Single-case designs

A single-case design, also called a single-subject design, evaluates an intervention by comparing phases of behavior within the same case, where a case may be a person, a classroom, or another unit treated as one series. It is not a case study. A case study describes. A single-case design arranges a baseline and a treatment so that a change coinciding with the intervention can be compared with periods in which the intervention was absent or not yet begun. Kazdin’s treatment of these designs is the clinical and applied reference methods courses rely on alongside the survey in Cozby and Bates (Kazdin, 2011; Cozby & Bates, 2020).

The logic is experimental even though there is no random assignment of many people. The researcher controls when the treatment starts and, in the stronger designs, when it stops or when it starts in a second behavior. The dependent variable is measured repeatedly, often every session, so that the level and the trend are visible before anyone claims an effect. A single dramatic day is not a phase. A phase is a run of observations stable enough that a later shift would be visible against it.

These designs grew up in the study of learning and in clinical behavior change, where the question is whether this intervention changes this person’s behavior, not whether a mean difference appears in a group that will never be seen again. That question is legitimate. It is answered by the pattern across phases, not by the richness of the narrative. A moving story with no baseline is a case study, whatever the author calls it.

A reversal design measures a baseline (often labeled A), introduces the treatment (B), and then withdraws it to see whether behavior returns toward baseline. An ABA design is baseline, treatment, baseline. An ABAB design adds the treatment a second time. The second introduction is powerful: if the behavior changes each time the treatment is added and relaxes each time it is removed, history and maturation have a harder time telling the whole story, because those rivals would have to turn on and off with the researcher’s decision. The withdrawal is also an ethical problem when the treatment is helping and taking it away harms the person. A design that requires a cruel withdrawal in order to be elegant is an ethics failure first. ABAB is for situations in which withdrawal is safe, or in which the behavior is not a matter of urgent protection.

A multiple-baseline design avoids withdrawal. The treatment is introduced at different times across several behaviors, several settings, or several people. If each series changes when, and only when, the treatment reaches it, the staggered onsets argue against a single historical event that would have hit all of them together. The baselines have to be independent enough that treating one does not spill into the others. A multiple baseline across school subjects fails if the skill trained in reading also changes the math baseline before math was treated. Spillover is a confound inside a design that was chosen to be clean.

Other variations exist: changing-criterion designs, in which the standard for reinforcement steps up and behavior is asked to follow the steps, and alternating-treatment designs, in which two interventions are toggled rapidly. The family resemblance is the repeated measure and the researcher’s control of the intervention’s timing. A reader who can say which phase changed, when the intervention started, and what else changed that day, can evaluate the design. A reader who only remembers that “they used single-case” cannot.

Internal validity in these designs comes from replication inside the series: the effect turns on and off, or it appears at staggered times. One shift at the moment of treatment is suggestive and is also what history looks like. The stronger patterns are the ones that would be an unlucky coincidence several times. External validity is a separate problem. A clear effect in one child does not estimate the effect in children. Single-case researchers build generality by direct replication with similar cases and by systematic replication that varies the case on purpose. A pile of successful case reports that never includes a failure is not that replication. It is a publication filter.

Statistical analysis of single-case data is a live area and is not required to see a large, immediate, repeated shift. Visual inspection can be abused, especially when the baseline was already trending in the direction the treatment “produced.” A baseline with a slope has to be described as a slope. Drawing a treatment line on a rising curve and declaring victory is the maturation threat in a new costume. The methods obligation is the picture of the whole series, baselines included, not a summary that starts on the first day of treatment.

Single-case designs sit beside group experiments rather than beneath them. Group experiments answer questions about average effects and about moderators that need many people. Single-case designs answer questions about whether an intervention changes a target behavior in the case that will actually receive it. Using a group mean to choose a treatment for one person, or using one person to announce a mean, mixes the questions. The next chapter stays with situations in which the researcher cannot assign people at all, and the comparison has to be built another way.