External validity

External validity asks where a finding travels

A study may show a convincing relation for one participant, response, implementer, and setting. External validity asks what supports extending that finding to another combination.

The dimensions should be named. “The intervention generalizes” can mean:

• another person shows a similar effect
• another response class changes
• the effect occurs at home, school, clinic, or community sites
• a caregiver or teacher implements the procedure successfully
• different materials or schedules produce similar results
• performance maintains weeks or months later

Evidence on one dimension gives limited information about another. Success with three people in one clinic does not automatically show that caregivers can implement the procedure at home.

Internal and external validity solve different problems

Internal validity concerns whether the planned independent variable plausibly produced the observed change in the studied conditions. External validity concerns the supported reach of that finding.

A study with weak internal validity provides an uncertain effect to generalize. A strong demonstration with one case can support a narrow causal conclusion while leaving generality open. Research programs need both clear demonstrations and planned replication.

The BACB BCBA Test Content Outline distinguishes internal from external validity and includes single-case prediction, verification, replication, and interpretation. It is examination content, not a population-inference rule.

Replication builds generality step by step

Direct replication repeats closely similar procedures with another participant or instance. It tests whether the effect occurs again under deliberately similar conditions.

Systematic replication changes a planned feature, such as setting, implementer, response, age group, or dosage. It tests whether the relation extends across that change.

A useful replication series reports both successes and failures. Selective reporting can make generality look broader than the complete record supports. Case descriptions, context, implementation, and outcome definitions must be detailed enough to compare.

Single-case generality differs from population estimation

Single-case designs can establish replicated functional relations across carefully studied cases. They usually do not estimate prevalence or an average treatment effect for a defined population through random sampling.

Population claims require attention to sampling, representativeness, heterogeneity, and the statistical question. A diverse series of single-case replications can strengthen confidence and identify boundaries, yet the conclusion should describe the evidence actually gathered.

The WWC Single-Case Design Technical Documentation focuses heavily on design standards, internal validity, visual analysis, and evidence of a relation. Its discussion of combining studies does not turn a small case series into a representative population sample.

A fictional three-learner replication

A team evaluates an accessible choice-making support with three learners. Each learner has six probes in a clinic and six probes in a familiar community setting, for 36 total probes.

Independent choices occur in 15 of 18 clinic probes and 12 of 18 community probes. Each learner improves relative to their own baseline in both settings. Procedural integrity meets criterion in 34 of 36 probes.

The pattern supports generality across these three people and two defined settings with trained implementers. It does not estimate how often the effect would occur across all autistic people, all ABA programs, or unfamiliar implementers. Two integrity misses and the smaller community effect remain visible.

The next systematic replication tests caregiver implementation with consent, coaching, and its own fidelity measure. That question needs new evidence.

Plan generality before the study

Identify the dimensions that matter for use:

1. people and eligibility features
2. target responses and measurement
3. settings, materials, and schedules
4. implementers and training
5. maintenance period
6. cultural, language, communication, health, and access context

Build opportunities to test those dimensions. If a support will be used in community settings, clinic-only data leave an important gap.

The WWC key-criteria resource summarizes criteria used in single-case reviews. Standards ratings speak to research evidence under a review framework; they do not by themselves establish fit for a particular person or service.

Report boundaries with the result

State who participated, what changed, where, when, under whose implementation, with what fidelity, and across how many opportunities. Describe exclusions and attrition. Report variation among cases rather than only a pooled percentage.

Use “replicated across these cases and settings” when that is the evidence. Reserve broader language for evidence designed to support it. Clear limits make a finding more useful because readers can see what still needs testing.

Track failures with the same care as successes. A boundary case may reveal which setting feature, skill prerequisite, support, or implementation condition deserves the next systematic replication.