Causal Diagrams

causal diagrams

the basic cause-effect relation (smoking → lung cancer)

mechanism or mediator

• explains how the main cause influences the main effect

  • (smoking → change of DNA in lung cells → lung cancer)

measures confounding variables (common causes)

• Spurious correlation: Can make it appear like there is a correlation even when there really not’t (false positive)

• Can make a relationship appear stronger than it actually is

• Can make a relationship appear in the opposite direction

common causes

• When we are worried that there might be a common cause, we must statistically control for it

• You should only control for variables that are plausibly direct causes of the DV

unmeasured confounding variables

• An alternative cause that you don’t know about, did not measure, or cannot measure

• Also known as noise variables

How do we figure out the causal structure

• We do studies about each cause-effect relationship to build it up link by link

• Evidence that is necessary for each causal link:

  • A correlation or statistical relationship

  • Temporal precedence (causal direction)

  • Eliminate (or reduce) confounds

observational studies

• Can provide evidence of a statistical relationship (correlation)

• Can provide evidence of temporal precedence, which provides evidence for the causal direction

• However, in all observational studies, there can still be confounds

How do experiments provide evidence for causality?

• Provide evidence of a statistical relationship

• Temporal precedence provides evidence for the causal direction

• Removes confounds or a least dramatically reduced the likelihood of confounds through randomization