Measurement
Measurement foundations
Measurement begins with an operational definition: observable, measurable, and complete enough that two observers could agree. Continuous methods capture every response: frequency/event count, rate (count divided by time), duration, latency (time from instruction or SD to response onset), and inter-response time. Discontinuous methods sample: whole-interval recording scores an interval only if behavior occurs throughout, so it usually underestimates; partial-interval scores any occurrence, so it usually overestimates; momentary time sampling checks whether behavior is occurring at the instant of each interval. Permanent products record tangible outcomes. Graph data to inspect level, trend, and variability, and compare baseline with intervention. Interobserver agreement (IOA) supports believability: total-count, mean-count-per-interval, exact-interval, and trial-by-trial formulas. Treatment integrity data show whether procedures were run as planned. Traps: choosing duration when the question asks how long until behavior starts (latency); reporting frequency when session lengths differ (use rate); assuming partial-interval equals true duration; and treating IOA as a measure of the client. Also watch for undefined response classes and missed recording opportunities. Worked example: In a 10-minute session a learner emits 20 hand raises. Rate = 20 ÷ 10 = 2 per minute. If a 20-interval partial-interval record shows 12 intervals with any hand raise, report 12/20 = 60% of intervals, not 20 responses. Rule of thumb: define the behavior and its dimension first, then select the measurement system that captures that dimension with the least bias.