Hc 1 & 2 Causality, Macro-Micro
What is a theory
- A set of logically interrelated propositions about empirical reality
- A systematic explanation of the observed facts and laws that relate to a
specific aspect of life
Inference: The goal of social science ‘to infer beyond the immediate data to
something broader that is not directly observed.
Quantitative & Qualitative inference:
- From sample to population
- From case to broader set of cases
Two types of inference: Descriptive & causal inference:
- Focus on explanation (XY)
- But, description is the necessary first step. Show what is happening
before they attempt to explain why it is happening (Goldthorpe 2001)
Inference can follow different logics:
- Inductive (Theory generating)
- Deductive (Theory testing)
Examples of descriptive questions:
- How high is election turnout in different states of the US (Y)
- How did election turnout in the US develop over the past three decades?
(Y)
- How is the war in Ukraine framed by right-wing media? (Y)
Causal relations:
- How do economic conditions impact cross-state differences in turnout?
(XY)
- What is the effect of social media use on the decision to vote (XY)
- Through which mechanisms does social media use influence people’s
voting decisions (XY)
Causation as robust dependence (Goldthorpe, 2001):
While correlation does not imply causation, causation must in some way or other
imply association. In other words, if two variables correlate (X goes up, thus so
does Y):
- Robust: Correlation remains after controlling for possible other
explanations (confounders)
- Typically studies that use regression-based models on large datasets.
For example:
,Being ashamed of flying Number of flights. You expect this relation to be
negative. However, after testing this you discover a positive effect. What can
explain this? For example, elitism. This can explain your flight shame, and your
number flights. This is then not a causal relation, but a spurious association. If
you would leave out this confounder of elitism, you are dealing with omitted
variable bias. It looks like this:
Expectation: -
Flying shame Number of flights
Z
What you find: +
Flying shame Number of
flights
Z
Explanation: +
Flying shame Number of
flights
+ +
Elitism
Critiques on this robust dependence approach:
- Critique 1: No strict causal test, as confounder are always possible. You
can always add extra confounders that may influence the relation. You can
never know when you are completely controlled for. It is always
provisional. An alternative for this is consequential manipulation:
Counterfactual reasoning: Observe Y in the presence and absence
of X. You want to test Y first when X is present, and then when X is
not present. In these two situations, all other factors should be the
same. It should be a totally the same situation but for X. This is
attempted with random assignments.
The robust dependence approach starts by looking at the X variable
instead of the Y. It does not examines what causes Y, but what X
causes. The starting point is with X, not with Y.
Besides that, some causes can not be taken away to test. For
example, age or gender. Furthermore, the intentions of people are
difficult to manipulate. Lastly, there are ethics, as you can not
manipulate causes always, everywhere.
- Critique 2: What is the mechanism? If you only look at correlation between
two variables, you are solely explaining what effects the cause have, and
, not why this is the case. The arrow between X and Y remains unexplained.
An alternative for this is causation as generative process (mechanism-
based approach), which tackles two problems:
Problem 1: It is important to think of what role we assign to different
variables. This distinguishes the difference between confounders
and intervening variables (mediators).
Problem 2: Multiple paths are canceled out. There may be a positive
or negative path in terms of mediators, or confounders that are
overlooked when you do not examine the mechanism itself.
Causal mechanisms
A causal mechanisms is a theoretical account that:
- Shows how specific outcomes or empirical regularities come about
- Shows structure of the causal process
- Shows the logical connection between X and Y
Can/Should we empirically study causal mechanisms?
- At minimum, a theoretical specification. You don’t do research saying X
causes Y, and then not explaining why this is the case.
- If possible, empirically explain this relation (observable implications)
- Mechanisms require some form of methodological individualism. This goes
a step further. We should not just examine what happens in between, but
the causal mechanism itself is something that happens at the level of
individual actors. Statistical techniques can show only relations among
variables, and not how these relations are actually produced, as they can
indeed only be produced, through the action and interaction of individuals
(Goldthorpe, 2001).
Methodological individualism
- Social phenomena should be explained as resulting from the behavior of
individual actors
- These behaviors should be explained as motivated by intentional states of
the actors. We study phenomena that are founded by people. These are
social factors.
- The strong versions treat social phenomena as either:
Irrelevant
Or strictly endogenous: Social phenomena are strictly understood
as the intended consequences actions of individual actors. They are
only the explanandum, not the explanans
- The weak versions treat social phenomena as:
Of fundamental importance to understand why actors act the way
they act.
Both endogenous and exogenous: Social phenomena are
understood as the intended and unintended consequences of
actions of individual actors. But action is shaped by pre-existing
social context
Coleman’s Bathtub explains this:
A ----------------------------------------------------D Macro level
What is a theory
- A set of logically interrelated propositions about empirical reality
- A systematic explanation of the observed facts and laws that relate to a
specific aspect of life
Inference: The goal of social science ‘to infer beyond the immediate data to
something broader that is not directly observed.
Quantitative & Qualitative inference:
- From sample to population
- From case to broader set of cases
Two types of inference: Descriptive & causal inference:
- Focus on explanation (XY)
- But, description is the necessary first step. Show what is happening
before they attempt to explain why it is happening (Goldthorpe 2001)
Inference can follow different logics:
- Inductive (Theory generating)
- Deductive (Theory testing)
Examples of descriptive questions:
- How high is election turnout in different states of the US (Y)
- How did election turnout in the US develop over the past three decades?
(Y)
- How is the war in Ukraine framed by right-wing media? (Y)
Causal relations:
- How do economic conditions impact cross-state differences in turnout?
(XY)
- What is the effect of social media use on the decision to vote (XY)
- Through which mechanisms does social media use influence people’s
voting decisions (XY)
Causation as robust dependence (Goldthorpe, 2001):
While correlation does not imply causation, causation must in some way or other
imply association. In other words, if two variables correlate (X goes up, thus so
does Y):
- Robust: Correlation remains after controlling for possible other
explanations (confounders)
- Typically studies that use regression-based models on large datasets.
For example:
,Being ashamed of flying Number of flights. You expect this relation to be
negative. However, after testing this you discover a positive effect. What can
explain this? For example, elitism. This can explain your flight shame, and your
number flights. This is then not a causal relation, but a spurious association. If
you would leave out this confounder of elitism, you are dealing with omitted
variable bias. It looks like this:
Expectation: -
Flying shame Number of flights
Z
What you find: +
Flying shame Number of
flights
Z
Explanation: +
Flying shame Number of
flights
+ +
Elitism
Critiques on this robust dependence approach:
- Critique 1: No strict causal test, as confounder are always possible. You
can always add extra confounders that may influence the relation. You can
never know when you are completely controlled for. It is always
provisional. An alternative for this is consequential manipulation:
Counterfactual reasoning: Observe Y in the presence and absence
of X. You want to test Y first when X is present, and then when X is
not present. In these two situations, all other factors should be the
same. It should be a totally the same situation but for X. This is
attempted with random assignments.
The robust dependence approach starts by looking at the X variable
instead of the Y. It does not examines what causes Y, but what X
causes. The starting point is with X, not with Y.
Besides that, some causes can not be taken away to test. For
example, age or gender. Furthermore, the intentions of people are
difficult to manipulate. Lastly, there are ethics, as you can not
manipulate causes always, everywhere.
- Critique 2: What is the mechanism? If you only look at correlation between
two variables, you are solely explaining what effects the cause have, and
, not why this is the case. The arrow between X and Y remains unexplained.
An alternative for this is causation as generative process (mechanism-
based approach), which tackles two problems:
Problem 1: It is important to think of what role we assign to different
variables. This distinguishes the difference between confounders
and intervening variables (mediators).
Problem 2: Multiple paths are canceled out. There may be a positive
or negative path in terms of mediators, or confounders that are
overlooked when you do not examine the mechanism itself.
Causal mechanisms
A causal mechanisms is a theoretical account that:
- Shows how specific outcomes or empirical regularities come about
- Shows structure of the causal process
- Shows the logical connection between X and Y
Can/Should we empirically study causal mechanisms?
- At minimum, a theoretical specification. You don’t do research saying X
causes Y, and then not explaining why this is the case.
- If possible, empirically explain this relation (observable implications)
- Mechanisms require some form of methodological individualism. This goes
a step further. We should not just examine what happens in between, but
the causal mechanism itself is something that happens at the level of
individual actors. Statistical techniques can show only relations among
variables, and not how these relations are actually produced, as they can
indeed only be produced, through the action and interaction of individuals
(Goldthorpe, 2001).
Methodological individualism
- Social phenomena should be explained as resulting from the behavior of
individual actors
- These behaviors should be explained as motivated by intentional states of
the actors. We study phenomena that are founded by people. These are
social factors.
- The strong versions treat social phenomena as either:
Irrelevant
Or strictly endogenous: Social phenomena are strictly understood
as the intended consequences actions of individual actors. They are
only the explanandum, not the explanans
- The weak versions treat social phenomena as:
Of fundamental importance to understand why actors act the way
they act.
Both endogenous and exogenous: Social phenomena are
understood as the intended and unintended consequences of
actions of individual actors. But action is shaped by pre-existing
social context
Coleman’s Bathtub explains this:
A ----------------------------------------------------D Macro level
