How should you conduct a quasi-experimental evaluation when randomization isn't feasible?

• A. Conduct a simple pre-post survey without comparison.
• B. Use only qualitative interviews without quantitative data.
• C. Use designs such as non-equivalent control groups or interrupted time series, apply matching, adjust for confounders, and pre-register analyses when possible.
• D. Randomly assign participants to control and intervention in a blinded fashion.

Answer: (C)

When randomization isn’t feasible, the goal is to approximate the counterfactual and distinguish the intervention’s effects from other factors by using robust quasi-experimental designs. Designs like non-equivalent control groups and interrupted time series provide a comparison or a clear pre/post pattern that helps attribute observed changes to the intervention rather than to aging, history, or other trends.

Non-equivalent control groups involve comparing the treated group to a similar untreated group chosen to match on key characteristics. Because the groups aren’t randomized, matching on observed covariates and applying statistical adjustments to account for remaining differences helps reduce selection bias and make the comparison more credible.

Interrupted time series looks at multiple measurements before and after the intervention to detect whether there is a real, abrupt or gradual change in the outcome that coincides with the implementation, beyond what would be expected from prior trends. This approach strengths causal inference by showing changes over time rather than a single pre/post snapshot.

Adjusting for confounders using statistical methods (such as propensity scores or regression adjustment) further helps isolate the intervention’s effect by accounting for factors that could influence outcomes. Pre-registering analyses when possible protects against analytic flexibility and enhances credibility by committing to planned methods and outcomes in advance.

In contrast, a simple pre-post survey without comparison lacks a guard against secular or external influences; relying only on qualitative interviews misses the quantitative signal needed to estimate effect sizes; and randomly assigning participants to groups in a blinded fashion would constitute a true experimental design, which isn’t permissible in a quasi-experimental context.