This document discusses causality and empirical methods in social sciences. It addresses why causality is an important epistemic norm that shapes how social phenomena are conceptualized and studied. Different views of causality - as something real in the world or as part of statistical models - lead to different modeling approaches. Quantitative and qualitative methods each have strengths and limitations, and combining the two may provide richer insights than either approach alone. Precisely defining and measuring concepts like socioeconomic status is challenging, and larger data sets and more sophisticated tools do not necessarily yield more meaningful results. Causality and choice of methods strongly influence research conclusions.
Causality and Mixed Methods in Social Science Research
1. Causality and Empirical Methods
in the Social Sciences
Federica Russo
Philosophy & ILLC | University of Amsterdam
russofederica.wordpress.com | @federicarusso
2. Overview
Causality and causal modelling
Why is it worth having a concept of causality in social research?
What kind of work does ‘causality’ do?
Empirical methods and causality
The question of measurements and the need of both quali and quanti approaches
2
3. Why causality?
Causality is more than just a
Concept
Condition to test
Causality is an essential epistemic norm:
It shapes the way we conceptualise the world and (decide) how to study it
It has also effects on normative (ethical/moral/political) aspects of research
3
5. Causality is in the world
A ‘thing’ in the world, but not a norm in our head
Typically held by philosophers of the natural sciences / scientists
Scientific realism and causal realism and laws
If causes are out there, we ‘just’ need tools to discover them
Models of increasing sophistication will allow us to pick out the true causes
The best tools are arguably (allegedly?) experimental
5
6. Causality is in the model
Statistical modelling
The whole literature from economics/econometrics, up to Pearl and Bayes Nets
The new holy grail: exogenity, causal Markov condition
Conditions to test
A test and a super test to ensure you get causality out of very sophisticated statistical
modelling
6
7. How to find causality in the social world
The gold standard of experimental methods spreads to social science (and
epidemiology)
See debate on potential outcome models
Sociologist and methodologist Goldthorpe: experimental methods do not suit the object of
study of the social sciences
Epidemiologists Vandenbrouke et al: one can’t treat e.g. race and ethnicity as proper causes
General methodological points
Pluralism
No gold standards
7
8. Causality is not in the social world
There are no causes in the social world
Too unstable and mutable to be pinned down
Two options
Pragmatic: We can describe, at best
Principled: Everything interacts anything else (systemism)
8
10. More than
‘no causes in, no causes out’
Cartwright’s law
Right overall
You can’t infer causes from correlations. You infer causes from causal hypotheses
But just talking about the model
What happens before and after the model?
10
11. Two functions of the epistemic norm
Before: Guiding modelling choices
Some models fundamentally depend on specific conceptualisations of causality (or lack
thereof)
After: Shaping conclusions and recommendations of empirical studies
One ought to reflect on how far causal conclusions should go beyond modelling
11
13. Contrast and compare
Causal modelling
Closure of the system and
mechanisms
The agent is external
Causal mechanisms are established
using prior information
Systemic approach
Every thing interacts with everything
else
The agent is internal
Structures are identified without prior
information
13
14. A case study:
health system and mortality
54
4
13
34
12
2
X1
Economic
development
X2
Social development
X3
Sanitary
infrastructures
X4
Use of sanitary
infrastructures
X5
Age structure
Y
Mortality
14
15. A criticism from within social science
Lauriaux (1994): theoretical weaknesses of causal analysis
Choice of variables, conceptualisation, closure of the system
Specifically:
Principal variables are theoretical constructs according to well established
economic and sociological theories
Assumption: economic development generates social development
Problem: counterexamples exist, the arrow might be reversed with serious
problems for policy
To intervene on an effect which is not an effect won’t deliver the planned
results
15
16. An alternative: systemism
Systems are homeostatic:
They keep themselves in a stable state by means of regulatory interdependent
mechanisms
Changes in the system re-establish the equilibrium in consequence of too strong
internal/external influences
In the process of balancing, components jointly evolve
Those joint evolutions are covariations we call causal
16
19. Causality as an epistemic norm
It is about a fundamental difference in
Thinking what the (social) world is like
Designing and implementing methods to study (social) reality
Explaining a (social) phenomenon
Making recommendations for intervention
See next part for specific considerations about measurement
19
20. The more you measure, the better
An easy solution a complex problem?
20
21. Measurement itself, especially if carried out using sophisticated instruments or analysed
using complex methodology, is seen to have the attributes of ‘science’, and often
taken effectively as a justification for believing the results that are presented as
if they have a meaningful relation to whatever social process they are
claimed to measure. […]
New technologies such as powerful dynamic computer graphics do have the potential to
convey findings and patterns in powerful ways, but whether they are used to
inform rather than merely impress, remains an open question.
[…] a better understanding is needed of the difference between data that
‘confirms’ a theory by providing a good model fit, and data that allows
us to explain observed data patterns using as much potentially falsifiable
information as possible.
Harvey Goldstein (2012)
21
22. What do we actually measure?
Measuring “socio-economic status” and “age”
22
23. At the extremes of measurement
Age
Very easy to measure
Does it just represent a definite
biological status?
Does it have any explanatory import?
SES
Very controversial how we should
measure it
How can it represent one’s status?
What is its import in explanation of social
or social / health outcomes?
23
24. Measuring socio-economic status
Theoretical approaches
Weberian, Marxist, Colemanian, …
Identification of different indicators, different types of variables
Procedure: class stratification
E.g., Goldthorpe Class Schema
Grouping of types of workers
Why measuring SES?
E.g., correlation with health outcomes, or other economic variables, …
24
25. What does SES do?
Categorise?
A classificatory variable
What part of the populations are more exposed, have higher prevalence …
Explain?
Active part in the explanation of diseases
Mixed aetiology!
What are the active causal pathways from exposure to outcome?
Social practices / norms / habits to explain (and to prevent) exposure
25
26. Measuring age
Easy to measure
Accessibility of data, straightforward question, …
How to measure it
Categorically, Continuously
Using Age
Control: Adjust results of statistical analyses (control for age)
Predict: Age structure helps predict results
Categorise: Grouping and collapsing multiple categories into fewer categories; Care with loss of
information, residual confounding
26
27. Why measuring age?
‘Demographic’ age: Locating individuals in the ‘right’ age group
Biological age: A typical health status, for that age
Social age: Social practices that are typical of that age
Epigenetic age: Our internal clock, possibly different from our chronological age
…
[these meanings of age do coincide, possibly they overlap]
Any explanatory import in using age?
27
29. Modelling data, interpreting outcomes
Quantitative studies
Large samples and data sets
Analysis of correlations
Validation via robustness tests, etc.
Qualitative studies
Small samples and groups
Description of practices
Difficult to generalise
29
30. The sample: the bigger, the better
Measurement: the more precise, the better
Another simple solution to an even more complex problem?
30
31. The extra information that statistics does
not give us
Description of
Practices
Interactions
Influences
Background
Norms
…
GO small FIRST!
31
32. The information that statistics does give us
Categorise the ‘practices, interactions, backgrounds, …’ into measurable variables
Is it generalisable?
An empirical question. Not a priori, determined by size
now go BIG!
32
34. How causality acts as a norm
Guiding fundamental choices about modelling
These are not unrelated to one’s worldview
Deeply influence modelling
Shaping conclusions of empirical studies
These may not remain epistemic and easily spill over the ethico-political domain
34
35. Quantitative measurement mantras: a social science trend, also to meet requests
from policy-making
What is evidence for decisions?
What methods we can trust?
…
Reflecting on measuring SES and Age we may conclude that
Quantitative measurement is not necessarily panacea
Qualitative measurement can help a great deal
Next item on the research agenda
How to combine quali- and quantitative data?
To what extent do QCA, mixed/multi-methods succeed?
35
36. Causality and Empirical Methods
in the Social Sciences
Federica Russo
Philosophy & ILLC | University of Amsterdam
russofederica.wordpress.com | @federicarusso
Thanks for your attention!