Causal inference Flashcards
What is causality
The relationship between cause and effect.
What is the difference between deterministic and non-deterministic causality.
Deterministic: the effect always follows the cause.
Non-deterministic: there is a causal relationship, but the cause does not always lead to the effect.
Why is causality important?
The business question is often not if there is a cause behind some effect, but what that cause is and how to change/maintain it.
Differentiate between causal effects and associations,
x has a causal effect on y if y “listens to” x.
x and y are associated if knowing one allows us to predict the other better on average.
What mathematical object can we use to understand the causal structure between two object?
Directed acyclic graphs, with nodes at each variable and directed edges representing causal relationshops.
How would direct/indirect causal relationships be represented using this structure?
Direct: arrow from x to y.
Indirect: arrow from x to some mediator variable, then to y.
What is conditioning on a variable?
Narrow our focus to the cases where the conditioning variable takes the value we are interested in. E.g. measure the health of people who take health supplements.
What is intervening on a variable?
Fix the variable to take the value we are interested in. E.g. measure the health of people half of whom you instruct to take health supplements.
When there are external factors that may come into play when looking for the effect of variable x on another, y, is it better to condition or intervene on x?
Intervene: this removes all dependence on external factors, revealing the true extent of the causal relationship between x and y.
Give an example of a chain.
x -> y -> z
Amount smoked -> tar buildup on lungs -> cancer.
Describe the causal structure of a chain using this example.
amount smoked and tar buildup are dependent
tar buildup and cancer are dependent
amount smoked and cancer are dependent
amount smoked and cancer are independent conditional on tar build up (since knowing someone smokes given that they have tar buildup gives no new information.)
Give an example of a fork
x z
ice cream sales crime rate
Describe the causal structure of a fork using this example.
ice cream sales and temperature are dependent
temperature and crime rate are dependent
ice cream sales and crime rate are dependent
ice cream sales and crime rate are independent conditional on temperature: knowing that ice cream sales are high given the fact that the temperature is hot provides no new information on crime rate.
Give an example of a collider
x -> y uni admission
Describe the causal structure of a collider using this example.
academic achievement and uni admission are dependent
uni admission and sport achievement are dependent (in US)
academic achievement and sporting achievement are independent
academic achievement and sporting achievement are dependent conditional on uni admission, since knowing someone is poor academically given that they are admitted to uni allows you to predict they have sporting achievements.
