Behavior analysts mostly rely on single-case designs, but group designs answer some questions better. Knowing what each design is built to show helps you read research critically, explain your evidence to families and administrators, and pick the right tool for the question.
Why it matters
Every intervention you recommend rests on some kind of evidence. Some of it comes from studies with three or four participants measured dozens of times. Some comes from studies with hundreds of participants measured once before and once after treatment. These designs answer different questions, so neither one is simply "better." As a BCBA, you need to judge whether a study's design fits the claim being made from it.
That means being able to name a strength or limitation of each approach, recognize which design fits a given question, and avoid confusing a small number of participants with weak experimental control.
Key ideas
- Single-case designs study the individual. Each participant's behavior is measured repeatedly under baseline and intervention conditions, and each participant serves as their own control. A study can include several participants, but each one's data are analyzed separately.
- Group designs study the aggregate. Participants are usually assigned (ideally at random) to a treatment group or a comparison group, and the groups' average outcomes are compared, often with inferential statistics.
- Repeated measurement is a major strength of single-case work. Because you see the data session by session, you can spot variability, extend a baseline that is not yet stable, or adjust an intervention that is not working. Group studies often measure each person only a few times, so these within-person patterns stay hidden.
- Averages can hide individuals. A group mean can improve even when some participants got worse or did not change. The average may not describe any real person in the study. Differences between participants (intersubject variability) are often treated as error rather than as something to explain.
- Group designs suit actuarial questions. When the question is about the overall effect of a program or policy across a large population, such as a district-wide initiative, a well-run group comparison with random assignment is a good fit.
- Generality is built differently. Group studies argue for generality through large, representative samples. Single-case research builds it through replication: repeating the effect with the same procedures (direct replication) and with deliberate changes in participants, settings, or behaviors (systematic replication).
Definitions
- Single-case design: an experimental design in which each participant's behavior is measured repeatedly, and that person's performance in one condition (such as baseline) is compared with their own performance in another.
- Group design: an experimental design that compares outcomes between groups of participants, usually by comparing group averages.
- Own control: the participant's baseline performance is the standard against which intervention performance is compared.
- Intersubject variability: differences in responding from one participant to another.
- Actuarial question: a question about the average or overall effect of a variable across many people, rather than its effect on a specific person.
Examples from a session
- A clinic evaluation. Maya, a BCBA, wants to know whether a visual schedule reduces transition-related crying for Theo, age 4. She measures crying every session in baseline and then with the schedule, reversing and reintroducing it. Theo's own graph shows whether the schedule worked for him, and Maya can respond to the data in real time.
- A district question. A school board asks whether a new classroom management program lowers suspension rates across 30 middle schools. The question is about overall impact on many students, so a group comparison with schools randomly assigned to the program or to business as usual fits the question better.
- Reading a group study. Dr. Ruiz reads a study reporting that a parent-training program reduced problem behavior by 25% on average. Before recommending it for the Nguyen family, she checks whether individual results were reported, because some families might have seen no change at all.
- Building generality. A research team shows a self-monitoring intervention increases on-task behavior for three high school students. Other teams later replicate it with adults at a supported employment site and with fourth graders. That chain of replications, not a single large sample, supports its generality.
- Non-example of a group design. A study reports results for eight children, but each child's graph is analyzed against that child's own baseline, with no group averages compared. Several participants do not make it a group design; it is still single-case.
Common mistakes / exam traps
- Equating few participants with weak internal validity. A single-case study with three participants can show strong experimental control through prediction, verification, and replication. The small number mainly limits what you can say about generality until the effect is replicated.
- Calling a study with several participants a group design. What decides it is the analysis: each person against their own baseline (single-case) or group averages compared (group).
- Forgetting the costs of group designs. They need large samples, often withhold treatment from a comparison group, and offer little room to change course mid-study.
- Treating the designs as rivals. The best evidence often combines both: single-case research to develop and refine an intervention, and group research to estimate its effect across a population.
Exam trap
If the question is about one learner's behavior or about refining a procedure, think single-case design. If it is about the average effect across a large population, think group design. Small N limits generality, not experimental control.
Exam trap
Group average vs. individual effect: a significant improvement in the group mean does not show that every participant, or this particular learner, improved. Only individual data (repeated measures for that person) answer "did it work for her?" Statistical significance is also not the same as socially significant change.
Quick check
Which statement best describes a limitation of relying on a group average to judge whether an intervention works?