SSR III
Critical thinking about causality
Correlation does not imply causation
Causal connection: A relationship between two events or variables in
which one directly influences or brings about a change in the other
- We think it is causality because event B happened after event A:
Thinking hack
o A causes B if and only if: (Mill causality
criteria)
Priority: Change A precedes
change B
Consistency: Change A varies systematically with
change B
Post hoc ergo propter hoc: “it happened after
this, therefore because of this”
o There is thus a priority relationship, but
consistency doesn’t hold
Exclusivity: There is no alternative explanation for the
relationship
Error: Inversion of cause
and effect
o Exclusivity requires a clear causal direction
o This error is called: Mistaking correlation for causation
INUS conditions: An INUS condition is a part of a combination of
conditions that, together, are enough to produce an effect, but alone it
cannot cause the effect
1. Insufficient: The condition by itself isn't enough to cause the effect
2. Necessary: It is a necessary component
within a larger set of conditions that
together cause the effect
3. Unnecessary: The overall set of
conditions isn’t the only way to bring
about the effect
4. Sufficient: However, that set of conditions, when present, is enough
to cause the effect
Often there are many factors required for an effect to occur, but we rarely
know them all
Counterfactual is used to analyze causation:
- Counterfactual: Knowledge of what would happen to each
participant if they had not undergone a certain manipulation
- If we compare that knowledge with what happened, we know
what the effect of the manipulation is
o Perfect counterfactual doesn’t exist:
If it doesn’t exist, what CAN we do:
, - Randomized controlled trials: This ensures the conditions have
similar variability and thus that the control condition is a
representation of counterfactuals
- Different conditions: Using a control group or another condition to
show how the groups differ with different (or lack of) manipulation
- Matching methods: Finding a close ‘match’ (individuals that share
a lot of characteristics) to see the difference between them in
different conditions
Causal reasoning threats: (Campbell’s threats to valid causal
inference)
Ruling our alternative explanations, so whether effect of treatment is
confused with…
1. Outside factors: (confound variables)
a. History: Influences (outside of intervention) over the course of
the research, which influence outcome
b. Maturation: Natural changes that may be confused with
effect treatment
2. Effects of selection: (selection bias)
a. Selection: Selection criteria for treatment related to outcome
of treatment
b. Attrition: Participants’ dropout, systematically correlated with
conditions
i. When attrition (loss of participants) is high or non-
random, it can change the composition of the study
groups
3. Unintended effects of study itself: (Hawthorne effect, placebo
effect)
a. Instrumentation: Change in measuring instrument resulting
in a difference between pre- and post-measurement
b. Testing: Effect of measurement itself on subsequent
measurement (fatigue, habituation, placebo etc)
4. Statistical artifacts:
i. Patterns in data that arise due to limitations of
measurement
b. Regression to the mean: Extreme scores will be followed by
less extreme scores
i. Regression to the mean can create an illusion of
causation by making it look like an intervention brought
about improvement (or deterioration), when participants
are simply returning closer to their average baseline
Causal diagram:
Cutting the ties of the confound factors
- Randomize control is the gold standard
, Beyond correlation does not imply causation:
1. Ignoring causality
a. Some authors write down only the correlations they find,
without making any statement about causality
b. These statements makes it difficult to think of the use for the
findings
2. Statements of causality, but unclear assumptions
a. Other researchers make statements about causal relationships
based on correlational data, but often without specifying
assumptions
i. Conceptual: which third variables should be included in
the model and why not?
ii. Confounds are not mentioned
b. Misspecification of this relationship may leave residual
confounding
i. Refers to the risk of incorrectly modeling or
understanding the relationship between the variables
in a study, which can result in residual confounding—
a situation where confounding effects still influence the
results, even after attempting to control for them
3. Pseudo-correlational statements
a. No direct statements about causality, but clearly the implied in
the conclusion
i. Highlight form Y1 Psych: ‘The role of attachment style on
relationship satisfaction’
Causal diagram: (adding a third variable Z)
1. Z is a mediator
a. Only control (keeping a variable equal across conditions)
when wanting to know the direct effect
2. Z is a collider (common effect)
a. Collider bias: We assume there is a connection between the
causal variables (when studying the variables where the
collider is present)
Critical thinking about causality
Correlation does not imply causation
Causal connection: A relationship between two events or variables in
which one directly influences or brings about a change in the other
- We think it is causality because event B happened after event A:
Thinking hack
o A causes B if and only if: (Mill causality
criteria)
Priority: Change A precedes
change B
Consistency: Change A varies systematically with
change B
Post hoc ergo propter hoc: “it happened after
this, therefore because of this”
o There is thus a priority relationship, but
consistency doesn’t hold
Exclusivity: There is no alternative explanation for the
relationship
Error: Inversion of cause
and effect
o Exclusivity requires a clear causal direction
o This error is called: Mistaking correlation for causation
INUS conditions: An INUS condition is a part of a combination of
conditions that, together, are enough to produce an effect, but alone it
cannot cause the effect
1. Insufficient: The condition by itself isn't enough to cause the effect
2. Necessary: It is a necessary component
within a larger set of conditions that
together cause the effect
3. Unnecessary: The overall set of
conditions isn’t the only way to bring
about the effect
4. Sufficient: However, that set of conditions, when present, is enough
to cause the effect
Often there are many factors required for an effect to occur, but we rarely
know them all
Counterfactual is used to analyze causation:
- Counterfactual: Knowledge of what would happen to each
participant if they had not undergone a certain manipulation
- If we compare that knowledge with what happened, we know
what the effect of the manipulation is
o Perfect counterfactual doesn’t exist:
If it doesn’t exist, what CAN we do:
, - Randomized controlled trials: This ensures the conditions have
similar variability and thus that the control condition is a
representation of counterfactuals
- Different conditions: Using a control group or another condition to
show how the groups differ with different (or lack of) manipulation
- Matching methods: Finding a close ‘match’ (individuals that share
a lot of characteristics) to see the difference between them in
different conditions
Causal reasoning threats: (Campbell’s threats to valid causal
inference)
Ruling our alternative explanations, so whether effect of treatment is
confused with…
1. Outside factors: (confound variables)
a. History: Influences (outside of intervention) over the course of
the research, which influence outcome
b. Maturation: Natural changes that may be confused with
effect treatment
2. Effects of selection: (selection bias)
a. Selection: Selection criteria for treatment related to outcome
of treatment
b. Attrition: Participants’ dropout, systematically correlated with
conditions
i. When attrition (loss of participants) is high or non-
random, it can change the composition of the study
groups
3. Unintended effects of study itself: (Hawthorne effect, placebo
effect)
a. Instrumentation: Change in measuring instrument resulting
in a difference between pre- and post-measurement
b. Testing: Effect of measurement itself on subsequent
measurement (fatigue, habituation, placebo etc)
4. Statistical artifacts:
i. Patterns in data that arise due to limitations of
measurement
b. Regression to the mean: Extreme scores will be followed by
less extreme scores
i. Regression to the mean can create an illusion of
causation by making it look like an intervention brought
about improvement (or deterioration), when participants
are simply returning closer to their average baseline
Causal diagram:
Cutting the ties of the confound factors
- Randomize control is the gold standard
, Beyond correlation does not imply causation:
1. Ignoring causality
a. Some authors write down only the correlations they find,
without making any statement about causality
b. These statements makes it difficult to think of the use for the
findings
2. Statements of causality, but unclear assumptions
a. Other researchers make statements about causal relationships
based on correlational data, but often without specifying
assumptions
i. Conceptual: which third variables should be included in
the model and why not?
ii. Confounds are not mentioned
b. Misspecification of this relationship may leave residual
confounding
i. Refers to the risk of incorrectly modeling or
understanding the relationship between the variables
in a study, which can result in residual confounding—
a situation where confounding effects still influence the
results, even after attempting to control for them
3. Pseudo-correlational statements
a. No direct statements about causality, but clearly the implied in
the conclusion
i. Highlight form Y1 Psych: ‘The role of attachment style on
relationship satisfaction’
Causal diagram: (adding a third variable Z)
1. Z is a mediator
a. Only control (keeping a variable equal across conditions)
when wanting to know the direct effect
2. Z is a collider (common effect)
a. Collider bias: We assume there is a connection between the
causal variables (when studying the variables where the
collider is present)
