Sampling, Bias, and Experimental Design
Sampling and Causation
Why randomness helps
Random sampling reduces systematic error in estimating population quantities, while random assignment in experiments helps isolate causal effects by balancing confounders across groups.
Sampling methods
Simple random sample
- Every unit has equal selection probability
- Best for minimizing selection bias
- Requires a full sampling frame
Stratified sample
- Population split into strata before sampling
- Improves precision when strata are internally similar
- Useful when subgroups must be represented
Bias sources
Correlation is not causation
An observed association does not prove that one variable causes the other. Experimental control or strong causal design is needed.
Designing a simple experiment
-
1
Step 1: State the treatment and response clearly.
-
2
Step 2: Randomly assign units to treatment groups.
-
3
Step 3: Use control or placebo groups when appropriate.
-
4
Step 4: Keep procedures consistent across groups.
-
5
Step 5: Measure outcomes and analyze differences with uncertainty in mind.
Which design feature most directly supports causal inference?
Random assignment helps balance confounders and supports causal interpretation.
Correct answer: Random assignment to treatment groups
What is the difference between random sampling and random assignment?
Sampling affects representativeness; assignment affects causality.
Correct answer: Random sampling selects units from a population; random assignment allocates sampled units to treatments.