Studies in which subjects are randomly assigned to treatments (and controls) and outcomes are then assessed.

Randomly divide sample of allergy sufferers into two groups. One gets new allergy medicine, the other gets placebo. Then symptom severity is measured for each person. If no statistical difference between the two groups, we conclude the medicine doesn't work

Suppose we want to examine effects of teacher expectations on student learning. We randomly assign each student a fake test score and inform the teacher. Then we see how well the students do on a real test at a later time

Random assignment means that statistical controls are less needed. Aside from simulations, experiments give us best shot at establishing causality.

For practical purposes, we basically accept results of experiments as indicative of causality

Ethical considerations prevent us from using experimental method in huge number of cases. Can't randomly assign people to smoking vs non-smoking groups.

Often have to be conducted in lab, limiting realism and therefore generalizability

Often a pain to construct

"Treatment" often refers to some kind of active intervention, such applying some therapy. But doesn't have to be as the complex example shows.

Complex example also shows that the independent variable does not have to be categorical

Data do not have to be explicitly collected at multiple points in time, but since the treatment/control assignment is done before measuring outcomes, in effect this is longitudinal

Studies in which subjects are non-randomly assigned to treatment and control groups

Studies in which pre and post measurements are taken.

Randomly divide sample of allergy sufferers into two groups. One gets new allergy medicine, the other gets placebo. Then symptom severity is measured for each person. If no statistical difference between the two groups, we conclude the medicine doesn't work.

Suppose we want to examine effects of teacher expectations on student learning. We randomly assign each student a fake test score and inform the teacher. Then we see how well the students do on a real test at a later time

Easier to implement than real experiments

Ethical considerations prevent us from using experimental method in huge number of cases. Can't randomly assign people to smoking vs non-smoking groups.

Often have to be conducted in lab, limiting realism and therefore generalizability

Often a pain to construct

"Treatment" often refers to some kind of active intervention, such applying some therapy. But doesn't have to be as the complex example shows.

Complex example also shows that the independent variable does not have to be categorical

Data do not have to be explicitly collected at multiple points in time, but since the treatment/control assignment is done before measuring outcomes, in effect this is longitudinal