842 PS • October 2018 © American Political Science Association, 2018 doi:10.1017/S1049096518000975
Process-Tracing Research Designs:
A Practical Guide
Jacob I. Ricks, Singapore Management University
Amy H. Liu, University of Texas at Austin
ABSTRACT Process-tracing has grown in popularity among qualitative researchers. However,
unlike statistical models and estimators—or even other topics in qualitative methods—process-
tracing is largely bereft of guidelines, especially when it comes to teaching. We address this
shortcoming by providing a step-by-step checklist for developing a research design to use
process-tracing as a valid and substantial tool for hypothesis testing. This practical guide
should be of interest for both research application and instructional purposes. An online
appendix containing multiple examples facilitates teaching of the method.
How does one develop a research design based on
process-tracing? This question highlights a major
challenge in teaching and adopting process-tracing
methods. Although there is an expanding body
of work on the approach (Beach and Pedersen
2013; Bennett and Checkel 2015; Humphreys and Jacobs 2015;
Mahoney 2012; Rohlfing 2014), we are still faced with Collier’s
(2011, 823) lamentation: “Too often this tool is neither well
understood nor rigorously applied” (see also Blatter and Blume
2008, 318; Zaks 2017). One central concern is that there are few
instructional materials in the qualitative-methods canon (Elman,
Kapiszewski, and Kirilova 2015; Kapiszewski, MacLean, and Read
2014). This article provides a short, practical guide for develop-
ing a process-tracing research design. The corresponding online
appendix applies this guide to four examples, thereby offering
a tool for researchers seeking to employ and instructors planning
to teach this method.
The material is organized in the form of a checklist
that provides introductory guideposts to help researchers
structure their research designs. This article is not a compre-
hensive literature review (Kay and Baker 2015), and neither
is it the final word on what constitutes good process-tracing
(Waldner 2015). There remains much work to be done in defin-
ing, delineating, and developing process-tracing methods,
and we advise graduate students and advanced researchers to
become familiar with these debates (Beach and Pedersen 2013;
Bennett and Checkel 2015). Instead, our contribution is to
make process-tracing accessible and more readily applicable to
beginners without being distracted by ongoing methodological
The discussion is limited to one type of process-tracing: theory
testing (Beach and Pedersen 2013). Speciﬁcally, we focus on the
systematic study of the link between an outcome of interest and
an explanation based on the rigorous assessing and weighting of
evidence for and against causal inference. By deﬁning process-
tracing in these terms, we emphasize the role of theory and the
empirical testing of hypotheses. The challenge is to assemble a
research design equipped to do so.
To craft a research design based on process-tracing, we suggest
that researchers must (1) define their theoretical expectations,
(2) give direction to their research, and (3) identify the types of
data necessary for testing a theory. Stated differently, the steps
outlined in ﬁgure 1 set the stage for implementing best practices
(Bennett and Checkel 2015). In the online appendix provided to
assist with teaching, we show how this checklist can be applied in
four diﬀerent examples: the rise of the Japanese developmental
state; the electoral success of the Thai Rak Thai party in Thailand;
the standardization of English in Singapore; and the bureaucratic
reforms of the Philippines irrigation agency. We recommend that
instructors start with the checklist before having students read
the appendix; these materials should be paired with Collier (2011).
Alternatively, instructors can present both the checklist and the
appendix simultaneously and then assign students to use the
checklist to evaluate a separate article based on process-tracing
methods (e.g., Fairﬁeld 2013 and Tannenwald 1999). The goal is to
ingrain in students’ minds what process-tracing is and how it can
be used. In the following discussion, we reference the example of
Slater and Wong’s (2013) process-tracing analysis of why strong
authoritarian parties sometimes embrace democratization.
Step 1: Identify Hypotheses
We adopt the maxim “Theory saves us all.” Research designs and
empirical analyses for causal analysis should be theoretically guided.
Jacob I. Ricks is assistant professor of political science at Singapore Management University.
He can be reached at email@example.com.
Amy H. Liu is associate professor in government at University of Texas at Austin.
She can be reached at firstname.lastname@example.org.
PS • October 2018 843
Therefore, establishing testable hypotheses based on our theories
is the first step in good process-tracing. In this sense, building
a research design for process-tracing is the same as in any other
attempt at causal inference. There is, however, one important
distinction. In process-tracing, we are concerned not only with
our theory of interest; we also must juxtapose rival explanations
that we intend to test (Hall 2013; Rohlﬁng 2014; Zaks 2017). It is
important that the concerned hypothesis is evaluated against
alternative(s) in a Lakatosian sense, creating a “three-cornered
ﬁght” that pits our observations against both our primary theory
and at least one alternative (Lakatos 1970).
The checklist is structured to allow for the testing of multiple—
that is, as many as required—rival hypotheses. In an oft-used
comparison, detectives in criminal cases begin their investigation
by focusing on those closest to the victim and then eliminating
suspects (i.e., hypotheses) along the way. Social scientists should
act similarly, remembering Ockham’s razor: seek ﬁrst hypotheses
that are clearly related to the outcome, simple, and testable before
employing more complex explanations. These theoretical expecta-
tions should be plainly established before moving to step 2.
Step 2: Establish Timelines
The second step is to sequence events. Timelines should be
bookended according to the theoretical expectations. The conclusion
of the timeline will be at or shortly after the outcome of interest—that
is, the dependent variable. The challenge is to identify how far back
in time we must go to seek out our cause. A good timeline begins
with the emergence of the theorized causal variable. For instance,
we hypothesize that the compounded eﬀect of antecedent party
strength, ominous signals, and legitimization strategies causes
strong authoritarian parties to embrace democratization (Slater and
Wong 2013). Therefore, we begin our timeline with the foundations
of the vital components of the theory—namely, the antecedent
strength of the party—and end it with the democratic transition.
The timeline has several purposes. First, it clariﬁes the research-
er’s thought process. Second, it establishes temporal precedence.
Third, it provides what can be constituted as a “face-validity” test
for the argument. Fourth, it helps to identify major events that
could have shaped the outcome of interest. In doing so, this allows
us to revisit our hypotheses and to ascertain whether we might be
missing an obvious probable cause for the concerned outcome.
In essence, we give ourselves the opportunity to verify whether
the events in question ﬁt the hypotheses. Analogously, criminal
investigators also use timelines to establish the victims’ histories
and points where they may have met foul play. Although these
timelines rarely ﬁnd their way into published works, they are an
imperative step in the research process. Researchers should keep
their timelines readily available with updates as they progress
through the many stages of ﬁeldwork. Timeline development is a
critical exercise before initiating evidence collection.
Step 3: Construct Causal Graph
After sequencing the timeline, the next step is to construct
a causal graph (Waldner 2015). This type of graph identifies
the independent variable(s) of
interest. It also provides struc-
ture, allowing us to focus on the
link between the explanation
and the concerned outcome. In
other words, a causal graph vis-
ually depicts the causal process
through which X causes Y. With
a causal graph, we can identify all
moments when the concerned
actor (e.g., individual, govern-
ment, party, or group) made a
choice that could have aﬀected
the result. This endogenous
choice need not be contentious,
but it does need to be theoreti-
We depart slightly from
Waldner (2015), however, in
two ways. First, we contend that
just as not all choices are rele-
vant moments, not all relevant
moments are choices. They also
Process-Tracing: The Checklist
In an oft-used comparison, detectives in criminal cases begin their investigation by focusing
on those closest to the victim and then eliminating suspects (i.e., hypotheses) along the way. Social
scientists should act similarly, remembering Ockham’s razor: seek ﬁrst hypotheses that are clearly
related to the outcome, simple, and testable before employing more complex explanations.
844 PS • October 2018
The Teacher: Process-Tracing Research Designs: A Practical Guide
can be exogenous events—that is, critical junctures that emerge
from events such as the discovery of oil or a natural disaster. What
matters is that these moments are “collectively suﬃcient to gen-
erate the outcome” (Waldner 2015, 131). Second, our use of causal
graphs potentially includes events that may not fit clearly into
the causal process being identiﬁed. We distinguish these events
with dashed lines. In contrast, the causal process remains out-
lined with solid lines. This highlights and clariﬁes—especially for
students—that not all interesting events are variables of interest.
Causal Graph of Slater and Wong (2013)
These activities are part of the background work that must be accomplished before engaging
in any type of ﬁeldwork—from visiting archives to conducting interviews, from administering
surveys to observing participants.
For an example, we offer a simple causal graph of Slater and
Wong’s (2013) theory about why strong authoritarian-party
states democratize (figure 2). Slater and Wong began by pre-
senting their scope condition: democratic transitions under
the watch of dominant authoritarian ruling parties. Given
this situation, our theoretical expectation would be a low like-
lihood of democratization. Yet, Slater and Wong (2013, 719)
claimed that “dominant parties can be incentivized to con-
cede democratization from a position of exceptional strength”
under a set of three specific conditions. First, they must enjoy
a high degree of antecedent strengths—that is, confidence that
the party can still dominate post-transition politics. Second,
this strength, however, must have been challenged by ominous
signals that the party is past its authoritarian prime. Third,
leaders must strategically choose to adopt democratic legitima-
Causal graphs follow the initial timeline; they build on the
series of events that are identified in the timeline. In other
words, we can pinpoint the hypothesized explanation and
the outcome in a temporal chain. We can specify where and
which types of empirical information are necessary for the
analysis. The timeline and the causal graph can be developed
together iteratively. Whereas the sequence of events will not
change, the creation of the causal graph might cause us to revisit
the timeline to clarify links or highlight important missing
Step 4: Identify Alternative Choice or Event
At each relevant moment in the causal graph, a diﬀerent choice
could have been made or another event could have happened. For
each distinct moment, we identify these alternative(s). It is impor-
tant, however, that these alternatives are theoretically grounded.
There must be a reason that the choice could have been made or
that the event could have manifested diﬀerently.
Step 5: Identify Counterfactual Outcomes
Next, for each moment, we identify the counterfactual outcome
that would have happened if the alternative choice had been
taken or the alternative event had transpired. Counterfactuals
are vital to process-tracing, especially when no alternative cases
are considered (Fearon 1991). When treating hypothetical predic-
tions, it is imperative that another outcome was possible. If there
is no plausible theory-informed alternative outcome, then no real
choice or event has taken place. Thus, the link between the input
and the outcome was predetermined; hence, process-tracing pro-
vides little value added. Note that steps 4 and 5 are closely linked.
An approach in lieu of counterfactuals is the use of controlled
comparisons, wherein the case of interest is compared with empiri-
cal alternatives rather than a hypothetical counterfactual (Slater
and Ziblatt 2013). However, if a researcher is primarily focused
on one single case—or perhaps multiple cases that are not explic-
itly comparable via the research design—then this counterfactual
exercise is important. Even if a researcher does use controlled
comparisons, we still recommend considering counterfactuals.
Note, however, that counterfactuals are heuristic devices that allow
us to identify hypothesized outcomes and thus potential data
to collect; they are not evidence per se.
It is important that steps 1 through 5 be conducted before data
collection. These activities are part of the background work that
must be accomplished before engaging in any type of ﬁeldwork—
from visiting archives to conducting interviews, from administer-
ing surveys to observing participants. They are essential to the
process of theory testing because they establish expectations about
what researchers should encounter during their data-collection
process. Because process-tracing often is iterative, researchers
likely will revisit these steps throughout the research project—
especially in light of new data. However, an initial plan for data
collection should be designed based on these ﬁve steps.
Step 6: Finding Evidence for Primary Hypothesis
After we have established a timeline, outlined our causal graphs,
and identified our theoretical expectations, we can design the
data-collection portion of our research project. At each iden-
tified relevant moment, we must plan to systematically find
evidence that the variable germane to the primary hypothesis
was the reason the concerned actor pursued the timeline path.
It is important that as we design our data collection, we must
recognize that not all evidence types are the same (Bennett
2014; Collier 2011; Mahoney 2012; Rohlﬁng 2014). Some data are
necessary to establish causation; others suﬃcient—and then there
are data that are neither or both. We suggest drawing on Van
Evera’s (1997) four types of evidence, summarized in table 1:
straw-in-the-wind, hoops, smoking gun, and doubly decisive.
Due to space constraints, we do not explain these evidence types
in detail (see Collier 2011 for an extensive discussion). Figure 1
utilizes these evidence types and the appendix demonstrates
PS • October 2018 845
Process-tracing involves rigor and attention to details and logic of causal inference similar to
that of a detective or a medical examiner.
When creating a data-collection plan, it is common for
researchers—especially those who spend time in the field—to
accumulate data in a “soak-and-poke” fashion. We do not con-
demn such efforts; however, we encourage researchers to think
carefully about the evidence types they are collecting because
most information gathered will be of the straw-in-the-wind
type. Stated differently, whereas much data gathered may offer
weak support for—or at least not negate—the primary hypothesis,
it is not the most useful for testing purposes. When designing
research, it is absolutely vital to remain cognizant of the evidence
type collected and its ability to support or negate the larger claims
(Fairfield 2013). The causal graph is particularly useful at this
point because it identifies the links that must be made between
Types of Evidence for Process-Tracing
Sucient for Arming Causal Inference
Necessary for Arming
1. Straw-in-the-Wind 3. Smoking Gun
a. Passing: Arms relevance of hypothesis but does not conrm it. a. Passing: Conrms hypothesis.
b. Failing: Hypothesis is not eliminated but is slightly weakened. b. Failing: Hypothesis is not eliminated but is
c. Implications for rival hypothesis:
Passing slightly weakens them.
Failing slightly strengthens them.
c. Implications for rival hypothesis:
Passing substantially weakens them.
Failing somewhat strengthens them.
2. Hoops 4. Doubly Decisive
a. Passing: Arms relevance of hypothesis but does not conrm it. a. Passing: Conrms hypothesis and eliminates
b. Failing: Eliminates hypothesis. b. Failing: Eliminates hypothesis.
c. Implications for rival hypothesis:
Passing somewhat weakens them.
Failing somewhat strengthens them.
c. Implications for rival hypothesis:
Passing eliminates them.
Failing substantially strengthens them.
Source: Collier (2011, 825)
interviews—for example, with military advisers from the authori-
tarian period who relayed growing disloyalty among the armed
forces and recommended the leadership to concede. It also can
be ascertained from archival documents—for example, minutes
from cabinet meetings discussing diﬀerent electoral rules for the
party to adopt on transition. Conversely, evidence describing the
personalities active in alternative rival parties might be consid-
ered straw-in-the-wind. Although interesting, these data are not
vital to establishing the strength of the theory; more important is
information on the level of threat they posed to the ruling party.
When we design data collection, we must be careful to focus on the
evidence types that matter lest we be left building our evidentiary
house with a pile of straw.
our variables of interest to establish causation. For instance, certain
evidence types simultaneously can support our proposed theory
and eliminate a rival one. Van Evera (1997) called this doubly-
decisive evidence. If such a datum is found, then we can exclude
all other hypotheses and step 6 becomes the final one in
our process-tracing efforts. Unfortunately, these cases are rare.
Therefore, we must increase our evidence pool to demonstrate
that our hypothesis is the best fit from a set of possible expla-
nations. This is outlined in step 7.
For step 6, we exhort researchers to make clear their expec-
tations about the evidence types needed to (1) further support
their argument, and (2) negate the rival hypotheses. For instance,
consider Slater and Wong’s (2013) assertion that democratiza-
tion can emerge from strategic decisions by a ruling party. Here,
we want smoking-gun evidence that links antecedent strength,
ominous signals, and legitimation strategies directly to the
decision to democratize. This type of evidence can be found in
Step 7: Find Evidence for Rival Hypothesis
Our final step is to repeat step 6; at each choice node, the focus
now should be on alternative explanations. This step may
require multiple iterations depending on the number of rival
hypotheses. The objective is to dismiss as many explanations as
possible, leaving only one hypothesis as the most likely. Here,
the most important evidence type is the exclusionary or—per
Van Evera (1997)—the hoops test. Hoops evidence, if absent, can
eliminate a hypothesis from consideration. If the hypothesized
variable was not present when the event happened, then we can
dismiss the rival hypothesis.
If the rival explanation is not easily discarded, we must move
on to other data types. Wherever possible, we look for opportuni-
ties to dismiss the hypothesis. However, if at some point we ﬁnd
evidence to the contrary, we cannot reject it. Instead, we must
consider that a rival hypothesis could explain the phenomenon of
interest better than the primary one.
846 PS • October 2018
The Teacher: Process-Tracing Research Designs: A Practical Guide
Because political phenomena are complex, it is possible
that the different explanations may not be mutually exclusive
(Zaks 2017). Therefore, pitting competing hypotheses against
one another can result in instances in which multiple hypotheses
all seem to have explanatory leverage. When these conditions
manifest, we must rely on a deep understanding of our cases to
weigh the evidence and judge which hypothesis best explains the
outcome. As in a criminal investigation, we must discern which
theory of the crime has the strongest evidence and proceed as
best we can to trial.
Despite the popularity of process-tracing in empirical research,
discussions on how to develop effective research designs based
on the method are largely absent in political science—especially
when we consider teaching materials. Frequently, there is a
disjuncture between theoretically driven research designs and
rigorously evaluated empirics. Beyond this, to those who do
not regularly engage in process-tracing, the method can be poorly
understood. The prime advocates of process-tracing continue
to make strides in pushing methodological understanding and
boundaries. This work, however, does not necessarily lend itself
to introducing the tool to the uninitiated. As a result, critics have
dismissed process-tracing as being ineﬀective in explaining polit-
ical phenomena beyond a singular case—if even that. We under-
stand but do not agree with these positions.
Process-tracing involves rigor and attention to details and
logic of causal inference similar to that of a detective or a medical
examiner. It requires establishing a sequence of events and iden-
tifying a suspect pool. With each piece of evidence, we can elimi-
nate a variable and/or strengthen one hypothesis against another.
We conduct this iterative process until we are ready for trial.
In this spirit, we oﬀer our checklist to help researchers develop
a causal research design and then evaluate pieces of evidence sys-
tematically against it. Such practical guidance is largely missing
in the process-tracing literature. This guide and the applications
in the online appendix attempt to address this shortcoming and
to demonstrate how process-tracing can be done rigorously.
We challenge advocates to adopt these standards in their own work
and skeptics to conceptualize process-tracing as more than glori-
ﬁed storytelling. We also hope that the method can be integrated
easily and clearly as a component of political science courses—not
only in methods classes but also in substantive courses. Indeed,
through careful application, process-tracing can serve as a strong
tool for hypothesis testing.
To view supplementary material for this article, please visit
We thank Marissa Brookes, Jason Brownlee, José Cheibub,
Travis Curtice, Jennifer Cyr, John Donaldson, Richard Doner,
Zach Elkins, Michael Giles, Anna Gunderson, Nicholas Harrigan,
Abigail Heller, Allen Hicken, Laura Huber, Kendra Koivu, James
Mahoney, Eddy Malesky, Joel Moore, Ijlal Naqvi, Sari Niedzwiecki,
Rachel Schoner, Dan Slater, Hillel Soifer, Kurt Weyland, and
the anonymous reviewers for helpful comments on this article.
An earlier version was presented at the 2016 Southwest Mixed
Methods Research Workshop at University of Arizona. We were
able to convert this project from an idea to an article under the
auspices of the Short-Term Research Collaboration Program at
Singapore Management University. Any errors belong to the
Beach, Derek, and Rasmus Brun Pedersen. 2013. Process-Tracing Methods. Ann Arbor:
University of Michigan Press.
Bennett, Andrew. 2014. “Process-Tracing with Bayes.” Qualitative and Multi-Method
Research Spring: 46–51.
Bennett, Andrew, and Jeﬀrey T. Checkel. 2015. “Process-Tracing: From Philosophical
Roots to Best Practices.” In Process-Tracing, ed. Andrew Bennett and Jeﬀrey T.
Checkel, 3–37. New York: Cambridge University Press.
Blatter, Joachim, and Till Blume. 2008. “In Search of Co-Variance, Causal Mechanisms
or Congruence? Towards a Plural Understanding of Case Studies.” Swiss Political
Science Review 14 (2): 315–56.
Collier, David. 2011. “Understanding Process-Tracing.” PS: Political Science & Politics
44 (4): 823–30.
Elman, Colin, Diana Kapiszewski, and Dessislava Kirilova. 2015. “Learning
through Research: Using Data to Train Undergraduates in Qualitative Methods.”
PS: Political Science & Politics 48 (1): 39–43.
Fairﬁeld, Tasha. 2013. “Going Where the Money Is: Strategies for Taxing Economic
Elites in Unequal Democracies.” World Development 47: 42–57.
Fearon, James D. 1991. “Counterfactuals and Hypothesis Testing in Political Science.”
World Politics 43 (2): 169–95.
Hall, Peter A. 2013. “Tracing the Progress of Process-Tracing.” European Political
Science 12 (1): 20–30.
Humphreys, Macartan, and Alan M. Jacobs. 2015. “Mixing Methods: A Bayesian
Approach.” American Political Science Review 109 (4): 653–73.
Kapiszewski, Diana, Lauren MacLean, and Benjamin Read. 2014. Field Research
in Political Science. New York: Cambridge University Press.
Kay, Adrian, and Phillip Baker. 2015. “What Can Causal Process-Tracing Oﬀer to
Policy Studies? A Review of the Literature.” Policy Studies Journal 43 (1): 1–21.
Lakatos, Imre. 1970. “Falsification and the Methodology of Scientific Research
Programmes.” In Criticism and the Growth of Knowledge, ed. Imre Lakatos and
Alan Musgrave, 91–196. New York: Cambridge University Press.
Mahoney, James. 2012. “The Logic of Process-Tracing Tests in the Social Sciences.”
Sociological Methods & Research 41 (4): 570–97.
Rohlfing, Ingo. 2014. “Comparative Hypothesis Testing Via Process-Tracing.”
Sociological Methods & Research 43 (4): 606–42.
Slater, Dan, and Joseph Wong. 2013. “The Strength to Concede: Ruling Parties and
Democratization in Developmental Asia.” Perspectives on Politics 11 (3): 717–33.
Slater, Dan, and Daniel Ziblatt. 2013. “The Enduring Indispensability of the Controlled
Comparison.” Comparative Political Studies 46 (10): 1301–27.
Tannenwald, Nina. 1999. “The Nuclear Taboo: The United States and the Normative
Basis of Nuclear Non-Use.” International Organization 53 (3): 433–68.
Van Evera, Stephen. 1997. Guide to Methods for Students of Political Science. Ithaca, NY:
Cornell University Press.
Waldner, David. 2015. “What Makes Process-Tracing Good? Causal Mechanisms,
Causal Inference, and the Completeness Standard in Comparative Politics.”
In Process-Tracing, ed. Andrew Bennett and Jeﬀrey T. Checkel, 126–52. New York:
Cambridge University Press.
Zaks, Sherry. 2017. “Relationships Among Rivals (RAR): A Framework for
Analyzing Contending Hypotheses in Process-Tracing.” Political Analysis
25 (3): 344–62.