Detecting false positives in A-B designs: potential implications for practitioners.

Bigham et al. (2013) ran computer simulations of A-B graphs. They wanted to know how often a stable baseline would fool you into thinking an intervention worked. They tested thousands of fake data sets that had no real treatment effect.

Each graph had a flat, steady baseline. The team then added random bounce and looked for false alarms — times the data crossed a preset decision rule and cried 'success' when nothing actually happened.

False positives stayed below two percent when the baseline showed no trend. In plain words, a flat, calm baseline rarely cried wolf. The design looked safe for everyday clinical use.

The result held across different amounts of bounce and different decision rules. Stability, not perfection, was the key guardrail.

Suzuki et al. (2023) extend the same idea to N-of-1 trend analyses. They also found that lower baseline variability boosts accuracy. Both papers tell you to eyeball a steady baseline before you trust any later change.

Lanovaz et al. (2019) add a twist. They show that when an effect is large and clear in the first A-B segment, it usually repeats in a full reversal design. K et al. give you confidence in simple A-B; Lanovaz et al. say you can often stop there if the jump is obvious.

Smith et al. (2022) widen the lens. They argue that non-concurrent multiple-baseline designs can be just as valid. Taken together, the four papers give you a menu: stable A-B, big visible change, or well-spaced baselines can all justify treatment without extra phases.

You can feel safer running a short A-B study when the baseline is flat and steady. Check for drift first; if you see none, the chance of a false alarm is tiny. This saves time in busy clinics and keeps parents from chasing ghost improvements. Pair the rule with Lanovaz’s big-effect shortcut and you will know when to push ahead or add more phases.