Measurement, Data Display, and Interpretation
Match the measure to the behavior
Prefer direct, continuous measurement: frequency and rate, duration (total or per occurrence), latency to onset, interresponse time, percentage of opportunities, and trials-to-criterion. Discontinuous methods are time sampling: whole interval (underestimates), partial interval (overestimates), and momentary time sampling (estimates duration and usually lands closest to true values).
Accuracy, validity, reliability
- Validity - the data measure what you claim; accuracy - they approximate the true value; reliability - they repeat.
- Believability comes from interobserver agreement plus a technological response definition.
- Operational definitions must be objective, clear, and complete: observable, measurable, with examples and non-examples.
IOA
- Total count = smaller ÷ larger × 100. Mean count-per-interval and exact count-per-interval are stricter; exact count requires identical counts within each interval, so it is always lower.
- Trial-by-trial = agreements ÷ (agreements + disagreements) × 100.
- Total duration and mean duration-per-occurrence for duration data.
- Scored-interval and unscored-interval IOA for time-sampling data.
- Conventionally acceptable is 80–90%, with 90%+ for behaviors targeted for reduction.
Traps
Confusing rate with count (rate needs a time denominator); reporting percent of intervals as percent of session; using whole-interval recording for a brief behavior and concluding it is rare; failing to recheck IOA after observer drift.
Worked example
Thirty responses in a 20-minute session gives 30 ÷ 20 = 1.5 responses per minute. If the target is a 3-second head turn, momentary time sampling estimates it far better than partial-interval recording, which would inflate it toward 100% of intervals.