Key Points
- An interval is scored the moment the behavior first appears, which frees the observer's attention until the next interval begins.
- A brief instance scores the same as a long one, so the method tends to overestimate behavior, an acceptable bias for behaviors a team wants to reduce.
- The result is a percentage of intervals rather than a count, and it compares cleanly only with data scored at the same interval length.
Partial Interval Recording Explained
Some behaviors are too frequent or too brief to count one by one, especially when the person observing is also running a program. Partial interval recording, also written partial-interval recording, handles them by sampling. The observation is divided into short, equal intervals, and the observer scores each one yes if the behavior occurred at any point within it and no if it did not. Once an interval is scored, nothing else that happens during it changes the score, and the observer can turn back to teaching until the next interval starts.
It is one of three interval methods, and the most sensitive. Whole interval recording scores an interval only if the behavior lasted the entire interval, and momentary time sampling looks only at the instant each interval ends. Partial interval recording counts any occurrence at all, which makes it the method least likely to miss a behavior and the most likely to overstate it.
The overstatement is built in. One second of behavior scores the same as a full interval of it, and several separate instances inside one interval score the same as one. That makes partial interval data a poor fit for behaviors a team wants more of, where an inflated number could end teaching too early, and a reasonable fit for behavior reduction targets, where an inflated number errs toward caution. Typical targets include vocal stereotypy, off-task behavior, and other high-rate behaviors without a clear start and end.
Interval length changes the result. Shorter intervals bring the estimate closer to the true amount of behavior, because fewer instances are folded into one score, but they ask more of the observer. Longer intervals are easier to run and inflate the estimate further. The length is chosen for the behavior and the setting, recorded with the data, and held constant from baseline onward, since the same session scored with different interval lengths produces different percentages.
Results are reported as the percentage of intervals in which the behavior occurred. That figure estimates how much of the session involved the behavior rather than how often it happened, and a progress report or payer submission should describe it in those terms instead of converting it into a rate of incidents.
In practice the method depends on a reliable signal for when each interval begins, usually a timer, a vibrating prompt, or the data collection app itself, and on a tight operational definition, because the observer has to decide quickly whether a brief movement counts. Interobserver agreement checks are especially useful here: two observers scoring the same intervals tend to disagree first on borderline instances, and those disagreements show where the definition needs tightening.