Quasi-Experimental Designs
Core summary
Quasi-experimental designs test interventions without random assignment. They include before-and-after studies, interrupted time series, and natural experiments. They are used when randomization is impractical, unethical, or impossible, but they are more vulnerable to confounding than true RCTs.
Detailed explanation
Detailed explanation
Sometimes you cannot randomize. A new hospital policy applies to all patients. A national law changes overnight. A natural disaster creates 'accidental' exposure groups. In these situations, quasi-experimental designs offer the next-best evidence. A before-and-after (pre-post) study compares outcomes before and after an intervention is introduced — simple but vulnerable to secular trends (things that would have changed anyway). An interrupted time series (ITS) is stronger: it collects multiple data points before and after the intervention, allowing you to see whether the trend changed at the intervention point. A natural experiment exploits an external event that creates exposure groups (e.g., a law change in one state but not a neighboring state). Difference-in-differences analysis compares the change in outcomes between affected and unaffected groups. The main threats to quasi-experiments are confounding (no randomization to balance groups), maturation (natural changes over time), history (other events coinciding with the intervention), and regression to the mean.
Clinical example
Your hospital introduces a new hand hygiene protocol. You cannot randomize wards (all must adopt the policy). Instead, you use an interrupted time series: track infection rates monthly for 24 months before and 24 months after the protocol. If rates drop sharply at the implementation point and stay low, the evidence is compelling — though not as strong as an RCT.
Research example
The impact of seatbelt laws on road traffic deaths has been studied using interrupted time series and natural experiments. States that enacted laws earlier served as the 'intervention' group while neighboring states served as controls, allowing difference-in-differences analysis.
Knowledge check
Q1. What makes a study 'quasi-experimental' rather than a true experiment?
Q2. Why is an interrupted time series stronger than a simple before-and-after study?
Q3. What is a 'natural experiment'?