Lecture 10

Question 1

What is Causality?

Answer 1

Causal modelling in observational data.

Question 2

Where is Causality crucial?

Answer 2

In Responsible and Explainable AI.

Question 3

What is the Classical rigid approach?

Answer 3

Causal interference is only possible in RCT’s (Randomized Clinical Trials).

Question 4

How does a RCT work?

Answer 4

All patients are randomly assigned to a control group and an experiment group.

Question 5

What is the goal of the Classical rigid approach?

Answer 5

To estimate the effect of the IV (Independent Variable) on the DV (Dependent Variable).

Question 6

What is the Independent Variable?

Answer 6

Usually an intervention, exposure or treatment?

Question 7

What is the Dependent Variable?

Answer 7

Usually represents a survival time, test score, measurement or a binary outcome.

Question 8

What can we assume due to randomisation (Classical rigid approach)?

Answer 8

That both groups are identical with respect to all measured/unmeasured variables (except the IV).

Question 9

What causes the differences between C and E with respect to the DV?

Answer 9

The IV (cause-effect relation).

Question 10

Why is RCT often not possible? (4 reasons)

Answer 10

Unethical, logically impossible, too expensive, too timeconsuming.

Question 11

What kinds of studies are done in a RCT without randomisation?

Answer 11

Observational studies.

Question 12

What can we not assume in a RCT without randomisation?

Answer 12

That both groups are identical with respect to all (un)measured variables.

Question 13

What does the absence of identity affect and jeopardize (RCT without randomisation)?

Answer 13

The established cause-effect relation is affected and the validity of the research is jeopardized (bias could occur).

Question 14

What is the Confounding Variable?

Answer 14

A variable that influences both the dependent and the independent variable.

Question 15

What is a disadvantage of the Confounding Variable?

Answer 15

It may cause a false association between IV and DV.

Question 16

What may the controlling for the confounder do? (Confounding Variable)

Answer 16

It may drop the association between IV and DV to zero.

Question 17

What may the 3rd variable do? (2 things) (Confounding variable)

Answer 17

1. It may weaken the effect of IV on DV.

2. It may cause the sign to switch (positive negative).

Question 18

Are Correlations / Associations enough to identify a confounding variable?

Answer 18

No.

Question 19

What is a Mediation Variable?

Answer 19

A mediation model is used to identify and explain the mechanism/process that underlies an observed relationship between an IV and DV by including a 3rd variable, the mediation variable.

Question 20

What kind of relation can you find between IV and DV with the Mediation model?

Answer 20

An Indirect causal relation.

Question 21

What is the Mediation Variable part of? What does it serve for?

Answer 21

It is part of the causal path and serves to clarify the nature of the relationship between IV and DV.

Question 22

When do Mediation analysis work best?

Answer 22

When the IV and DV don’t have an obvious connection.

Question 23

What is Full Mediation?

Answer 23

When the inclusion of the mediation variable drops the relationship between IV and DV to zero.

Question 24

What is Partial Mediation?

Answer 24

When the inclusion of the mediation variable accounts for some of the relationship between IV and DV (direct and indirect effects).

Question 25

What does the Moderator Variable do?

Answer 25

It explains the strength of the relationship between the dependent and independent variable.

Question 26

Can one draw causal arrows when only analysing a single cotingency table?

Answer 26

No.

Question 27

What kind of concept is association?

Answer 27

A symmetric concept.

Question 28

What is the relation of correlation to causality (Simpsons paradox)?

Answer 28

correlation IS NOT causality.

Question 29

What is the relation of skewness to bias (Simpsons paradox)?

Answer 29

skewness IS NOT bias.

Question 30

What kind of process is the Simpsons paradox?

Answer 30

A mediation process.

Question 31

When can we speak of the Simpsons paradox?

Answer 31

When a positive association between two categorical variables is negative in all classes of a third variable we control for.

Question 32

How do we determine the weights of the arrows when making a model?

Answer 32

The weights of arrows can be estimated with statistical techniques such as multiple regression analysis, path analysis and structural equation modelling (SEM).