Data Collection, Measurement, and Analysis
This domain tests measurement decisions, interobserver agreement computation, and visual analysis within single-case designs. Items frequently give you raw observation records and ask for the correct IOA formula, or give a graph and ask whether experimental control is demonstrated.
Choosing a measure
- Operational definitions must be observable, measurable, and complete, with stated inclusions and exclusions.
- Continuous measures: frequency and rate, duration, latency, interresponse time, percentage of opportunities. Match the measure to the dimension of interest - count for discrete responses, duration for sustained behavior, latency for promptness, IRT for temporal patterning.
- Discontinuous measures: whole-interval recording underestimates, partial-interval overestimates, and momentary time sampling is generally least biased when intervals are short relative to the behavior.
Interobserver agreement
- Total count: smaller ÷ larger × 100 - hides distribution differences.
- Mean count-per-interval and exact count-per-interval capture within-session correspondence.
- Interval-by-interval: agreements ÷ (agreements + disagreements) × 100 - inflated when rates are very high or very low.
- Scored-interval (both observers scored occurrence) suits low-rate behavior; unscored-interval suits high-rate behavior; trial-by-trial suits discrete trials. Accept 80% generally and 90% for new observers or critical targets.
Visual analysis and design logic
- Within condition: level, trend, variability. Between conditions: immediacy of change, overlap, consistency across similar conditions, and replication.
- Reversal (ABAB) demonstrates control but requires reversibility and ethical withdrawal. Multiple baseline across behaviors, settings, or participants is used when reversal is inappropriate. Alternating treatments compares interventions with counterbalancing. Changing criterion evaluates stepwise shaping.
Worked example. Two observers record 20 intervals. Fourteen intervals agree (6 both-scored occurrence, 8 both nonoccurrence); 6 disagree, all intervals where Observer A scored occurrence and B did not. Interval-by-interval IOA = 14/20 = 70%, below criterion. Scored-interval IOA = 6/(6+6) = 50%, which pinpoints the problem: the definition is ambiguous about brief or doubtful instances. Revise the definition and retrain before collecting more data.
Traps that cost marks. Using percentage when no opportunities are defined. Treating a high, stable baseline as evidence of good measurement. Withdrawing treatment for a nonreversible or dangerous behavior. Claiming experimental control from a single A-B comparison.