Avoiding Common Errors in QCA: A Short

Guide for New Practitioners

Claude Rubinson
University of Houston–Downtown
Lasse Gerrits
Erasmus University Rotterdam
Roel Rutten Thomas Greckhamer
Tilburg University Louisiana State University
November 27, 2024

QCA is increasingly being adopted by researchers from a variety of disciplines

and industries. As is only natural when learning a new methodological approach,
not all of these applications are free of errors. In this paper, we provide a list
of common errors that we have seen when reviewing QCA projects. We provide
this list as a “cheat sheet” in the hopes of helping researchers improve their
projects and avoid some of the most common mistakes encountered in the wild.
For each item we briefly explain what the error is and why it is considered a
problem. We also describe possible remedies whenever applicable. The order of
items follows the logic of the research cycle. These discussions are by no means
comprehensive and we strongly recommend that the reader review the following
texts for more information:

ˆ Ragin, C.C. (1987). The Comparative Method

ˆ Ragin, C.C. (2008). Redesigning Social Inquiry

ˆ Oana, I., C.Q. Schneider and E. Thomann (2021). Qualitative Compara-

tive Analysis with R: A Beginner’s Guide

ˆ Mello, P.A. (2022). Qualitative Comparative Analysis: An Introduction to

Research Design and Application
ˆ Rutten, R. (2024). Qualitative Comparative Analysis: Learning from
Cases

Also, for discussions of recommendations for standards of good practice in QCA

research, readers are referred to the following articles:

1
ˆ Schneider, C. Q., and Wagemann, C. (2010). Standards of good practice
in qualitative comparative analysis (QCA) and fuzzy-sets. Comparative
Sociology, 9, 397–418.
ˆ Greckhamer, T., Furnari, S., Fiss, P., and Aguilera, R. (2018). Studying
configurations with qualitative comparative analysis: Best practices in
strategy and organization research. Strategic Organization, 16, 482–495.
ˆ Rubinson, C. (2019). Presenting qualitative comparative analysis: No-
tation, tabular layout, and visualization. Methodological Innovations,
May–August, 1–22, DOI:10.1177/2059799119862110.

A List of Common Errors in QCA

Approaching QCA
1. Substituting QCA for small-N statistical analysis
An important advantage of QCA is that it can be used to analyze small-
and medium-N data (as well as large-N). However, this does not mean
that it is always appropriate for small- and medium-N analysis. Keep
in mind that conventional statistical analysis and QCA are designed to
answer different types of research questions.
2. Using variable-oriented language
Causality in QCA is expressed as conditions being sufficient or necessary
for the outcome, not in terms of independent variables having an effect
on a dependent variable. That is, QCA focuses on the causes of effects,
rather than the effects of causes. This is particularly important when
verbalizing the sufficiency of a conjunction of multiple conditions. For
example, from a configurational perspective, it is not that having friends
and having a job have independent, positive effects on happiness. Rather,
it is the combination of having both friends and a job that is sufficient for
being happy. (Note also, that singular tense is used here because there is
a single combination, comprised of multiple conditions.)
Whereas a variable reports the magnitude of an empirical phenomenon, a
condition is a constituitive component of the case(s) under consideration.
Conditions do not report magnitudes but qualitative states. In QCA, cases
have (degrees of) membership in conditions—both explanatory conditions
and the outcome condition.
3. Asserting causation without identifying mechanisms
QCA is a descriptive, not inferential, technique. If you wish to estab-
lish causation, you need to identify the underlying mechanism(s) at work,
drawing upon substantive and/or theoretical knowledge. (This is why
QCA is frequently applied in conjunction with in-depth case studies.)

2
 Setting up research
4. Including only cases for which the outcome is present
QCA is a method for studying diversity. This means that your data set
must include observations where the outcome is absent. It also means that
your truth table must include rows where the outcome is False (i.e., consis-
tency is less than your specified threshold). If all of your (non-remainder)
truth table rows are consistent with the presence of the outcome, recon-
sider your calibration strategy. It may be that you have miscalibrated
one or more conditions, or that you need to recalibrate to a more spe-
cific target set (e.g., from “moderate civic engagement” to “strong civic
engagement”).
If your data set only includes observations where the outcome is present,
you may instead employ analytic induction (AI), as described in Ragin
(2023) Analytic Induction for Social Research, which is closely related to
QCA and available open access.
5. Using nouns to name conditions instead of adjective phrases
Nouns refer to variables; a condition must have an adjective attached. For
example, “GDP” is a variable while “Developed country” is a condition.
“Income” is a variable; “Very rich individual” is a condition. “Years of
education” is a variable; “Highly educated person” is a condition. In
each of these examples, only the adjective phrase refers to a set in which
observations may have membership. Clarifying the adjective phrase for
each of your conditions will help with calibration.

Calibration
6. Using symmetric calibrations
It is important to remember that calibration is asymmetric. That is: the
negation of “rich” is not “poor” but, rather, “not rich.” When calibrating
your conditions, you need to carefully consider what is meant by “fully
in” (fuzzy score=1.0) and “fully out” (fuzzy score=0.0). “Fully out” is
only the opposite of “fully in” in the case of true dichotomies, which is
rare (e.g., “biologically male” is equal to “not biologically female” if and
only if we ignore intersex conditions).
Other examples:
(a) the negation of “developed country” is not “under-developed coun-
try” but, rather, “not-developed country”
(b) the negation of “large company” is not “small company” but, rather,
“not-large company”
(c) the negation of “happy family” is not “sad family” but, rather, “not-
happy family”

3
 7. Calibrating to 0.5
When an observation is assigned a fuzzy-set score of 0.5 for a given condi-
tion, that observation will have equal membership in all truth table rows
but will not have maximum membership in any truth table row. It is
unclear what this score means substantively because the crossover point
is the point of maximum ambiguity, where the observation is neither in
nor out of the target set. The case will appear to drop out of the truth
table analysis, as it doesn’t belong to any corner of the vector space. If
you have many observations scoring at (or close to) 0.5, it often means
that there is something wrong with your calibration strategy. (This may
be a consequence of using symmetric calibrations or conceiving of your
conditions as variables rather than sets, discussed above).
8. Failing to explain calibrations
It is crucial that you explain what your calibrations mean substantively.
It is not sufficient to simply report the values that correspond to, e.g., 0.0,
0.5, and 1.0. You also need to explain why, e.g., 14+ years of education
corresponds to “fully in the set of educated people.” Especially when space
is limited, we recommend providing a table that describes each condition,
the calibration rules and this justification.
9. Mechanistic calibration
Successful calibration requires one to carefully reflect upon the nature
of one’s measures and their meaning. Automated and statistical clus-
tering/rescaling techniques rarely produce useful calibrations. Also avoid
simply adopting another researcher’s calibrations without first considering
how they may need to be adjusted for your particular project.
10. Using a measure of central tendency and/or the variable’s distribution to
calibrate
In general, you should avoid using the variable’s distribution as the basis
for your calibration. Having a higher-than-average income does not mean
that one is rich. Instead, you need to rely upon substantive and theoretical
knowledge of your domain in order to develop meaningful calibrations. If
substantive and theoretical knowledge does not exist, you need to get to
know your cases better. If you nevertheless use the variable’s distribution
as a basis for your calibration, make this clear when naming the condition:
“Above-average family size” or “Below-average family size.”

4
11. Using the full range of a Likert-type scale or index to calibrate
This is not always an error but often is. When calibrating a Likert-type
scale or index, you should not automatically assign the bottom value to
0.0, the middle value to 0.5, and the maximum value to 1.0. Instead, you
should carefully consider the meaning of the scale. In particular, consider
whether some values may need to be collapsed together. For example,
with a 7-point scale, it may be the case that scores 1-3 are “fully out” of
the target set, 4 is “more out than in,” 5 is “more in than out” and 6-7
are “fully in.” The precise calibration depends upon the what the scale
means.
You also should not automatically use the same calibration strategy for
all conditions. Just because you have multiple 7-point measures does not
mean that you should calibrate them all in the same way. Instead, think
about what each item means substantively.

Analysis and interpretation

12. Using a sufficiency consistency threshold < 0.80
In Redesigning Social Inquiry, Ragin states that the consistency threshold
for sufficiency testing should not be less than 0.75 and recommends a
threshold of at least 0.85, especially for macro-level data. Most QCA
reviewers today anticipate a sufficiency consistency threshold of at least
0.8, but this is only a rule of thumb. Lower thresholds may be acceptable
when accompanied by a substantive and/or theoretical justification.
13. Ignoring very low unique coverage scores
Very low unique coverage scores indicate substantial overlap among your
sufficiency recipes. This often indicates that you really have a single recipe
with slight variants due to substitutable conditions. To understand the
nature of the overlap, begin by identifying the observations that belong
to the overlapping recipes.
14. Not running a separate QCA for the negation of the outcome
If you wish to explain both the presence and the absence of the outcome,
as is common in QCA, you need to conduct a separate analysis for each.
Recall that QCA is asymmetric and the combination of conditions that
explains the presence of the outcome may be different from the one that
explains its absence.
Do not be surprised, however, if analyzing the absence of the outcome
yields indeterminant results and/or low consistency scores. Cases lacking
the outcome are often quite heterogenous. In such instances, it is ap-
propriate to focus on explaining the presence of the outcome and simply
acknowledge that understanding the absence of the outcome will require
further research.

5
15. Neglecting to conduct a necessity test
QCA is not only about sufficiency. It is important to also test for necessity,
unless you have a theoretical or methodological justification for not doing
so.
16. Analyzing conditions that are always present
If a condition is always present, for all observations in your data set,
it has no explanatory power because it is necessary for the presence of
the outcome and also for the absence of the outcome. There are two
ways of thinking about such a condition. It may be a “trivial necessary
condition” (e.g., oxygen is necessary for both war and peace) or it may be a
scope condition that describes your sample. Alternatively, it could simply
indicate that you have miscalibrated and need to revise your calibration
of that condition.
17. Ignoring low solution coverage scores
A low solution coverage indicates that there are many instances of the
outcome that are not explained by your model. Whether that is prob-
lematic needs to be discussed. It may be that there are good theoretical
reasons for this, that your model is poor or that your model simply does
not explain everything and that further research is needed. A low coverage
score is not necessarily an indication of poor research design but should
not be ignored and needs to be explained.

18. Ignoring equifinality because coverage is low

One of QCA’s strengths is that it can identify when there are multiple
pathways to the same outcome (equifinality). However, researchers some-
times ignore pathways with low coverage. Just because a pathway only
explains a relatively small number of cases doesn’t mean that it is unim-
portant. For example, just because most instances of lung cancer are due
to smoking does not mean that we should not study other causes.
19. Failing to interpret the results
Do not simply conclude your project by presenting your Boolean solu-
tion(s) and their consistency/coverage scores. Rather, you must explain
what your results mean. A statement of necessity or sufficiency verbalizes
a cross-case regularity that must then be interpreted into a causal mecha-
nism. For example, if having both friends and a job is sufficient for being
happy, this indicates that having a social network to satisfy your emotional
needs and having an income to satisfy your material needs produces hap-
piness. Return to your cases to assess the substantive meaning of your
empirically observed cross-case regularity. Remember that in QCA causa-
tion is established through substantive knowledge, not empirical metrics.

6
 Author biographies
Claude Rubinson (corresponding author; [email protected]) is direc-
tor of COMPASSS, the international, interuniversity consortium of QCA
and CCM practitioners, co-organizer of the Annual QCA Conference of
the Americas and Associate Professor of Sociology at the University of
Houston–Downtown (Houston, Texas, USA) where he teaches courses on
research methods, social inequality, and culture. His research program
has three tracts: methodological research on developing formal methods
of qualitative research, focusing on QCA; sociological research on the re-
lationship between the global political-economy and aesthetic form; and
health services research on the conditions that facilitate/hinder the success
of interventions designed to improve patient outcomes.
Prof. Dr. Dr. Lasse Gerrits ([email protected]) is Professor in the Gover-
nance of Complex Urban Transformations and academic director at the
Institute for Housing and Urban Development Studies at Erasmus Univer-
sity Rotterdam. His research focuses on the nexus of technology, policy
and politics. He is interested in methodological issues with regard to so-
cial complexity and the ways in which this complexity can be unboxed.
He has published extensively on this topic, among others in relationship
to qualitative comparative analysis, critical realism, social complexity and
qualitative methods.
Dr. Roel Rutten ([email protected]) researches the geography of kn-
owledge creation and the organization of knowledge creation. He works at
the School of Social and Behavioural Sciences of Tilburg University (the
Netherlands) and at European Regional Affairs Consultants (ERAC).
Thomas Greckhamer ([email protected]) is the William W. Rucks IV En-
dowed Chair and Professor of Management in the E.J. Ourso College of
Business at Louisiana State University. His research interests are at the
intersection of organization studies, strategic management, and research
methods, focusing on empirical applications of as well as theoretical and
methodological contributions to set-theoretic and qualitative approaches.