The quality of ABC data begins before the first observation
An operational definition should allow trained observers to agree on whether the response occurred and when it began and ended. Avoid inferred states such as “frustrated,” “noncompliant,” or “seeking.” Select a measurement dimension that fits the response: frequency, rate, duration, latency, or another defensible unit.
Record events, not categories too early
Write what was presented, removed, said, or changed. Record the response topography. Then record the immediate consequences and whether they were contingent. Note relevant setting events or motivating operations separately when they can be measured. Also record opportunities: ten episodes after demands and one after denied access cannot be compared as raw counts alone.
Descriptive association is not experimental function
ABC data are vulnerable to biased sampling, observer reactivity, inconsistent definitions, and naturally correlated events. A consequence may frequently follow behaviour because adults respond to risk—not because that consequence maintains it. Conditional probabilities can improve summaries, but they do not create experimental control.
The correct product of descriptive assessment is a testable hypothesis and a better next decision—not certainty written in stronger language.
Strengthen confidence in the data
- Train observers with examples and nonexamples.
- Sample across people, times, routines, and low-risk comparison periods.
- Measure interobserver agreement for a meaningful proportion of observations.
- Graph relevant dimensions and examine within-session patterns.
- Check whether the observation system captures communication responses and successful participation, not only target behaviour.
Decide what assessment comes next
The risk, severity, clarity of patterns, treatment history, feasibility, and practitioner competence shape the next step. Experimental functional analysis involves planned manipulation of environmental variables and should be individualized and conducted only with appropriate expertise and safety safeguards. In some cases additional direct observation or a carefully designed analysis is needed; in others a low-risk hypothesis-informed intervention with close measurement may be reasonable.
