Measurement, Data Display, and Interpretation
Choosing a measure
Measurement begins with a measurable dimension: count, rate, duration, latency, interresponse time, or percentage of opportunities. Continuous measurement captures every instance; discontinuous methods (partial- and whole-interval recording, momentary time sampling) sample behavior. Each method carries known bias: partial-interval recording overestimates high-rate behavior, whole-interval underestimates it, and momentary time sampling is generally the most accurate of the three for estimating duration.
Validity and reliability
Accuracy is agreement with the true value; reliability is consistency. Interobserver agreement is computed differently by design: total count, exact count, mean count-per-interval, trial-by-trial, and total duration IOA. Know which formula fits which data set-mixing them is a common trap. Treatment integrity data describe whether the independent variable was delivered as planned, and without it a null result is uninterpretable.
Displaying and reading data
Line graphs suit repeated measures over time; cumulative records show rate as slope; bar graphs compare conditions without a time dimension; scatterplots reveal temporal patterns across the day. When interpreting, look for level, trend, and variability, and judge immediacy of effect, overlap, and consistency across tiers. Never call a change significant from visual inspection of a single data point.
Rule of thumb
Name the dimension before choosing the tool. If a behavior is discrete and frequent, count or rate. If it is prolonged, use duration. If it occurs in bursts, consider IRT.
Worked example
Two observers record 10 and 12 responses in a session; smaller over larger gives 83% total count IOA. If they instead record interval-by-interval, use interval-by-interval IOA, not the total count formula.