Quasi-Experimental and Non-Experimental Designs

When You Can't Randomly Assign

Not every research question can be studied with a true experiment. Sometimes the variable of interest cannot ethically or practically be randomly assigned — researchers cannot randomly assign people to smoke, to experience trauma, or to be born male or female. Quasi-experimental designs compare groups that differ on some variable of interest (like a naturally occurring event, a pre-existing condition, or group membership) WITHOUT random assignment. A natural experiment, comparing regions before and after a policy change that researchers did not control, is a classic quasi-experimental design — informative, but weaker than a true experiment at ruling out confounds, since the groups may have differed in other ways before the study even began.

Non-experimental designs go further still, involving no manipulation of any variable at all. Correlational studies simply measure the relationship between two variables as they naturally occur — for instance, measuring whether hours of sleep and reported mood are related. Correlational designs can reveal that two variables are associated, but critically cannot establish causation: a correlation between sleep and mood could mean poor sleep causes low mood, low mood causes poor sleep, or some third variable (like chronic stress) causes both. Case studies examine a single individual or small group in great depth, often used for rare phenomena that would be impossible to study with a large sample (like an unusual brain injury). Archival research analyzes existing records or data collected for another purpose entirely.

Common pitfall: treating a strong correlation as proof of causation. Even a very large, statistically robust correlation between two variables never, by itself, tells you which one (if either) causes the other — establishing that requires a true experiment with random assignment, or very careful additional evidence ruling out alternative explanations.