Case selection
Lukas Lehotsky & Petr Ocelik
Should offer more
insight than cross-case
Case selection
Cross-case
• Sample selection
• Random
• Based on knowledge
about population
• Sampling and analysis
sequential tasks
• Large sample = easy
elaboration of external
validity
• Internal validity more
concerning
Case study
• Specific case selection
prone to bias (cherry-
picking)
• Sampling and analysis not
separate from each other
• “Casing” – what is this
case a case of?
• External validity more
concerning
Case selection
• Does case study provide any advantage?
• What is the population? What is this case a case of?
• Am I interested in particular case? Why? What is it
that I want to study?
• Do I want to test or build theory?
Normal (typical) case
• Represents typical relation between variables
• Least residuals – closest to the linear model
prediction – onlier
• Representative by nature
• Theory testing
Diverse cases
• Two or more cases
• Full scope of variation of the variable/variables
• Standard-deviation
• Theory testing/building
• Represents full spectrum of variation (but not
distribution)
Extreme case
• One or more cases
• Extreme value of variable/variables
• Many standard deviations from the mean
• Theory building
• Representative in comparison with larger
population
Deviant case
• One or more cases
• Outlier – the most distant case from the linear
model prediction – highest residual
• Theory building, modification
• Unrepresentative by nature
Influential case
• One or more cases
• Similar to deviant case – seemingly disproves
theory and has high leverage – influences the
model
• Special configuration of variables – exception which
(if explained) strengthens theory
• Theory testing, modification
• Representativeness irrelevant
Crucial case
• One or more cases
• Case which must fit the theory – proves validity of
theory
• Least-likely
• Predicted to contradict the hypothesis, but does not
• Confirms the theory
• Most-likely
• Predicted to support the hypothesis, but does not
• Disconfirms the theory
• Most-difficult test for an argument
• Theory testing
• Selecting on theoretical grounds
Pathway case
• One or more cases
• Case, where value of dependent variable is most
likely caused by independent variable – apparent
impact of indep. X on dependent Y
• Cross-tabulation or analysis of residuals (residuals
of full model vs. model without causal variable)
• Theory testing – especially causal mechanisms
Most similar cases
• Two or more cases
• Comparative design
• All variables except those of interest are
similar/same
• Selection from cross-case based on values of
variables
• Theory testing and theory building
Case V1 V2 V3 V4 V5
Case 1 0 + + + Case
2 + + + - Most
similar cases
Most different cases
• Two or more cases
• Comparative design
• All variables except those of interest are different
• Selection from cross-case based on values of
variables
• Theory testing and theory building
Most diverse cases
Case V1 V2 V3 V4 V5
Case 1 + - 0 - +
Case 2 + + + - -
Comparability
• Cases must be comparable for a comparative CS to
be valid.
• Achieving comparability
• Synchronical/spatial - more units in same time period
• Diachronical - same unit in more time periods
Overview
Case type Theory # Case selection
Test Build
Normal + 1 Min. ε of linear regression Y=aX+bZ+ε
Diverse + + 2 Max. range of variance of X, Y, X/Y, …
Extreme + 1 Max. variance of Y
Deviant + 1 Max. ε of linear regression Y=aX+bZ+ε
Crucial + 1 ML predicts H1, proves to be H0
LL predicts H0, proves to be H1
Influential + 1 Max. influence of a case on slope a of
Y=aX+bZ+ε
Pathway + 1 X1 and not X2 likely to have caused Y
Most similar + + 2 Cases similar on var. other than X1/Y
Most different + + 2 Cases different on var. other than X1/Y
SelectionResearch goalFocus
Case study
Theory-centered
Building
hypothesis/theory
Distribution-based
(Div., Ext., Dev.,
MS, LS)
Testing
hypothesis/theory
Distribution-based
(Norm., Div., Infl.,
Path., MS, LS)
Theory-based
(Crucial ML,
Crucial LL)
Modifying
hypothesis/theory
Distribution-based
(Deviant, MS, LS)
Theory-based
(Crucial ML,
Crucial LL)
Case-centered