very present in the contemporary debate on CSR, even though in a slightly revised form
as one can easily tell from some very common definitions of case study actually in use.23
However, that the conclusions be generalizable to a broader class of elements is not
the only modality for CSR to achieve external validity, or any other design for that matter.
Successful extrapolation from case to case is another modality that is, according to some
scholars, more amenable to CSR (Forrester 1996, Creager et al. 2007). This alternative
approach gives different answers to each of the three questions posed above. As to the
object of the generalization, it is not the empirical regularity as such, but the causal
mechanism responsible for the phenomenon of interest and the causal conclusion it elicits.
The target of the generalization is another or a few other cases where similar phenomena
are observed. Finally, the inference is made by analogical reasoning. The studied case is
not regarded as an instance that is representative of a broader class of elements but is,
more generally, a situated phenomenon that is similar to others in some relevant respects.24
In the third approach, the case study is seen as the occasion for formulating conceptual
and theoretical generalizations rather than empirical results to be directly applied to other
cases, be they classes of elements or concrete instances. Robert Yin describes this
alternative as a form of analytic generalization and contrasts it with the statistical
generalization typical of other research strategies. In CSR the mode of generalization is
analytic generalization, in which a previously developed theory is used as a template with
which to compare the empirical results of the case study 2003: 33). In a similar spirit,
Clifford Geertz talks of clinical inference, or inference within cases. Rather than beginning
with a set of observations and attempting to subsume them under a governing law, such
inference begins with a set of signifiers, or symptoms, and attempt to place them within
an intelligible frame (1973: 165). The task is constructing an analytic system in whose
terms what is generic to the studied case will stand out against the other determinants of
human behavior. The generality the case study contrives to achieve, Geertz suggests,

23
See section 1.3.1 above for my discussion of some recent definitions of case study; see Chapter 3 for a
thorough discussion and criticism of what I call the traditional view on external validity.
24
It can be argued that the former approach discussed above is interpretable to some extent as a special
case of the latter, more general, approach.
 ertz
1973).
In Chapter 3, I shall explore and defend the second approach to external validity. This
line of reasoning about external validity seems in fact the most promising. On the one
hand, it rescues CSR from the statistical viewpoint that informs the first approach and
opens the way to a better understanding of CSR as an autonomous epistemic genre.
Furthermore, unlike the third approach, it focuses on the causal conclusions that can be
elicited from the case studies and carried over to other contexts. What is under scrutiny in
this thesis is the capacity of case studies to generate valid causal knowledge under the
assumption that this is the evidence we primarily need for policy making. In line with the
general spirit of the present work, in the chapters that follow I only focus on the
formulation and extrapolation of causal claims in CSR and omit analytic generalizations
instead.

1.3.3 RELEVANCE
Case studies can be assessed in a further respect besides their internal and external validity,
namely with respect to the relevance of their results. Valid hypotheses are those hypotheses
that are regarded as correct by the extant background knowledge; relevant hypotheses are
those conclusions that are adequate with respect to purposes. Validity claims thus differ
from relevance claims first and foremost because they address distinct properties of
scientific hypotheses: the former express judgments about the correctness of the
inferential procedure by means of which conclusions are derived within the study, or
outside of it; the latter express judgments on the adequacy of those conclusions to further
purposes, fulfill tasks, or solve problems. As they target different properties of scientific
conclusions, judgments of validity and judgments of relevance are based on different sorts
of considerations.
Validity claims are directed to evaluate whether a hypothesis is justified by the
evidence: matter for this the inferential logic of the method employed and the assumptions
on which it works. Relevance claims are directed to evaluate whether the results delivered
by a given method are epistemically adequate to try to solve and achieve the problem and
purpose in question. This type of analysis presupposes:
1) The definition of the purpose/problem at stake;
2) The characterization of the kind of conclusions required;
3) The assessment of what kind of results a given method provides.
Relevance claims are justified if 1), 2), and 3) are also justified. This implies that they
can be contested by impugning either 1), 2), or 3). One can impugn 3) by objecting that
the conclusion licensed by the method in question is not exactly the type of information
that is required to fulfill the purpose of interest as specified in 2). For instance, if control
is the goal of interest and one regards counterfactual claims as necessary to this end, one
might object to the relevance of case study evidence because it corroborates singular causal
hypotheses that cannot support counterfactuals. Or, one can impugn 2) by objecting that
the specified kind of scientific result, even if actually provided by the method in question,
is not adequate for the purpose or problem as defined in 1). For instance, one might
object that counterfactuals are not the type of claims one need for control purposes, after
all, and that one needs evidence of causal processes instead.
Alternatively, one can impugn 1) by arguing that the problems and purposes as actually
defined in 1) are either illegitimate or incorrectly posed. Purposes and problems can be
defined at a rather abstract level, for instance by referring to the fourfold distinction
between description, explanation, prediction, and control, or at a higher level of detail.
When considering control, one might define it as the formulation of policies to bring about
intended effect in the social realm.25 And one might object to this aim as an illegitimate
goal of science on the ground that science should stay aloof from all that concerns politics,
policies and power. Alternatively, one might specify in greater detail the type of problem,
policy, or yet effect of interest, for instance, by suggesting that the aim of economics is
erasing poverty in developing countries and leveling out inequalities. The objection to
purposes defined at this level of specification is eminently an empirical matter, and draws

25
See section 1.3.3.1 for a similarly general, though more precise, characterization of policy making.
to a large extent on factual considerations about the context in which the purpose at hand
is to be reached or the problem solved.
Even though validity and relevance are distinct concerns, relevance claims do relate in
fact to validity claims. In particular it seems reasonable to argue that when assessing the
kind of results that a given method provides (point 3 above) one should only pay heed to
those results that are reached by way of a valid inference. In this sense, relevance presupposes
validity. Validity is, however, far from sufficient for relevance. In particular, relevance
claims draw on a much broader set of considerations than the validity of those scientific
conclusions whose relevance is ultimately at stake. This set of considerations encompasses
among others formulating the aims which can legitimately be pursued by way of a scientific
inquiry, and assessing which methods actually help further those aims.

1.3.3.1 CASE STUDY RELEVANCE FOR POLICY MAKING

Besides the traditional aims of scientific research, such as description and explanation,
some contributors to the recent debate on CSR have started considering more seriously
control. Case studies thus seem to be methodologically interesting not only for their
descriptive and explanatory virtues, but also as evidence for policy makers. As a matter of
fact, case studies are used for intervention purposes in several fields. In medicine,
psychology or psychiatry the prominent goal of the clinical study is the treatment of the
patient both directly and indirectly by building clinical knowledge to be used in future
treatments; similarly, social and economic policies implemented by regional and
international institutions are also largely built on case-based knowledge.
There are specific features of case studies that seem to render them relevant to policy
making on an intuitive ground. The case study can be in fact characterized as providing
contextual, concrete, and processual evidence.
In the first place, it is an investigation into phenomena that are circumscribed in time
and space, where the boundary between context and subject of investigation remains fluid
during the research (Morgan 2012). 26
The output of this procedure is thus a
characterization of the phenomenon that is left permeable for contextual information.
Furthermore, case studies are thick descriptions that use dense evidential material. This
type of analysis is extremely rich in detail and, in this sense, formulated at a relatively high
level of concreteness. By focusing in fact on localized phenomena, the case study can
afford to retain information that ought instead to be sacrificed in the process of
abstraction that more general analyses demand. Finally, in certain circumstances case
studies explain by describing the causal process, or mechanism, by means of which certain
effects are brought about.27 In particular, this form of explanation amounts to identifying
the conditions, the factors, the events, and their configuration that were responsible for
the phenomenon of interest.
Thus seen, case studies are relevant to policy making understood as a practice aimed
at producing intended effect in the social world by manipulating some feature of it.
First, policies operate in contexts: they produce the intended effect in conjunction with
other causal factors. Using a more technical language, policies are INUS conditions, that
is, they are an insufficient but non-redundant part of a condition which is itself
unnecessary but sufficient for the occurrence of the outcome of interest.28 Consider an
example used by Nancy Cartwright and Jeremy Hardie to illustrate this point (Cartwright
and Hardie 2012). California implemented a project to improve the academic achievement
of its school pupils by reducing class-size in the mid90s. Class-size reduction is an INUS
On its own it does not contribute to the effect of interest,
unless supported by other factors. For instance,
availability of space and quality of teachers are other non-redundant but insufficient parts
,
thus, policy making is to a large extent a situated practice: when involved in actual decision
making policy makers target contexts in their specificity. From this follows the relevance

26
See point II in section 1.3.1.
27
They can do so by employing process-tracing as a technique of causal analysis (see section 1.3.2).
28
.
of evidence such as case studies that is contextual and thus provides clues to what other
factors besides the policy of interest need to be present for the outcome to occur.
Furthermore, policy makers try to sort the intended effect by modifying certain
features of the target contexts, by erasing existing practices, or by creating new ones. In
other words they manipulate the context. Manipulation requires detailed information about
what is exactly to be modified and how it can be modified: policy making needs thus to
be informed by descriptions fine-grained enough to enable the decision maker to have an
answer to this type of questions; this is why evidence that is concrete plays an important role
in it. Finally, the idea of manipulating certain aspects of reality in order to produce the intended
effect presupposes causal justification. Policy making so understood is rational only if
supported by causal beliefs that is, by the confidence that there is some relation between
the object to be manipulated and the outcome of interest such that the former can be
exploited to modify the latter. Processual knowledge is causal knowledge of a specific type
which, as such, is prima facie relevant to this end.29
Even though the idea of case studies as relevant to policy making seems to have some
appeal, the debate among philosophers and social scientists is still poorly developed.
Several research avenues have, however, been opened and might be worth exploring. As
early as the mid-70s, leading political scientist Alexander George worried about the gap
between policy maker e is
concerned about the conditions present in the field of international relations at that time
and fears a bias in current research towards grand theories at the expense of research with
higher practical relevance (George 1976, 1994, 1997).
George takes a modest though challenging stance. He admits that the gap between
policy making and scholarly research cannot be eliminated. Political, financial, ethical, and
practical considerations, in fact, inform decision making besides the factual knowledge
that science provides. Nevertheless, George urges, the gap could be bridged by scholarly
knowledge that fulfills the standards for relevance. In sum, scholars face a choice between

29
Relevance as defined in section 1.3.3 presupposes validity to some extent. In Chapter 2 I will
characterize the conditions in which case studies are likely to deliver valid causal knowledge.
taking into consideration the need of policy makers and ignoring it while being driven by
purely scholarly interest.
The plea for more relevant research by field specialists is voiced again and articulated
further in the later work by George and Bennett (2005). They remark that the grand theories
of scholarly research in international relations are generalizations broad in scope and
probabilistic in form; as such, this form of knowledge is insufficient to guide the policy
maker in the choice of the strategy to implement in target contexts. It needs to be
integrated by usable knowledge,30 which consists of propositions that, though narrower in
scope, specify the conditions under which these limited generalizations hold. In this way
usable knowledge can be guide to action because it allows the policy maker to better explain

for the adoption of a given strategy. George and Bennett see case studies as a fruitful
instrument for the production of usable knowledge.
The fruitfulness derives from the case study, which has to identify the causal
mechanism by means of which explanatory variables bring about the outcome of interest.
The focus on mechanisms, as opposed to broad generalizations, has two distinctive
benefits. First, the conditions under which the more modest generalization holds will be
clarified. Secondly, focusing on mechanisms yields a description at a lower level of
abstraction; in this way, they would identify variables that are likely to be more easily
manipulated, and hence of utmost interest to the policy makers.
The proposal by George and Bennett is certainly insightful and praiseworthy. They
make an effort to spell out what the positive contribution of case studies can be to policy

consideration. It leaves, however, much unspecified. Policy making is a complex practice

organized in various stages each of which relies on a variety of evidential inputs. The role
of case studies can be clarified further by specifying at which stages they are relevant, why,
and how they complement the other sources of evidence.

30
George and Bennett borrow the term usable knowledge from the work by Lindblom and Cohen Usable
Knowledge, Social Science and Social Problem Solving (1979).
 In this effort, social scientists and philosophers working on case studies would be
assisted by more developed strands of literature on policy evaluation and evidence-based
policy. As to the former,31 several contributors emphasized the specific and crucial role of
case studies with respects to other forms of evaluation (Eck 2006, Scriven 2008). In the
discussion on evidence-based policy, CSR still occupies a minor role if compared to other
more popular strategies of research. 32 This does not come as a surprise when one
considers the fact that case studies lie at the bottom in the hierarchies of evidence
formulated by most institutional evaluators. The scholars who see the potential of case
studies in policy making should thus feel the urge to engage in this debate as well. This
would force them to clarify further why case study evidence is distinctive, how it differs
from evidence generated otherwise, and how exactly it complements it.

1.4 CONCLUSION
Understanding CSR as an autonomous epistemic genre, rather than a (weaker) form of
the statistical or experimental methods, adds complexity to the analysis of case studies
while making room at the same time for new research avenues. This perspective draws
attention to the striking features of CSR and leads to a re-conceptualization of the old
definitions and problems. What was simply understood as a problem of control is thus re-
framed as an issue of internal validity, of which control constitutes just one specification.
Similarly, the generalizability potential of CSR is not limited to the formulation of
regularities that apply to the class of elements of which the case constitutes a representative
instance. The forms of possible generalizations from, and within, case studies are various
in content, target, and inferential modality. Finally, the discussion of the relevance of the
CSR is also liberated from the old distinction between the heuristic and justificatory
moment of scientific inquiry. In the old view the usefulness of the case study was in fact
determined by the degree of support it gives to the theory at hand. Possible uses of case
studies were either the formulation of new theories or the testing of extant hypotheses. In

31
Reference point in this literature is Pawson (2006), and Pawson and Tilley (1997).
32
As an example of very recent and important contribution to the field see Cartwright and Hardie (2012).
a broadened perspective other uses of case studies become apparent and their relationship
with theory, now more broadly understood, is accordingly more complex.
The chapters that follow are motivated by these considerations. They are informed by
a view on CSR as an autonomous epistemic genre and address the methodological
problems case studies confront by exploring some of the avenues outlined above.
In Chapter 2, I address internal validity in historical narratives. Historical narratives
are case studies that aim to formulate and substantiate causal hypotheses by articulating
descriptions of the sequences of events leading to the outcome of interest. They typically
make use of process-tracing to draw causal inference, and often rely on the additional use
of the methods of comparison. Despite the important role of historical narratives in the
social sciences, how process-tracing operates in the narratives is still poorly understood.
The debate on process-tracing in fact, even though it is growing thanks to a number of
recent contributions, is still muddy and under-developed. In particular, there are no shared
criteria to assess its epistemic contribution; moreover, the conditions proposed so far tend
to tie the validity of the findings to the use of specific kinds of evidence and are thus
unhelpful when this specific evidence is not available.
I argue that the proposed conditions are unduly restrictive and fail to acknowledge the
actual contributions which process-tracing can offer to valid causal inference. I formulate
new conditions to assess process-tracing performance in cases in which the favorable
evidential circumstances do not occur and existing criteria fail to apply. By discussing
process-tracing, I touch upon what is currently a hot topic in philosophy of science; that
is the role of causal mechanisms and mechanistic knowledge in the sciences. The majority
of scholars who are interested in process-tracing also agree that it somehow helps the
detection of causal mechanisms. The agreement, however, stops there: scholars entertain
many different notions of mechanisms and disagree on the role mechanisms actually play
in explanation and causal inference. From my discussion, it will become apparent how the
complexity of this debate contributes to render our understanding of process-tracing even
more problematic.
 In Chapter 3, I address the problem of generalizability. I provide an outline of what I
define as the traditional view on external validity. This approach is conditioned by a
statistical viewpoint on CSR and reduces external validity to issues of mere
representativeness.33 In so doing it leads the debate on the generalizability of case-study
results to a dead end as it quickly dismisses external validity as the downside of CSR. At
the same time, it suggests that CSR is comparatively stronger in providing results internally
valid. On this ground this approach recommends the use of case studies when internal
validity is the main research goal of interest, while turning to other methods when one
pursues generalizations instead. This outcome is unfortunate because, as a matter of fact,
case studies are often performed with the explicit or implicit purpose of drawing lessons
from the studied case to be carried over to new contexts yet unstudied.
I attempt to release this tension by examining the assumptions behind the traditional
view on the external validity of CSR. Some of these assumptions have already been
addressed, and actually disputed, in the current debate. In Chapter 3, I focus instead on
those assumptions that, to the best of my knowledge, have not been addressed yet and
seem to be responsible for the dead end in which the discussion among social scientists
seems to be trapped now. In particular, I suggest that the debate should focus on how
make case studies comparable rather than how select the typical case. Typicality and
comparability are concepts closely related but distinct. The traditional view conflates the
two and thus runs into confusion about what external validity is really about and how it
can be addressed in a fruitful manner. I surmise that by enhancing the comparability of
studies unnoticed room for improvement is made for formulating more reliable
assessment of the external validity of results obtained in case studies.
In Chapter 4, I discuss issues of relevance when policy making purposes are at stake.
In particular, I focus on the debate on the use and usefulness of randomized controlled
trials (RCTs) to find the key to economic and social development. The participants to this
debate agree that RCTs are affected by limited external validity, and that this impinges on

33
This is the first of the three approaches to external validity I briefly characterize in section 1.3.2.
their usefulness for policy making. They diverge, however, on the strategies to overcome
this problem. I analyze three alternatives that are found in the economic literature:
replication of RCTs, which has been proposed by the promoters of RCTs; cross-country
regressions, which have been typically endorsed by RCT-skeptics; and the causal models
proposed by James Heckman. I argue that these strategies succeed in their attempt to a
different, and limited, extent.
Proponents of the first two strategies fail to take into adequate consideration the
distinction between external validity and relevance, and treat the latter as a spill-over of
the former. Their strategies, in fact, aim to improve the external validity of causal effects
on the assumption that relevance will automatically follow. I argue that this is not the case,
because external validity and relevance are distinct concerns and should thus be addressed
separately. The proposal by Heckman succeeds in delivering causal effects that are, as a
matter of fact, more relevant to policy maker
cannot adequately address the type of problems policy makers are likely to face in
developing contexts. f prediction,
in developing contexts policy makers face problems of planning. Planning is a complex
procedure that depends on various pieces of evidence and raises several concerns. Causal
effects are but one epistemic input in this procedure; case-study evidence is also relevant
to the crucial phases of planning.