Single-case designs are how behavior analysts show that an intervention, and not something else, changed a person's behavior. They are not defined by how many people are in a study. They are defined by a particular way of comparing conditions and a particular logic for drawing conclusions.
Why it matters
Every time you write "the intervention was effective" in a progress report, you are making a causal claim. To support it, you need more than a before-and-after picture. You need data that let you predict what would have happened without treatment, check that prediction, and show the effect again.
Practice items on this topic often describe part of a graph (a stable baseline, a withdrawal phase, a tier that stays flat) and ask what that piece contributes. If you can name the job each phase does, you can answer them.
Key ideas
- Each participant serves as their own control. Behavior under the intervention is compared with the same person's behavior under baseline, not with a different group of people.
- The dependent variable is measured repeatedly. Many measures within each condition show the level, trend, and variability of responding, so you can see whether behavior is stable and when it changes.
- Steady state responding makes prediction possible. When baseline data show little variability (a flat level or a consistent trend), you can reasonably say where the data path would go if nothing changed.
- Baseline logic has three elements. Prediction, verification, and replication work together to show a functional relation between the independent variable and behavior.
- Data are analyzed for each individual. Effects are judged by visual analysis of each person's graph rather than by averaging people together.
Definitions
- Single-case design: an experimental arrangement in which each participant's behavior is measured repeatedly across conditions and compared with their own baseline. A study can include one participant or several.
- Repeated measurement: measuring the dependent variable many times within each condition, so changes can be detected over time.
- Steady state responding: a pattern of responding with relatively little variation in level or trend over a period of time.
- Prediction: the expected outcome of a future measurement, based on the steady state data already collected, if conditions stay the same.
- Verification: showing that the baseline level would have continued had the intervention not been introduced, which strengthens the original prediction.
- Replication: repeating the effect of the independent variable, for example by reintroducing treatment and seeing behavior change again.
Examples from a session
Baseline logic in three settings
- Home, reversal design: Ines, 12, completes about 20% of her homework across eight stable baseline sessions. That stable level is the prediction: without a change, completion should stay near 20%. With a checklist and praise, completion rises to about 85%. When the checklist is withdrawn, completion falls back near 20%, which verifies the prediction. When the checklist returns and completion climbs again, the effect is replicated.
- Clinic, multiple baseline across behaviors: Omar, 5, has three skills in baseline. Teaching begins on the first skill only. While it improves, the two untaught skills stay flat, which verifies the prediction for the first skill. When teaching starts on the second skill and it improves, the effect is replicated.
- Adult day program, the weak design: Staff measure greetings by Grace, 41, for a week, start a peer-buddy system, and see greetings rise. This A-B comparison gives a prediction and shows a change, but nothing verifies the prediction or replicates the effect, so other events could explain the change.
A quick habit: for any phase on a graph, ask which job it does. Does it set up a prediction, check one, or repeat the effect?
Common mistakes / exam traps
- Thinking "single-case" means one participant. Many single-case studies have several participants. The defining feature is the within-person comparison across conditions.
- Treating an A-B design as proof. A change after treatment starts is a start, but without verification and replication, history, maturation, and other confounds remain possible.
- Changing conditions too soon. If baseline is highly variable, you cannot make a confident prediction. If it is already trending in the direction you expect treatment to produce, a later change will be hard to credit to the intervention. Extend baseline or look for sources of variability first.
- Confusing replication with interobserver agreement. IOA checks the believability of the measurement. Replication concerns the effect of the independent variable.
Exam trap
The trap: verification vs. replication. Verification checks the baseline prediction; replication repeats the intervention's effect. In a reversal design, the return to baseline levels when treatment is withdrawn is verification. Reinstating treatment and seeing the change again is replication. In a multiple baseline, stable untreated tiers verify, and a change in the next tier after treatment begins there replicates.
Exam trap
The trap: maintenance vs. replication after a failed reversal. If behavior stays high when treatment is withdrawn (for example, because peers now reinforce it), the intervention has not failed; the baseline prediction is simply unverified. A level that holds is maintenance, not replication, because replication needs the effect to appear again after a change in conditions. For behavior unlikely to reverse, verify with untreated tiers in a multiple baseline.
Quick check
Theo, a BCBA, graphs out-of-seat behavior for Amara, 9, at school. Baseline is stable at about 12 per hour, then drops to about 3 per hour after a token board starts. Theo stops collecting data there. What does this A-B graph lack before he can claim a functional relation?