Research Design, Causal Inference, and Translation
From Correlation to Causation
Why causal inference matters
Applied work must distinguish association from intervention effects. In practice, this means choosing designs that can address confounding, selection bias, reverse causality, and spillover effects. Randomized controlled trials remain a gold standard, but field constraints often require quasi-experimental designs.
Design principle
Start with the decision you want to inform, then choose the weakest design that still yields a credible causal answer.
Key Designs and Tradeoffs
Common causal designs in applied psychology
| Design | Strengths | Limitations |
|---|---|---|
| Randomized experiment | High internal validity | Cost, ethics, compliance issues |
| Cluster randomized trial | Reduces contamination | Requires more clusters, complex analysis |
| Difference-in-differences | Useful for policy change evaluation | Parallel-trends assumption |
| Regression discontinuity | Strong local causal inference | Only near threshold, requires large samples |
| Instrumental variables | Addresses unobserved confounding under assumptions | Hard to find valid instruments |
What is a major advantage of randomized assignment in applied research?
Randomization balances observed and unobserved confounds on average, strengthening causal claims.
Correct answer: It reduces confounding by balancing groups on average
Name one assumption required for a difference-in-differences design.
The treated and comparison groups should have followed similar outcome trends absent the intervention.
Correct answer: Parallel trends.
Implementation and transportability
A central frontier topic is transportability: whether an effect estimated in one population, institution, or time period can be generalized elsewhere. Applied psychologists increasingly combine causal models with data on context moderators, implementation fidelity, and mechanism variables to predict where interventions will succeed.
Transportability is primarily concerned with what question?
Transportability asks whether a causal effect or intervention can be moved to a new setting.
Correct answer: Whether an effect generalizes across settings and populations
Give one reason implementation fidelity matters.
Low fidelity confounds the interpretation of null or weak effects.
Correct answer: Without fidelity, an intervention may fail because it was not delivered as intended.