Quasi-experiment

A study that estimates the effects of an intervention without randomly assigning participants to treatment conditions. It uses design or statistical methods to address differences between groups.

Difference-in-differences: A method that estimates an intervention effect by comparing outcome changes over time between treated and comparison groups. It separates the intervention’s estimated effect from shared time trends.
Causal inference: The study of methods for estimating cause-and-effect relationships from data. Quasi-experiments are one family of designs for estimating causal effects.
Randomized controlled trial: An experiment that assigns participants to treatment conditions by chance and compares their outcomes. Random assignment balances causes of outcomes in expectation; quasi-experiments lack this safeguard.
Policy evaluation: The systematic assessment of a public policy’s implementation, effects, or value. Policies often begin without researcher-controlled assignment, making quasi-experimental designs useful.
Causal estimand: A precisely defined causal quantity a study aims to estimate for a specified population and intervention. Different designs may estimate effects for different populations or treatment margins.
Regression discontinuity design: A design that estimates treatment effects by comparing cases just above and below a threshold governing treatment assignment. A cutoff can create near-comparable groups without random assignment.
Potential outcomes framework: A framework that represents each unit’s outcome under each possible treatment condition, though only one is observed. It defines the treatment effect through outcomes that cannot both be observed for the same unit.
Observational study: A study that observes exposures and outcomes without assigning participants to conditions. Quasi-experiments are a more design-driven subset of nonrandomized research.
Program evaluation: The systematic assessment of a program’s design, implementation, outcomes, or impacts. Programs commonly reach participants through administrative or practical selection rather than random assignment.
Unmeasured confounding: Confounding caused by factors that were not measured or adequately represented in an analysis. Statistical adjustment cannot directly balance unknown or unrecorded causes of treatment and outcome.

Linked from 15 pages