Randomization

• Purpose of Randomization: Solves key issues in causal inference
  • Consistency
  • Positivity
  • Does not address interference

Ideal vs Realizstic Randomized Trials

• Ideal Randomized Trials
  • Achieve exchangeability
• Realistic Randomized Trials
  • May violate exchangeability with non-adherence / dropout

Protcol elements mapped to assumptions

Mapped to diagraming causal claims

Target Trials

• There are many reasons why randomization may not be possible
  • it might not be ethical to randomly assign people to a particular exposure
  • there may not be funding available
  • there might not be enough time to conduct a full trial

Target Trial

• In these situations, we rely on observational data to help us answer causal questions by implementing a target trial
• A target trial answers: What experiment would you design if you could?

Let’s try a simple example

# A tibble: 10,000 × 3
   confounder exposure outcome
        <int>    <int>   <dbl>
 1          0        0 -0.804 
 2          0        0 -1.06  
 3          1        1 -0.0354
 4          1        1 -0.186 
 5          0        0 -0.500 
 6          1        1  0.475 
 7          1        0  0.698 
 8          1        0  1.47  
 9          1        0  0.752 
10          0        0  1.26  
# ℹ 9,990 more rows

Let’s try a simple example

Let’s try a simple example

Causal Diagram

Application Exercise

Application Exercise