Studies in History and Philosophy of Science 62 (2017) 6e13

Contents lists available at ScienceDirect

Studies in History and Philosophy of Science

journal homepage: www.elsevier.com/locate/shpsa

Process tracing in political science: What’s the story?q

Sharon Crasnow
Norco College, 2001 Third Street, Norco, CA 92860, United States

a r t i c l e i n f o a b s t r a c t

Article history: Methodologists in political science have advocated for causal process tracing as a way of providing evi-
Available online 14 March 2017 dence for causal mechanisms. Recent analyses of the method have sought to provide more rigorous
accounts of how it provides such evidence. These accounts have focused on the role of process tracing for
Keywords: causal inference and specifically on the way it can be used with case studies for testing hypotheses.
Narrative While the analyses do provide an account of such testing, they pay little attention to the narrative el-
Process tracing
ements of case studies. I argue that the role of narrative in case studies is not merely incidental. Narrative
Causal mechanism
does cognitive work by both facilitating the consideration of alternative hypotheses and clarifying the
Case studies
relationship between evidence and explanation. I consider the use of process tracing in a particular case
(the Fashoda Incident) in order to illustrate the role of narrative. I argue that process tracing contributes
to knowledge production in ways that the current focus on inference tends to obscure.
Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction democratic peace.1 A narrative of the incidentda story that pulls

During the late-19th and early-20th centuries, as European
powers scrambled for control of the African continent, British and
French interests intersected in the North Nile River Valley. In July of
1898, a small French expedition occupied the village of Fashoda in a
move to secure a trade route between their colonies in the east
(Senegal) and west (Djibouti) of the continent. Anglo-Egyptian
forces led by Sir Herbert Kitchener were moving south from Cairo
as part of an effort to reassert control over the Sudan. After
receiving orders to investigate the French presence in Fashoda,
Kitchener marched on the village and arrived there on September
18. While cordial on the ground, the meeting sparked outrage at
home, touching off “a diplomatic crisis over the division of colonial
rights in the Upper Nile” (Schultz, 2001, p. 176). The two forces sat
waiting instructions from their respective governments with ten-
sions rising until early November, when the French government
backed down and withdrew their troops thus ending the crisis.
Although a seemingly minor episode during the European
imperialist conquest of Africa, the Fashoda Incident has been
thought interesting for a variety of reasons: its resolution shaped
colonial control of Northern Africa; it marked the last time that
French and British forces appeared on opposite sides of a conflict,
marking a shift in international alliances; and it came close to
providing a counterexample to the robust empirical generalization
that democracies do not go to war against each otherdthe
q This paper appears in a special issue of SHPS on ‘Narrative in Science’.
E-mail address: Sharon.crasnow@norcocollege.edu.
together the events surrounding the encounter at Fashodadmay
thus be told in a variety of ways. If we are interested in the history of
the European colonial presence in Africa, there are aspects of the
case that will be more relevant than othersdthe location of the
village is crucial, for example. Fashoda is roughly the intersection of
a proposed trade and communication routedthe Cape to Cairo
railroad proposed by Cecil Rhodes in 1892dand a potential east-
west link between the French colonies. The French withdrawal
gave the British their north-south pathway and secured their
control of Egypt, which had been in dispute. The story of the
Fashoda Incident can also be told so that it bears on international
relations debates about militarized conflict and the democratic
peace. On this telling, the location is less important than the nature
of the government and the dynamics between the nations. That the
nations are democracies that nearly go to war is more important
than where the incident takes place.
Different narratives of the same case carry with them frame-
works within which the story needs to make sense. Paul Roth
makes this point in his discussion of Clifford Geertz’s classic “Deep
Play: Notes on a Balinese Cockfight.” There he understands Geertz
as constructing “a particular storyline, that is, a way of reading the
event of the cockfight as a tale about Balinese society” (Roth, 1989,
1
The democratic peace is one of the few widely accepted empirical generaliza-
tions in international relations. Bruce Russett, for example, begins Grasping the
Democratic Peace with, “Scholars and leaders now commonly say ‘Democracies
almost never fight each other’,” and refers to “The Fact of Democratic Peace”
(Russett, 1993, p. 3).

http://dx.doi.org/10.1016/j.shpsa.2017.03.002
0039-3681/Ó 2017 Elsevier Ltd. All rights reserved.
 S. Crasnow / Studies in History and Philosophy of Science 62 (2017) 6e13
p. 450). As Roth puts it, “A narrative is not determined by
sequencing some prior set of events. Rather, what comes first is
some more general view of what counts; the particular eventsdthe
elements relevant to one’s narrativedemerge from this” (Roth,
1989, p. 455).
For political science, such frameworks, and consequently the
stories they inform, are usually causal. If, for example, the Fashoda
Incident is the story of a “near miss”dthat the two democracies,
Britain and France, nearly went to wardthen the causal factors that
sustained the peace are most salient. But how are such factors to be
identified? One answer to this question is that they are identified
through hypothesized causal mechanisms. Such hypotheses might
be either empirically or theoretically drivendthat is, suggested by
empirical observation or by theories of why nations do and do not
go to war. A number of such theories appear in the international
relations literature. Testing various hypotheses against the events
as they occurred through a case study of that event is one way of
providing evidence for or against these competing hypotheses. One
method proposed for uncovering this evidence is causal process
tracing.
Causal process tracing has been characterized as a way of
“formulating and testing explanations with case studies”
(Mahoney, 2015, p. 200). The value of process tracing is thus
thought to rest in how it provides evidence for causal mechanisms.
Statistical and experimental methods may be able to establish a link
between a dependent and independent variable, but they cannot
reveal what is in the “black box”dthe mechanisms through which
the cause brings about the effect. The idea is that if we are able to
trace the process, we will be able to identify the causal mechanism
as it is operating in a particular case.
Recently, characterizations of process tracing have been accom-
panied by an effort to provide a more rigorous account of how the
method works to provide evidence.2 The focus on the use of process
tracing for testing hypotheses in these efforts has indeed provided a
more rigorous characterization. These accounts have helped to
clarify the variety of approaches that are described under that name,
offered suggestions for how the method provides evidence for
causal inference, and urged greater transparency in how process
tracing is used, but these approaches have also moved the under-
standing of the method away from its roots in the narrative elements
of case studies. Nina Tannenwald remarks that while these recent
accounts have value because they make process tracing more
transparent and rigorous, she warns that the “‘armature’ of method
should not become such a fetish that it overwhelms the narrative”
(Tannenwald, 2015, p. 227). She continues, “It is a tall order to be both
methodologically rigorous and narratively engaging but that is a
worthy goal to which security studies can continue to aspire.”
(Tannenwald, 2015, p. 227). Although Tannenwald couches her
remark in terms of aesthetic appeal, in this essay I argue that the role
of narrative is not merely aesthetic. A consideration of the narrative
elements of case studies illuminates how process tracing contributes
to knowledge production in ways that the current focus on inference
tends to obscure.
The structure of the essay is as follows. I start by giving an ac-
count of the key concepts used and the relationships among them:
process tracing, causal mechanism, and narrative. I next consider
Bennett’s reconstruction of Kenneth Schultz’s use of process tracing
to evaluate various hypothesized causal mechanisms offered to
account for the Fashoda Incident. I argue that Bennett treats
Schultz’s reasoning as a matter of constructing arguments using
process tracing as a means of identifying evidence to be used in
2
I specifically have in mind Beach and Pedersen (2013) and Bennett and Checkel
(2014).
7
those arguments. Bennett’s account serves as an example of
problems with an approach to process tracing that isolates the el-
ements of the hypothesized causal mechanism uncovered through
process tracing and downplays the cognitive work played by the
narrative elements of case studies.
Specifically, I identify two areas in which elements of narrative
are core to process tracing: 1) consideration of alternative hy-
potheses and 2) the relationship between evidence and explana-
tion. The narrative of the case guides process tracing so that it does
more than uncover events, actions, and entities that are relevant for
testing hypotheses. To illustrate this, I offer an alternate reading of
Schultz’s use of process tracing, arguing that his reasoning depends
on a more holistic reading of the eventsdone in which they are
made coherent through a narrative. In this way, I argue that process
tracing requires creating a narrative through a hypothesized causal
mechanism and that involves more than uncovering pieces of ev-
idence to test a hypothesis. The hypothesized causal mechanism
not only provides a framework from which information to test
hypotheses can be gleaned, but the narrative does cognitive work
through making the causally salient elements of the case coherent.
2. Process tracing
The key idea behind process tracing is that through seeking the
key elements of a hypothesized causal mechanism within a case, it
should be possible to identify whether the mechanism is oper-
atingdthat is, one should be able to trace the mechanism from the
cause to the effect. This might be forward lookingdstarting with
the cause and moving to the effectdor backward lookingdstarting
with the effect and tracing its origin in the cause. The case might be
taken to be exemplary and so provide the grounds for some
generalization or hypothesis (theory-building) or it might be a case
in which a hypothesized causal mechanism is thought to be oper-
ating. The examination of the case provides evidence that the
mechanism is or is not operating (theory-testing). George and
Bennett put it this way: “Process-tracing attempts to empirically
establish the posited intervening variables and implication that
should be true in a case if a particular explanation of that case is
true” (George & Bennett, 2005, p. 147).
Bennett and Checkel describe process tracing as “the examina-
tion of intermediate steps in a process to make inferences about
hypotheses on how that process took place and whether and how it
generated the outcome of interest” (Bennett & Checkel, 2014, p. 6).
This characterization closely follows George and Bennett’s
description of process tracing as a method that “attempts to iden-
tify the intervening causal processdthe causal chain and causal
mechanismdbetween an independent variable (or variables) and
the outcome of the dependent variable” (George & Bennett, 2005, p.
206).
Unlike quantitative observational and experimental methods
that work with larger populations and produce evidence of average
effects, process tracing requires the careful examination of a single
case or a few cases and provides evidence for singular causation.
That is, rather than providing evidence for the distribution of Y, given
X in a population, process tracing provides evidence that an instance
y of Yoccurred in this case. Furthermore, process tracing is thought to
give insight into how y occurred. For methods that are appropriate
for populations, we may find ourselves both wondering how X
brought about the distribution of Y in the population (what is the
causal connection between X and Y?) and whether instances of Y
will occur in individual/event/activity a or not since there is no
assurance that any singular case will exhibit an instance of Y just
because Y is on average likely to be present in the population.
Conversely, if we know that X brought about Y in n we cannot infer
that X will do the same in different populations (or subpopulations
8 S. Crasnow / Studies in History and Philosophy of Science 62 (2017) 6e13
of the original examined population) n1, n2, n3, or . nz. Different
methods thus appear to answer different questions and so it is not
surprising that combining methods has been advocated in order to
answer both sorts of questions.
The idea that process tracing will allow for a more complete
causal understanding is intuitively appealing, but how it does so is
not obvious, although the link appears to lie somewhere in the ap-
peal to causal mechanisms. Advocates of process tracing argue that
close examination of the intervening events, activities, or entities
occurring between purported cause and purported effect can pro-
vide evidence for the causal mechanism and thus provide the how.
This suggests that a first step for understanding process tracing is
understanding what is meant by “mechanism” and what the rela-
tionship between the process that is traced and the mechanism is.
Mechanisms have been explicated in a variety of ways and the
preferred understandings vary among disciplines. Still, the main
features of mechanisms that are relevant to the discussion here can
be seen in most characterizations. One of the best known and
frequently referred to in the social science literature is Machamer,
Darden, and Craver’s: “Mechanisms are entities and activities
organized such that they are productive of regular changes from
start or set-up to finish or termination conditions” (2000, p. 3).
While they have biological mechanisms in mind, there appear to be
similarities with social mechanisms of the sort in which political
scientist are interested. Thus, we see a similar characterization
offered by Glennan,: “A mechanism for a behavior is a complex
system that produces that behavior by the interaction of a number of
parts, where the interactions between parts can be characterized by
direct, invariant, change-relating generalizations” (2002, p. S344).
The main difference between these two is that Machamer,
Darden, and Craver emphasize entities and activities organized
with a starting point and ending point whereas Glennan is focused
on the interactions that take place in a system. Several philosophers
of science have claimed that these two characterizations of mech-
anism are not so different. James Tabery has argued that these
definitions should be synthesized (Tabery 2004) whereas Phyllis
Illari and Jon Williamsons suggest a “consensus concept” of
mechanism (Illari and Williamson, 2012). Illari and Williamson’s
preferred characterization is “A mechanism for a phenomenon
consists of entities and activities organized in such a way that they
are responsible for the phenomenon” (Illari and Williamson 2012,
p. 123) and. to clarify their understanding, they look at each of the
three parts of their consensus approach in turn: 1) responsible for
the phenomenon; 2) entities and activities; and 3) organization.
These accounts capture important aspects of the notion of
mechanism. However, if the idea of a synthesis is taken seriously, I
propose the following as a better characterization of the core fea-
tures of mechanisms: 1*) mechanisms have partsdthey may be
identified as activities, events, or entities, but the parts are in some
sense discrete (if only analytically); 2*) the parts are organized in
some waydthe mechanism has a structure; and 3*) there is an
active element that is seen through the inter-relationship of the
partsdall characterizations of mechanism include the idea that an
effect is brought about, produced, propagated, or maintained
through the inter-relationship of the parts.3
3
These elements appear to be implicit in Illari and Williamson’s account with 2*
and 3* packed into their 2. Machamer, Darden, and Craver and Glennan also include
that the changes produced are regular e the interactions among the parts are
invariant and general, however it is not a consensus that regularity should be
required for mechanisms (Illari and Williamson, 2012 do not include regularity, nor
do Bogen, 2005; Reiss, 2008) nor that regularity should always be understood as
embedded in a structure (Reiss, 2008, p. 110). Since this is a disputed feature and
nothing that follows depends on understanding mechanisms as regular or invariant
I leave it out of the characterization offered here.
For process tracing to be possible, mechanisms must have parts
and those parts need to be accessible to researchers. But the other
features that characterize mechanisms play important roles as well.
In the next section, I consider hypothesized causal mechanisms
offered to account for the outcome of the Fashoda crisis Incident as
concrete examples. and I argue that an analysis of process tracing as
a search for diagnostic pieces of evidence fails to capture the way
that a mechanistic account seeks to address the inter-relationship
of the parts in a way that the narrative elements of case studies can.
3. Process tracing and the Fashoda Incident
Andrew Bennett (2010, pp. 207e220) has analyzed Kenneth
Schultz’s treatment of the Fashoda Incident in Democracy and Co-
ercive Diplomacy (2001) as a use of process tracing to empirically
test several different approaches to understanding international
conflict and ultimately to support Schultz’s preferred account. As
Bennett sees it, Schultz seeks out pieces of evidence each of which
provides a specific sort of inferential test for the various hypothe-
sized causal mechanisms under consideration. To provide back-
ground for the analysis, I first give a brief account of the alternative
approaches under scrutiny.
There are two main approaches that can be identified in the
literature on international conflict. The first stems from the robust
empirical generalization that democracies do not go to war with
each otherdthe democratic peace. The democratic peace suggests
a role for domestic politics in explaining international conflict
however leaves open the question of how democracy might play
this role. The two main approaches offered to account for the
democratic peace are normative and institutional. Normative ap-
proaches propose two sets of norms. The first is the socialization of
democratic leaders to have a preference for nonviolent dispute
resolution that decreases the probability of war between demo-
cratic nations. The second set of norms is the mutual recognition on
the part of leaders that governments in liberal democracies reflect
the wishes of their citizens. Democracies, it is claimed, recognize
that they will be seen by other democratic states as legitimate
because of this. This mutual respect increases the likelihood of the
peaceful resolution of disputes since demands are recognized as
legitimate and both parties seek common ground to resolve dif-
ferences (Schultz, 2001, pp. 12e13).
Institutional approaches begin with the supposition that states
are less likely to go to war when those who would bear the cost of war
are involved in the decision making. Democratic institutions provide
the means for the representation of those who would bear the costs of
war since they make the government answerable to the citizens.
“Thus, democratic leaders face costs of war to which their nondem-
ocratic counterparts are less vulnerable” (Schultz, 2001, p. 14).
While these approaches seek to explain the democratic peace, a
second theoretical framework, neorealism, challenges that de-
mocracy or any other domestic political configuration is relevant in
considering the causes of war. Neorealists argue that conflicts of
interests among nations are inevitable and that wars occur when
the balance of power is such that one nation can militarily domi-
nate another. Cases of democracies that do not go to war in spite of
conflicting interests can be explained based on differential military
power (Layne 1994).4
4
Neorealists also point out that the period most relevant for evidence of the
democratic peace e the rise of democracies during post World War II e coincides
with the Cold War, which they identify as bearing the primary responsibility for
peace among democracies during this same period.
 S. Crasnow / Studies in History and Philosophy of Science 62 (2017) 6e13
Schultz offers theoretical, statistical, and case-based arguments
against each of these approaches, and proposes an alternative.5
Using a fairly simple bargaining game model as a tool for anal-
ysis, Schultz proposes an account that he identifies as falling be-
tween the democratic peace and neoliberal frameworks. He
identifies informational asymmetries (lack of information about
interests, intent, resolve, or military strength) as among the causes
of the failure in bargaining between nations that lead to war and
elaborates by identifying features of international disputes that
make it difficult to overcome such asymmetries (Schultz, 2001, p.
18).
As part of his argument, Schultz explores several cases through
which he tests the viability of these various hypotheses, including
his own. The Fashoda Incident is one such case. He argues that
neither neorealism nor the democratic peace hypotheses can ac-
count for the details of this case as well as the asymmetrical in-
formation hypothesis he proposes. According to Schultz, the
dynamics surrounding international conflict resolutiondboth
successful negotiations that preserve peace and unsuccessful ne-
gotiations that end in wardcenter on incomplete and asymmetrical
information. Although factors such as the balance of power are
important, these are mediated through beliefs and information and
so their effects are not necessary predictable (Schultz, 2001, p. 24).
Resolution of conflict fails primarily because of lack of access to the
information that is needed to assess the credibility of a nation’s
threats to go to war. When interests conflict, parties involved in
negotiations have the motivation to deceive their rivals making it
difficult to assess the credibility of threats. Such assessment relies
primarily on interpretation of signals, both those directed to the
rival nation and those directly internally to the domestic audience.
Schultz argues that attention to information and signaling issues
provides causal mechanisms through which the democratic peace
can be understood.
He spells this out by looking at the effects of democracy on
signaling the credibility of threats issued by state leaders. He
identifies two mechanisms that work to create clearer signals: a
restraining effect of democracy and a confirmatory effect. The fact
that democracies must answer to their constituents in a trans-
parent environment has a restraining effect on democratic leaders’
ability to hide their true intentions from their adversaries resulting
in less opportunity for deception. The result is that threats are
perceived as more credible. In addition, because of these same
features, when democracies make threats that are clearly backed by
public opinion these threats are highly credible. They are rendered
even more credible when opposition parties in democracies sup-
port them. Again, because in a democracy the voicing of opposition
support is visible to rival nations, it has a confirmatory effect,
increasing the credibility of threats. Schultz thus proposes causal
mechanisms through which the peace is sustained in democracies.
e the government is restrained in making threats because of the
transparency, thus thereby increasing their credibility when threats
are made. Likewise, public and opposition support also increases
and the credibility of the threats. is also increased when confirmed
by public and opposition support. The credibility of the threats thus
5
In response to normative accounts he notes that a shared preference for peace
does not always prevent war. In considering institutional arguments, he points out
that frequently the cost of war for dictators is greater than it is for democratic
leaders. Dictators frequently risk losing their freedom or their lives as a result of
taking their nations into unpopular wars whereas democratic rulers merely lose
office. Against neorealists he argues that they focus on conflict of interests and
balance of power but do not do a good job of explaining why rational actors ever
chose such a costly option as war.
6 9
Recall that the costliness of war raises a serious question about why nations
would ever go to war in the first place.
9
motivates alternative solutions to costly military conflict.6 Autoc-
racies do not have this transparency, making signals are harder to
read and thus, Schultz argues, increasing the likelihood of a crisis
escalating into military conflict.
As Bennett analyzes Schultz’s use of process tracing, Schultz
seeks pieces of evidence that either conflict with or comport with
the hypothesized causal mechanisms that are offered in each the
approaches. For example, Bennett considers Schultz’s argument
against the neo-realist balance of power approach. On that account,
the explanation for France backing down rests in Britain’s superior
military strength (Layne 1994); e the democratic nature of the two
governments is not relevant. According to Bennett, Schultz shows
that this hypothesis cannot explain a number of features of the
casedthat the crisis happened in the first place, that it lasted two
months, and that it came so close to wardsince the fact of Britain’s
superior military strength was known to the French from the outset
(Bennett, 2010, pp. 211e212). Since each of these pieces of infor-
mation is visible in the case, the balance of power account is
incomplete and its failure to account for aspects of the case means
that this hypothesis is inadequate.7
In contrast, Schultz’s own signaling hypothesis is supported by
an in-depth examination of the case that brings forth relevant
diagnostic pieces of evidence. These show that the credibility of the
British refusal to back down was signaled by the government, the
opposition, and the public. Specifically, Bennett mentions the
public support by the opposition for the British refusal to back
down (confirmatory signaling). In addition, he cites evidence of
apathy on the part of the French public (restraining the French
government from successful deception and thus making it difficult
to force a compromise). “In sum, the close timing of these events,
following in the sequence predicted by Schultz’s theory, provides .
evidence for his explanation; this, combined with the alternative
explanations’ failures ., makes Schultz’s explanation of the
Fashoda case convincing” (Bennett, 2010, p. 212).8
While Bennett does mention that Schultz predicts a particular
sequence of events, Bennett’s focus is primarily on the pieces of
evidence and how they bear inferentially on the various hypotheses
under consideration. The failure of the neorealist account to pass a
test is interpreted as a refutation of that hypothesisdan inference
that the causal mechanism proposed is not in operation in this case.
Bennett next interprets the evidence in support of Schultz’s
asymmetrical information hypothesis as a form of Bayesian
updating. The subjective degree of belief that the hypothesized
mechanism is the best available account is increased with each
piece of evidence produced. The relationship between these
discrete bits of evidence and the hypothesized causal mechanisms
for which they provide evidence is that they are events, activities,
or entities that would be expected if the causal mechanism were
operating in this case. Finding them should increase the degree of
belief in the hypothesis.9
Bennett’s emphasis on the discrete units of evidencedthe
partsdseems to pull away from the idea that process tracing is
7
The idea is that minimally the hypothesis should account for all the facts of the
case and in failing to do so the hypothesis should be eliminated. Of course, doing so
does not confirm the hypothesis but only means that it is still on the table. Bennett
identifies various types of tests that evidence might provide following a categori-
zation proposed by van Evera (1997). I discuss these tests in Crasnow, 2012, p. 663.
8
Bennett also briefly discusses Schultz’s treatment of the democratic norms and
institutions hypotheses, however, Schultz’s own discussion of them differs from his
discussion of neorealism in that he identifies them as conflicting with the case
whereas the neorealist account is merely incomplete. I will return to this point in
Section 5.
Bennett also develops this Bayesian analysis in more detail in Bennett, 2014.
Beach & Pedersen, 2013 also analyze process tracing in terms of Bayesian updating.
10 S. Crasnow / Studies in History and Philosophy of Science 62 (2017) 6e13
seeking evidence for hypothesized causal mechanism as a whole.
Another way of putting this, referring back to the core features of
mechanisms in Section 2, is that he emphasizes that mechanisms
have parts (1* of my account of mechanisms), but downplays or
neglects that the parts are organized in some waydthey have a
structure (2*)dand the active element that is seen through the
inter-relationship of the parts (3*). While it seems clear that finding
parts of the mechanismdfor example, the speeches made in sup-
port of the British government’s stance by the oppositionddoes
count as evidence, the telling of the tale (the narrative) involves
taking the mechanism as a skeleton (organizing structure) and
moving the story forward through the causal nature of the mech-
anism. These organizational features make the story more
compelling than the additive effect of such bits of evidence in-
dicates. The mechanism is the skeletondthe framingdof the
narrative through which the story is told and the holistic role that
this framing plays is not apparent if process tracing is treated as a
project of finding diagnostic pieces of evidence with which to make
inferences. To see what is missing more clearly, I turn to a discus-
sion of narrative before reconsidering Schultz’s account of the
Fashoda Incident.
4. Narrative, process tracing, and causal mechanisms
I have claimed that uncovering the “diagnostic pieces” of evi-
dence that are found within case studies through process tracing
only provides a partial understanding of how process tracing con-
tributes to knowledge production. Tracing the process depends on
being able to tell the story of the case. A case study is not a list of
relevant features (nor is it a list of all of the features surrounding
the event) and process tracing is not merely a matter of sifting
through those features for evidence.
The first feature of narrative relevant here is that narratives have
a starting point and an end point. I will refer to this feature as
“narrative closure.” The narrative takes us from the start to the end
of the story by moving us along through a series of events that
unfold over time. Although they need not be presented in the exact
temporal order over which they unfold, their temporality is
important since it gives the narrative structure through revealing
the inter-related nature of the elements of the story. The “closure”
that I am referring to is the “endedness” of the story that de-
termines which pieces of the story are to be connected.
To relate this to the discussion of the previous section, tempo-
rality is one way of organizing the parts of the mechanism and the
movement of time from the starting point to the end point offers a
path along which to trace the process. (See Morgan in this issue for
other ways narrative organizes parts of a whole.) In the case of the
Fashoda Incident, when the point of telling the story is its relevance
for the democratic peace, the end point of the story is the peaceful
resolution of the conflict. The story is over when the conflict has
ended e. the hypothesized mechanism directs attention to the el-
ements of the story that lead to that end point.
Process tracing also must choose a starting pointdsome event,
entity, or activity that sets the mechanism in motion. The case of
the Fashoda Incident takes as its starting point the initiation of the
crisisdthe point at which the British meet the French outpost at
Fashoda. There is a sense in which that beginning point is some-
what arbitrary. Where we start the storydwhere we begin tracing
the processddepends on a variety of things. It depends on our
interests, background knowledge, and other features of context. If
the story of the Fashoda crisis is told in the context of the more
general question of colonial power in Africa, the story might begin
earlier. Again, when the concern is the resolution of crisis, the point
at which the crisis becomes evident seems appropriate.
In the case of process tracing in political science, the beginning
points and end points often derive from a theoretical framework
that informs the hypothesis. The Fashoda Incident example is like
this in that it becomes interesting in relation to the theories of
international conflict and more specifically in relation to the
democratic peace. Theoretical frameworks and background
knowledge point to possible causes and effects in a similar way. For
example, a multiple regression analysis that focuses on the rela-
tionship between levels of wealth and democracy treats them as
the relevant variables in part because of theories about the rela-
tionship that have made it an established research area in political
science.
A second feature of narrative relevant for thinking about process
tracing is the generative nature of events and actions. In a narrative
the events do not merely follow one another, they are connected to
one another. Noël Carroll (2001) has argued that the sequence of
events in a narrative depends on causal relations among the
events.10 While not everyone agrees with this account, it surely
seems that at least some narrative is moved from its beginning to
its end through causal relationsdthe reasons of the actors, the
events occurring that produce changes in the options available to
the actors, and the structures in which the actors must act, all
operate as causes. The story not only has structure but the sequence
of events is one in which they are driven along towards the end by
these causes. Such causal relations are similarly the focus of process
tracing. Finding the traces of the mechanism in the story means not
only finding the parts but finding the organizational structure and
the active elements through which the endpoint is reached.
Just as this causal nature of “narrative connection” (Carroll,
2001) distinguishes narrative from a temporal series of events,
the generative, productive, causal nature of mechanisms also
distinguish a mechanism from set of elements that could be treated
as independent and dependent variables even when taken together
with all the intervening variables.11
Another way that narratives work to move the story from its
beginning to its end is through an elimination of alter-
nativesdpossible paths that could have been taken but were not.
Sometimes storytellers are explicit about what these options are,
sometimes the reader fills in the alternative scenario with implicit
background knowledge of the circumstances the actors face. As
John Beatty puts it, “A narrative indicates or alludes to non-
actualized possibilities, and their consequences, by treating
events . not just as points along the way to the outcome, but as
“‘turning points’” .” (Beatty, 2016, 36; also see Beatty in this issue).
Beatty goes on to distinguish two senses of “contingency” that are
illustrated through this feature of narrative: 1) forward looking
contingencydthe idea that there are possibilities in front of us that
are eliminated at the turning points, and 2) backward looking
contingencydwhat appeared to be a matter of chance latter comes
to seem necessary.
There are several features of Beatty’s characterization that seem
particularly relevant for process tracing. One worry often raised
about the use of case studies is the idea that it may devolve to
cherry picking or just-so stories. This is indeed a concern but one
10
The sense of cause he uses is “a necessary or indispensable contribution to a
sufficient, though non-necessary, condition for the occurrence of the relevant later
event in the narrative complex (or, in other words, it is, at least, what J. L. Mackie
calls an INUS condition)” (Caroll 2001, pp. 123e124). I am not committed to un-
derstanding cause in only this way. I am more inclined to think that there are a
wide variety of ways that “cause” is understood and that any of them could provide
the narrative connection Carroll describes.
11
I mention this alternative specifically because King, Keohane, and Verba (1994)
have argued that mechanisms have no unique characteristics that cannot be un-
derstood through such a sequence of variables.
 S. Crasnow / Studies in History and Philosophy of Science 62 (2017) 6e13
that can be addressed by requiring that all of the relevant details of
the case be considered and not just those that are relevant to the
favored hypothesis. In order to assure that these details are
addressed, alternative hypothesesddifferent ways that the story
could have gone, different paths that could have been taken,
different mechanisms through which the case can be under-
stooddneed to be explored.12 Forward looking contingency directs
the researcher to identify crucial turning points. If the details of the
case cohere with the mechanismmechanism, then the turning
points should be apparent in the mechanism as points at which the
operation of the mechanism could be disrupted. The successful
tracing of the mechanism suggests that the outcome was inevita-
bledor at least highly probable. But alternative pathsdalternative
mechanismsdmust also be explored. The case study has to be
detailed enough that alternatives are clear, can be explored, and the
story makes clear why the alternatives were not actualized.
There is another way that the narrative of the case aids in
thinking through the viability of various accounts. The counter-
factual thinking that causal claims support requires considering
options that are as close as possible to the original case. But how are
we to know what is a minimally different counterfactual scenario
without a clear understanding of the actual? And what is actual
includes more than the “furniture”dthe diagnostic bitsdof the
case. It also includes the connections between the actors, circum-
stances, and events. Counterfactual thinking parallels the backward
looking contingency that Beatty describes through exploring how
changes in the past might have resulted in a present unlike that
actual. When the story makes clear why such alternative scenarios
did not come to pass, the outcome seems inevitable or nearly so.
(See Beatty in this issue for more on narratives and counterfactuals.)
These parallels suggest that if narrative has cognitive force then
that force could well be present in process tracing. I have thus far
argued that we can see this in the causal explanatory aspect of
narrative that Carroll focuses on and in both the forward and
backward looking contingency that Beatty discusses. I turn now to
re-examine Schultz’s use of process tracing in the case of the
Fashoda Incident with these insights about narrative in mind.
5. Telling the story e: The Fashoda Incident again
Schultz’s account of the Fashoda crisis Incident goes on for 20
twenty pages of text (Schultz, 2001, pp. 177e196). In those 20
twenty pages, he tells the story of the crisis in detail, focusing on
events, actors, and circumstances of the case. While he focuses on
particular events that are relevant to the each of the examined hy-
potheses, in the course of considering them he gives a rich account
of events and actions surrounding the crisis. The details I highlight
here are intended to give a sense of how Schultz weaves the
narrative but they do not reproduce it. His strategy is to provide a
narrative through the various hypothesized causal mechanism. In
doing so he seeks out events, actions, and entities that the hypoth-
esis in question relies on to connect those events and generate the
outcome. The balance of power mechanism, the democratic peace
shared norms and institutional mechanisms, and then ultimately his
signaling and asymmetrical information mechanism are each
explored in this way. Each provides, as Roth suggests of narrative, a
way of reading the events. Schultz’s strategy is not to uncover events
or actions that refute or support the hypotheses, as in Bennett’s
analysis, but is the more nuanced argument that telling the story
through one causal mechanism is more compelling than through
12
Stephen Gould and Richard Lewontin emphasize this in their 1979 paper “The
Spandrels of San Marcos and the Panglossian Paradigm: A Critique of the Adapta-
tionist Programme”.
11
others. Each hypothesis proposes a mechanism that serves as an
abstract organizing principle through which he concretizes the
details of the case.13 Schultz’s argument is that his account does so in
a way that provides a more coherent storydin a sense of “coher-
ence” that includes completeness, consistency, and credibility.14
He starts with the neorealist hypothesis. He makes clear that the
superior military strength of the British both in Fashoda itself and
more generally (he particularly notes the superior naval power of
Britain) was known to the French. This creates a puzzle for the
balance of power story in that it leaves unexplained why the French
would have challenged the British at all, why they would have
remained in Fashoda for two months as the crisis escalated rather
than leaving once it was clear that the British would not compro-
mise, and why they ultimately did concede when they did. Schultz
considers how the balance of power account might be preserved.
He notes for example, that there is fairly strong evidence that the
French never intended to stay in Fashoda but rather hoped to force
a compromise that would improve their position in this portion of
Africa and so challenge the presumed dominance of Britain in the
area. The challenge is consistent with recognizing the superior
military strength of Britain, since it takes France as never intending
to go to war, but it leaves unexplained why the French held their
position when that strength was so overwhelming and a compro-
mise did not appear to be forthcoming. Schultz considers the
possibility that the French had hopes that a third nation, either the
Germans Germany or the Russians, would join them as allies
against the Britishdsomething that would account for their
persistence. But he eliminates these alternative pathwaysdother
ways that the story might have gonedby producing evidence of
events and circumstances that close them off. French and German
animosity was deep and so they were unlikely allies in any case. The
Russian navy would be ice bound for the winter and so unable to be
counted on as an ally against the powerful British navy.
In order to avoid telling just just-so stories, process tracing
needs to be done in a way that does not pick out only the parts of
the story that fit. The consideration of alternative hypotheses and
their elimination is crucial for a convincing story. But the focus on
the elimination of alternative hypotheses tells us little about what
alternatives to considerdwhich options among all the possibilities
are “live.” While alternative hypotheses can be suggested by the
political science literature and general background knowledge, the
details uncovered in the case study also suggest alternatives.
Beatty’s notion of backward looking contingency is helpful here.
The review of options and their elimination as Schultz considers the
neorealist approach illustrates this. He asks, how could we have
gotten here, given that the balance of power hypothesis makes it
unlikely? In this way, he generates alternative possibilities that
need to be examined and eliminated if he is to build an argument
against the balance of power account.
In relation to the democratic peace accounts, Schultz notes that,
“Whereas the neorealist story fails primarily by being incomplete,
elements of the democratic peace story are directly contradicted by
many aspects of the crisis” (Schultz, 2001, p. 180). During the two
months in which the Fashoda crisis Incident unfolded, the French
government was under pressure due to the Dreyfus Affair and there
was a real concern that the democratic regime could collapse. Both
the democratic norms and institutions hypotheses for the
13
That the relationship between cases and hypotheses or theories is one of the
concrete to the abstract is proposed by Chang (2012) in discussing the relationship
between historical cases in science and the history and philosophy of science.
14
My use of “credibility” here follows Morgan (2012) and differs from the sense of
“credibility” in the discussion of the credibility of threats. That terminology is
common in international relations discussions of signaling and refers to the belief
that the threat will be acted upon.
12 S. Crasnow / Studies in History and Philosophy of Science 62 (2017) 6e13
democratic peace seem inconsistent with the British response to this
French domestic crisis. “Rather than working against the jingoism of
the public, Salisbury and his ministers cultivated the public’s hos-
tility in order to convince French decision makers that they could not
grant concessions” (Schultz, 2001, p. 186). Upping the pressure on a
weakened democracy does not appear to be consistent with the idea
that mutual respect for democratic values and institutions is what
moderates bargaining between democratic nations.
Bennett does not distinguish between these two different sorts
of failingsdincompleteness and inconsistency. This too is because
Bennett focuses on pieces of evidence that he sees as playing a role
in telling inferencesdtestsdof the theory. Schultz, on the other
hand, is more focused on how well the story hangs together. The
gaps in the story correspond to failures to show the in-
terconnections between the actions, entities, and events that are
implicated in the processdthe elements that we should see if the
hypothesized mechanism is operating. The inconsistencies, on the
other hand, are expectations that contradict the account of events
as they happened. We can treat these as two ways that the coher-
ence of the story can fail.
In describing an account as “coherent” I mean to indicate
minimally that it is both consistent and complete. For the case at
hand, Schultz argues that thinking of the Fashoda Incident in terms
of balance of power will not produce a complete account and that
telling the story in terms of the democratic peace will not produce a
consistent account. But coherence is not just completeness and
consistency of a story. It is also the way these features contribute to
our confidence that the story is a good onedto its credibility, as
Morgan puts it (2012, 674). In considering how case studies work as
support for a theory, Morgan uses Donald Campbell’s idea that they
can both infirm (weaken) and confirm (strengthen) our confidence
in a hypothesis (Morgan, 2012, p. 670). Case studies contain so
many potential points of fitdso many details that might not be able
to be accounted fordthat when they do all fit into an account, our
confidence that this account is a good one should be increased. This
is not just a matter of fitting the points to the story as one does in
statistical curve fitting, but rather having the case fit into the story.
Credibility is the credibility of the case when told through the
causal mechanism.
Schultz argues that understanding the case as told through his
hypothesized signaling mechanisms produces the bestdmost
complete, consistent, and credibledstory of the options considered
e, and that it is the most coherent. To support this argument, he
considers both the British and French domestic environments and
their transparency to each other. First, he notes that incomplete
information contributes to the French move on Fashoda in the first
place. The prime minister, Lord Salisbury had a reputation for
making “graceful concessions” (Schultz, 2001, p. 185) and so chal-
lenging the British might not have seemed unreasonable to the
French given that their assessment was that the risk of war was not
high even though Britain’s military strength was so much greater.
However, over the next few months the signals from Britaindboth
the public outcry and the outcry from the Liberal opposi-
tiondindicated that the British resolve and the pressure on Salis-
bury not to compromise were to be taken seriously. Backward
looking contingency is also useful in that it suggests counterfactual
scenarios. Again, the details of the casedthe context in which the
hypothesized mechanism operatesdare important. The counter-
factual scenarios that are worth considering are those that are
minimally different from the actual events. To know which they
aredto know what makes sense to probe as an alternative sce-
nariodrequires knowing what could possibly have been different
in the case at hand. Rather than treating the details of the story as
context against which we evaluate the hypothesis, they are the
relevant content of the story. This supports backward looking
contingency, indicating alternative paths that events could have
taken. Finding that they did not go that way solidifies our under-
standing of the actual course of events.
In this case, the causal mechanism signals the pivotal
pointsdforward looking contingency. The mechanism directs the
search for features that will produce a coherent story by looking for
“moments” when crucial information might have been conveyed
but was not. If the French had realized that Salisbury would be
under pressure not to compromise they would not have misjudged
the risk of war. These pivot points are then used to account for the
path taken. The French back down when it is clear that there is
unified British support for war (confirmatory signals) and when it is
also clear that the French public is apathetic about Africa and
opposed to engaging the British (restraining signals).
In a narrative, one element that makes the story worth telling is
the awareness that things might have been otherwisedthat the
story could have turned out differently. If we think of process
tracing in terms of testing hypotheses it is not clear how we get that
sense of contingency., but if we think of the case as constructing a
narrative, these elements are highlighted. The evidence that we
find through process tracing is evidence that things did not go in
some other way that they could have. The possible thus informs the
actual and clarifies the nature of the causal relationships outlined.
In this way, it is related to and perhaps buttressed by counterfactual
reasoning rather than an alternative to it (as Reiss, 2015 argues).
In summary, while Schultz’s case study can be reconstructed as a
series of tests of alternative hypotheses as Bennett does, to do so
fails to address both the way constructing a narrative guides pro-
cess tracing and the way that the story provided through one hy-
pothesized mechanism has greater coherence than through the
others considered.
6. Conclusions
I have argued that reconstructing process tracing as inference
does not capture key elements of how process tracing produces
knowledge. If we understand reading the case as seeing it through a
causal mechanism, then those connections are more clearly visible.
In addition, what appears to be context on an inferential reading of
process tracing becomes content when we understand the case as
narrative.15 It is through the narrative elements of the case that the
salience of this content to other aspects of knowledge production is
revealed.
The causal mechanism shapes the story line that helps us
identify what we should look for in the case. The details of the
narrative told through that mechanism also clarify what counts as
relevant alternative hypotheses and supply the dynamic or gener-
ative features that are hallmarks of a mechanism. Successful pro-
cess tracing reveals the coherence of the events when understood
through the causal mechanism. While process tracing can be
analyzed as providing a basis for causal inference, to do so is to
minimize key features that make it a powerful tool for evaluating
hypothesized causal mechanisms.
When we think of process tracing through narrative, the role
that case studies play becomes clearer. The case itself includes a
variety of different descriptive elements from which the causal
story is constructed. While an account of process tracing that fo-
cuses on inference draws evidence from the constructed story, it
leaves out the inter-relationships among the partsdthe generative
and temporal elements of the mechanismdand pulls the focus
away from the coherence of the case as a whole that is relevant for
consideration of alternative hypotheses and for counterfactual
15
Thanks to Mary Morgan for providing this turn of phrase.
 S. Crasnow / Studies in History and Philosophy of Science 62 (2017) 6e13
analysis. These features support the evaluation of hypotheses
because they constitute the total evidence that must be made sense
of in the telling of the causal storydthey contribute to the narra-
tive’s completeness, consistency, and credibility.
The narratives through which case studies are presented are
indeed engaging and, as Tannenbaum suggests, it would be a
shame to sacrifice that appeal. But the greater loss is to the
cognitive power of the narrative when cases are treated merely as
sources of bits of diagnostic evidence.
Acknowledgements
I am grateful to Stephan Haggard, Julian Reiss, and an anony-
mous referee for comments on an earlier version of this paper. In
addition, I owe a special thanks to Mary Morgan and Norton Wise
both for including me in this project and their guidance in thinking
about narrative, process tracing, and causal mechanisms.
References
Beach, Derek, & Pedersen, Rasmus Brun (2013). Process-tracing Methods: Founda-
tions and guidelines. Ann Arbor: University of Michigan Press.
Beatty, John (2016). What are narratives good for? Studies in the History and Phi-
losophy of Biological and Biomedical Sciences, 58, 33-40.
Bennett, Andrew (2010). Process tracing and causal inference. In Henry Brady, &
David Collier (Eds.), Rethinking social inquiry: Diverse tools, shared standards (p.
220). Lanham, MD: Rowen & Littlefield Publishers, Inc.. Lanham.
Bennett, Andrew (2015). Appendix: Disciplining our conjectures, systemizing pro-
cess tracing with bayesian analysis. Process Tracing: From Metaphor to Analytic
Tool, 276-298. Cambridge: Cambridge University Press.
Bennett, Andrew, & Checkel, Jeffrey (2014). Process Tracing: From metaphor to an-
alytic tool. Cambridge: Cambridge University Press.
Bogen, Jim (2005). Regularities and causality; generalizations and causal explana-
tions. Studies in the History and PhilosohyPhilosophy of Biological and Biomedical
Science, 36(2), 397-420.
13
Carroll, Noël (2001). Narrative connection. In Noël Carroll beyond Aesthetics: Philo-
sophical essays (pp. 118-133). Cambridge: Cambridge University Press.
Chang, Hasok (2012). Beyond case studies: History as philosophy,” integrating
history and philosophy of science: Problems and prospects. In
Seymour Mauskopf, & Tad M. Schmaltz (Eds.), Boston studies in the history and
philosophy of science (pp. 109-124). Dordrecht: Springer.
Crasnow, Sharon (2012). The role of case study research in political science: Evi-
dence for causal claims. Philosophy of Science, 79(5), 655-666.
George, Alexander L., & Bennett, Andrew (2005). Case studies and theory develop-
ment in the social sciences. Cambridge: MIT Press.
Glennan, Stuart (2002). Rethinking mechanistic explanation. Philosophy of Science,
S342-S353.
Gould, Stephen J., & Lewontin, Richard C. (1979). The spandrels of san marcos and
the panglossian paradigm: A critique of the adaptationist programme. Pro-
ceedings of the Royal Society London B, 205, 581-598, 1979.
Illari, Phyllis McKay, & Williamson, Jon (2012). What is a mechanism? Thinking
about mechanisms across the sciences. European Journal of Philosophy of Science,
2, 119-135.
King, Gary, Keohane, Robert O., & Verba, Sidney (1994). Designing social Inquiry:
Scientific inference in qualitative research. Princeton. NJ: Princeton University
Press.
Layne, Christopher (1994). Kant or Cant: The myth of democratic peace. Interna-
tional Security, 19(2), 5-49.
Mahoney, James (2015). Process tracing and historical explanation. Security Studies,
24(2), 200-218.
Morgan, Mary (2012). Case studies: One observation or many? Justification or
discovery? Philosophy of Science, 79, 667-677.
Reiss, Julian (2008). Error in economics: Towards a more evidence-based methodology.
New York: Routledge.
Reiss, Julian (2015). Causation, evidence, and inference. New York: Routledge.
Roth, Paul (1989). How narratives explain. Social Research, 56(2), 449-478.
Russett, Bruce (1993). Grasping the democratic Peace: Principles for a post-cold war
world. Princeton, NJ: Princeton University Press.
Schultz, Kenneth (2001). Democracy and coercive diplomacy. New York: Cambridge
University Press.
Tabery, James (2004). Synthesizing activities and interactions in the concept of a
mechanism. Philosophy of Science, 71(1), 1-15.
Tannenwald, Nina (2015). Process tracing and security studies. Security Studies, 219-
227.
Van Evera, Stephan (1997). Guide to methods for students of political science. Ithaca:
Cornell University Press.