BSSn4495:
Qualitative
research in
security
studies
Strategies for causal
identification:
experiments and QCA
March 21, 2024
Miriam Matejova, PhD
Agenda
• Why experiments/QCA?
• When should/can we use
experiments/QCA?
• What are the
advantages/disadvantages of
the experimental/QCA
method?
Spurious correlation
We may observe a covariation (correlation) between C
and E.
BUT, this may be because C is NOT a cause of E, but
because Z is a cause of BOTH C and E.
C E
Z
Correlation
“Assignment” of Causes
• Causal claim: Attending University (X) Causes
Higher Future Earnings (Y).
• Each case represents an individual (a potential
student)
• How is Attending University (X) “assigned”
across cases in the real world? X is typically
chosen by individuals on the basis of some Z (e.g.
ambition).
Case 1 Case 2 Case 3 Case 4 Case 5
X
Low Z High ZHigh Z Low Z High Z
Here, only cases with
High Z get X=>
Spurious Correlation
between X and Y
“Assignment” of Causes
• But what if we let researchers assign X across
cases in such a way that it does not depend on
Z
Case 1 Case 2 Case 3 Case 4 Case 5
X
Low Z High ZHigh Z Low Z High Z
Case 6
Low Z
Here Cases with High and Low Z are
equally likely to be assigned X => No
Spurious Correlation between X
and Y
“Random Assignment”
“Random assignment” is a procedure for assigning
X to cases that ensures that the difference in the
value of the Zs between the cases that are assigned
X and the cases that are not assigned X disappears
as the number of cases gets large (law of large
numbers)
Case 1 Case 2 Case 3 Case 4 Case 5
X
Low Z High ZHigh Z Low Z High Z
Case 6
Low ZThis procedure works even if the researcher does not know
what the Z variables are or cannot measure them
Limitations of “random assignment” in
social sciences
• Cost and ethics
• Artificial intervention by the researcher vs. real
world applicability
– The problem of generalization
• Cannot study the effects of things that have already
happened
• Can get biased result if inappropriately designed
Qualitative comparative analysis (QCA)
• A set-theoretic method
• QCA as an approach
and a data analysis
technique
Set-theoretic methods
• The data consist of set membership scores
– crisp, fuzzy, multi-value
• Relations between social phenomena
modeled in terms of set relations
– necessity, sufficiency, etc.
• The focus is on causal complexity
– equifinality, conjunctural causation, etc.
When do we use QCA?
• Causal complexity
– Multifinality: same factor, different outcomes
– Equifinality: different factors, same outcome
– Asymmetric causality:
• presence and absence of outcome have different
explanations
– economic growth → democratization
– clientelism → non-democratization
• Presence and absence of condition produce different
outcomes
• Mid-sized N
Sets: necessary and sufficient
conditions
What are sets?
• Establish qualitative, not quantitative,
differences between cases
– height  not a set
– tall person  set
Sets vs. variables
Schneider 2017
Types of sets: crisp set
• Dichotomous sets
• Full member (1) vs. full non-member (0)
– Establishes qualitative, not quantitative,
differences between cases
– E.g., set of big countries
• China, Russia (1) vs. Hungary, Lichtenstein (0)
Types of sets: fuzzy sets
• Allow for degree of membership in set
• Partial membership in sets
– Any value between 0 and 1
– Three qualitative anchors (0, 0.5, 1)
– Qualitative and quantitative differences
• NOT probabilities
Schneider 2017
QCA: Steps
1) Assemble the universe of cases
2) Collect raw data
3) Calibrate conditions sets and outcome sets
4) Search for necessary conditions
5) Represent empirical evidence in a truth table
6) Identify sufficient conditions by logically
minimizing the truth table
7) Do within-case analyses in typical and deviant
cases
QCA challenges
• Location of qualitative anchors
• Sometimes false impression of precision
• Resources, time, data availability