The concept of variation in causal discovery
Or, why causality needs difference
Federica Russo
Dipartimento di Studi Umanistici, Università di Ferrara
https://blogs.kent.ac.uk/federica
Overview
Causal reasoning and ‘variational’ epistemology
Ordinary, experimental, statistical
Foundations of variational reasoning
Mill, Durkheim
Why difference? Why (not) regularity?
Learning causes and the role of regularity
2
CAUSAL REASONING
3
‘Ordinary’ causes
“Had I left home earlier,
I wouldn’t have missed the flight”
Pin down the cause to understand
why something did (not) occur
4
‘Experimental’ causes
Hypothesise the function of a gene, say TP53
Knock out that gene
Observe changes in appearance, behaviour, physical &
biochemical characteristics
Reconstruct mechanisms to understand disease
causation and act in response to that knowledge
5
‘Statistical’ causes
Gather a large number of observations,
organise them in variables
E.g. socio-biological characteristics (exposure) and cancer
rates (disease)
Study the (in)dependencies between variables,
robustness and stability of correlations
Establish stable patterns of (in)dependencies
to identify risk factors and possible interventions
6
EPISTEMOLOGY OF CAUSAL
REASONING
7
Different questions, different answers
What is causation?
What are causes?
What does causality / cause mean?
How do we find out about causes?
What notions guide causal
reasoning?
What to do with causes?
How to use causal knowledge?
Metaphysics / Semantics /
Conceptual analysis
Epistemology /
Methodology
Use
8
Use
Epistemology
Metaphysics Methodology
Semantics
Different questions, different answers
What is causation?
What are causes?
What does causality / cause mean?
How do we find out about causes?
What notions guide causal
reasoning?
What to do with causes?
How to use causal knowledge?
Metaphysics / Semantics /
Conceptual analysis
Epistemology /
Methodology
Use
10
THE RATIONALE OF VARIATION
11
Causal discovery is reasoning about variations.
To establish causes we need difference.
12
‘Ordinary’ variations
“Had I left home earlier,
I wouldn’t have missed the flight”
Leaving home on time / late makes a difference to
missing the flight
Counterfactual reasoning: search for the element
changing the chain of events
13
‘Experimental’ variations
“Knock out TP53 and observe what happens to cell
division or apoptosis”
Change putative causal factors to see
what changes (don’t) follow.
Experimental reasoning: search for those manipulable
factors changing causal structures
14
‘Statistical’ variations
“Gather data about socio-economic status, occupation,
diet, smoking behaviour and see how steadily they
are associated with cancer”
Study how variations in exposure are related to
variations in disease.
How different levels of exposure change the probability
of disease.
Statistical reasoning: search for those factors explaining
the variance of the outcome.
15
FOUNDATIONS
16
Variations in MillAgreement:
comparing different instances in which the
phenomenon occurs.
Difference:
comparing instances in which the
phenomenon does occur with similar
instances in which it does not.
Residues:
subducting from any given phenomenon all
the portions which can be assigned to
known causes, the remainder will be the
effect of the antecedents which had been
overlooked or of which the effect was as
yet an un-known quantity.
Concomitant Variation:
in presence of permanent causes or
indestructible natural agents that are
impossible either to exclude or to isolate,
we can neither hinder them from being
present nor contrive that they shall be
present alone. Comparison between
concomitant variations will enable us to
detect the causes.
Mill (1843), System of Logic
The experimental method is based
on the Baconian rule of varying
the circumstances
The Four Methods are all based on
the evaluation of variations
17
Variations in Durkheim
Durkheim (1897), Le suicide
A study into the variability of suicide rate.
A search for the causes making suicide rate vary.
Durkheim (1885), Les règles de la méthode sociologique
The method of concomitant variations
makes sociology scientific.
18
WHY DIFFERENCE?
19
Regular causes?
Day follows night, night follows day
Days follow nights regularly
Day and night are different
It is the variation from day to night and night to day
that is regular
Search for the element that makes C↣E regular
20
Epistemology
/
methodology
No causation without variation
Problem:
? Effects of Islamic culture on
gender equality ?
Consider only
Islamic countries
No variations detected!
Compare Islamic with
non-Islamic cultures
Variations can be detected >
variations assessed > causal
relations established
We can’t detect causal
relations in populations
too homogeneous
too heterogeneous
To detect causal relations
we need
variation, change, difference
21
WHY NOT REGULARITY?
22
Learning ‘ordinary’ causes
‘Humean’ causal learning
Instances of smoke follow instances of fire
Can’t establish logical, necessary link
Create expectation, project causal belief onto the future
Studies in causal cognition: we learn causal relations
From covariation
Manipulating difference-makers
We learn from difference, before regularity
23
Learning ‘scientific’ causes
Causal discovery (experiments, statistics)
Search for differences
Probability of the outcome given the cause; effects of
manipulations; …
Explain a phenomenon by appealing to causes
Most causes are ‘difference-makers’ in mechanisms
regulating health and behaviour.
Variation at the basis of causal methodology
24
Regularity too
Statistical regularity
Causal methodology needs regularity as a constraint on
variations, differences
Regular causal relations are ‘generic’
Population-level, repeatable
Hence we need regularity to establish generic level
25
SUM UP AND CONCLUDE
26
Metaphysics /
Conceptual Analysis
• What is
causation?
• What are causes?
• What does
causality / cause
mean?
Epistemology /
Methodology
• How do we find
out about
causes?
• What notions
guide causal
reasoning?
Use
• What to do with
causes?
• How to use causal
knowledge?
27
Concept of variation
in causal reasoning
The rationale of variation…
… underpins causal discovery
Ordinary
Experimental
Statistical
Variation, difference
the common denominator of various forms
of causal reasoning
28
Causal discovery is reasoning about variations.
To establish causes we need difference.
29
Further ‘variational’ readings
Russo F. (2009). Causality and Causal Modelling in the Social Sciences.
Measuring Variations. Springer.
Russo F. (2011). Correlational data, causal hypotheses, and validity. Journal for
General Philosophy of Science, 42(1), 85-107.
Russo F. (2012). On empirical generalisations. In D. Dieks, W.J. Gonzalez, S.
Hartmann, M. Stoeltzner, M. Weber (eds), Probabilities, Laws, and
Structures, 133-150, Springer.
Russo F. (2009). Variational causal claims in epidemiology, Perspectives in
Biology and Medicine, 52(4), 540-554.
Russo F. (2006). The rationale of variation in methodological and evidential
pluralism. Philosophica, 77. Special Issue on Causal Pluralism, 97-124.
Illari P. and Russo F. Causality: Philosophical Theory Meets Scientific Practice.
Oxford University Press. Almost in press!
30

The concept of variation in causal discovery

  • 1.
    The concept ofvariation in causal discovery Or, why causality needs difference Federica Russo Dipartimento di Studi Umanistici, Università di Ferrara https://blogs.kent.ac.uk/federica
  • 2.
    Overview Causal reasoning and‘variational’ epistemology Ordinary, experimental, statistical Foundations of variational reasoning Mill, Durkheim Why difference? Why (not) regularity? Learning causes and the role of regularity 2
  • 3.
  • 4.
    ‘Ordinary’ causes “Had Ileft home earlier, I wouldn’t have missed the flight” Pin down the cause to understand why something did (not) occur 4
  • 5.
    ‘Experimental’ causes Hypothesise thefunction of a gene, say TP53 Knock out that gene Observe changes in appearance, behaviour, physical & biochemical characteristics Reconstruct mechanisms to understand disease causation and act in response to that knowledge 5
  • 6.
    ‘Statistical’ causes Gather alarge number of observations, organise them in variables E.g. socio-biological characteristics (exposure) and cancer rates (disease) Study the (in)dependencies between variables, robustness and stability of correlations Establish stable patterns of (in)dependencies to identify risk factors and possible interventions 6
  • 7.
  • 8.
    Different questions, differentanswers What is causation? What are causes? What does causality / cause mean? How do we find out about causes? What notions guide causal reasoning? What to do with causes? How to use causal knowledge? Metaphysics / Semantics / Conceptual analysis Epistemology / Methodology Use 8
  • 9.
  • 10.
    Different questions, differentanswers What is causation? What are causes? What does causality / cause mean? How do we find out about causes? What notions guide causal reasoning? What to do with causes? How to use causal knowledge? Metaphysics / Semantics / Conceptual analysis Epistemology / Methodology Use 10
  • 11.
    THE RATIONALE OFVARIATION 11
  • 12.
    Causal discovery isreasoning about variations. To establish causes we need difference. 12
  • 13.
    ‘Ordinary’ variations “Had Ileft home earlier, I wouldn’t have missed the flight” Leaving home on time / late makes a difference to missing the flight Counterfactual reasoning: search for the element changing the chain of events 13
  • 14.
    ‘Experimental’ variations “Knock outTP53 and observe what happens to cell division or apoptosis” Change putative causal factors to see what changes (don’t) follow. Experimental reasoning: search for those manipulable factors changing causal structures 14
  • 15.
    ‘Statistical’ variations “Gather dataabout socio-economic status, occupation, diet, smoking behaviour and see how steadily they are associated with cancer” Study how variations in exposure are related to variations in disease. How different levels of exposure change the probability of disease. Statistical reasoning: search for those factors explaining the variance of the outcome. 15
  • 16.
  • 17.
    Variations in MillAgreement: comparingdifferent instances in which the phenomenon occurs. Difference: comparing instances in which the phenomenon does occur with similar instances in which it does not. Residues: subducting from any given phenomenon all the portions which can be assigned to known causes, the remainder will be the effect of the antecedents which had been overlooked or of which the effect was as yet an un-known quantity. Concomitant Variation: in presence of permanent causes or indestructible natural agents that are impossible either to exclude or to isolate, we can neither hinder them from being present nor contrive that they shall be present alone. Comparison between concomitant variations will enable us to detect the causes. Mill (1843), System of Logic The experimental method is based on the Baconian rule of varying the circumstances The Four Methods are all based on the evaluation of variations 17
  • 18.
    Variations in Durkheim Durkheim(1897), Le suicide A study into the variability of suicide rate. A search for the causes making suicide rate vary. Durkheim (1885), Les règles de la méthode sociologique The method of concomitant variations makes sociology scientific. 18
  • 19.
  • 20.
    Regular causes? Day followsnight, night follows day Days follow nights regularly Day and night are different It is the variation from day to night and night to day that is regular Search for the element that makes C↣E regular 20
  • 21.
    Epistemology / methodology No causation withoutvariation Problem: ? Effects of Islamic culture on gender equality ? Consider only Islamic countries No variations detected! Compare Islamic with non-Islamic cultures Variations can be detected > variations assessed > causal relations established We can’t detect causal relations in populations too homogeneous too heterogeneous To detect causal relations we need variation, change, difference 21
  • 22.
  • 23.
    Learning ‘ordinary’ causes ‘Humean’causal learning Instances of smoke follow instances of fire Can’t establish logical, necessary link Create expectation, project causal belief onto the future Studies in causal cognition: we learn causal relations From covariation Manipulating difference-makers We learn from difference, before regularity 23
  • 24.
    Learning ‘scientific’ causes Causaldiscovery (experiments, statistics) Search for differences Probability of the outcome given the cause; effects of manipulations; … Explain a phenomenon by appealing to causes Most causes are ‘difference-makers’ in mechanisms regulating health and behaviour. Variation at the basis of causal methodology 24
  • 25.
    Regularity too Statistical regularity Causalmethodology needs regularity as a constraint on variations, differences Regular causal relations are ‘generic’ Population-level, repeatable Hence we need regularity to establish generic level 25
  • 26.
    SUM UP ANDCONCLUDE 26
  • 27.
    Metaphysics / Conceptual Analysis •What is causation? • What are causes? • What does causality / cause mean? Epistemology / Methodology • How do we find out about causes? • What notions guide causal reasoning? Use • What to do with causes? • How to use causal knowledge? 27 Concept of variation in causal reasoning
  • 28.
    The rationale ofvariation… … underpins causal discovery Ordinary Experimental Statistical Variation, difference the common denominator of various forms of causal reasoning 28
  • 29.
    Causal discovery isreasoning about variations. To establish causes we need difference. 29
  • 30.
    Further ‘variational’ readings RussoF. (2009). Causality and Causal Modelling in the Social Sciences. Measuring Variations. Springer. Russo F. (2011). Correlational data, causal hypotheses, and validity. Journal for General Philosophy of Science, 42(1), 85-107. Russo F. (2012). On empirical generalisations. In D. Dieks, W.J. Gonzalez, S. Hartmann, M. Stoeltzner, M. Weber (eds), Probabilities, Laws, and Structures, 133-150, Springer. Russo F. (2009). Variational causal claims in epidemiology, Perspectives in Biology and Medicine, 52(4), 540-554. Russo F. (2006). The rationale of variation in methodological and evidential pluralism. Philosophica, 77. Special Issue on Causal Pluralism, 97-124. Illari P. and Russo F. Causality: Philosophical Theory Meets Scientific Practice. Oxford University Press. Almost in press! 30