An analytics concept defining statistical methods used to estimate relationships, evaluate interventions, and generate forecasts. It specifies data requirements, estimation procedures, and uncertainty measures used to support decision-making. It does not prove causation without an appropriate identification strategy, data quality checks, and sensitivity analysis. It supports performance management by translating data into estimates, predictions, and quantified uncertainty. The concept is generally stable, though tooling and best practices for measurement evolve over time.

Principle

Identification rests on the parallel trends assumption: absent treatment, the average outcome trajectory of treated and control groups would have followed the same path, so differences in pre-post changes isolate the treatment effect.

Demonstration

Evaluate a minimum wage increase implemented in one state by comparing employment changes before and after the policy in that state versus a neighboring state without the increase, controlling for observable time-varying factors.

Misapplication

Applying DiD when treated and control groups had different pre-trends, failing to test for parallel trends, or not accounting for time-varying confounders and clustering of standard errors, leading to biased inference.

Consequence

When parallel trends and related assumptions hold, DiD yields credible causal estimates of average treatment effects over the study period; robustness checks (event studies, placebo tests) and clustered inference strengthen credibility.

Reversal

A randomized controlled trial where assignment to treatment is random and does not rely on trend assumptions, or a simple before-after comparison without a control group which cannot separate treatment from time effects.

Boundary

Requires panel data or repeated cross-sections with comparable groups; not valid if there is anticipation, spillovers, composition changes, or treatment timing heterogeneity without appropriate adjustments.

Semantic Tension

Close to synthetic control and fixed‑effects panel methods; tension appears when choosing between DiD and synthetic control approaches or when staggered adoption requires specialized DiD estimators.

Synthesis

Difference-in-differences attributes differential changes between treated and control groups to an intervention under a common-trends assumption; its credibility hinges on pre-treatment evidence of parallelism and robustness to alternative specifications.