Assessment and Data Collection
Getting the measure right
An operational definition must be observable, measurable, clear, and objective: describe what the behavior looks like, not why it happens. Match the dimension to the behavior's shape. Use count/rate for discrete responses, duration for how long a behavior lasts, latency for time to onset (do not confuse it with inter-response time, the gap between responses), percentage for accuracy given a known number of opportunities, and permanent products when behavior leaves a record.
Observation and sampling
Continuous recording is the gold standard but often impractical. Whole-interval recording underestimates behavior (an interval counts only if behavior lasts throughout); partial-interval overestimates; momentary time sampling is usually the best approximation and is close to unbiased with short intervals. Scatterplots reveal when behavior occurs, not why.
Assessment logic
Functional behavior assessment can be indirect (interviews, rating scales), descriptive (ABC recording), or experimental (functional analysis). Only a functional analysis demonstrates a causal relation; descriptive data suggest hypotheses. Preference assessments identify potential reinforcers; a stimulus preference assessment does not prove a stimulus functions as a reinforcer, a reinforcer assessment does.
Interobserver agreement
For total count, IOA = (smaller count ÷ larger count) × 100. For interval-by-interval, IOA = agreements ÷ (agreements + disagreements) × 100. Acceptable IOA is typically 80% (90% for discrete, high-stakes behaviors). Low IOA means retrain observers or refine definitions before trusting the data, not that you change the intervention.
Worked example
Two observers record 20 and 25 instances. Total-count IOA = 20/25 × 100 = 80%, which is at the floor. Check the operational definition and retrain the observers before drawing conclusions about the intervention.