Experimental Design
Baseline logic
Every design must demonstrate prediction (baseline would have continued), verification (a return to baseline confirms it), and replication (the effect occurs again). A functional relation requires that the independent variable, not an extraneous variable, produced the change - and that requires at least two demonstrations.
The designs
- Withdrawal/reversal (ABAB). Strongest for reversible behavior; ethically questionable when behavior is dangerous or the learning is irreversible.
- Multiple baseline across subjects, behaviors, or settings. Stagger baselines; each tier must change when its intervention begins. Traps: baselines already trending, or behaviors that are not independent and covary.
- Multiple probe. A variant that avoids prolonged baseline measurement of dangerous behavior.
- Alternating treatments (multielement). Rapid comparison of two or more interventions; vulnerable to carryover and sequence effects.
- Changing criterion. For stepwise improvement; behavior must track each criterion change.
- Component and parametric analyses. Isolate which part of a package works and at what dose.
Threats to internal validity
Maturation, history, testing, instrumentation (observer drift), regression to the mean, attrition, and sequence effects. Each maps onto a control: concurrent baselines handle history and maturation, IOA and calibration handle instrumentation, stable baselines handle regression to the mean.
Rule of thumb
Ask three questions of any vignette: (1) Is there a baseline? (2) Is there at least one replication? (3) Does the change line up with the manipulation? If any answer is no, a functional relation cannot be claimed. Treat external validity as a separate claim about generality across settings, people, and time.