When to use interval recording
Interval recording fits behaviors that are hard to count accurately: high-rate behaviors such as vocal or motor stereotypy, behaviors without a clear beginning and end such as being on task or engaged, and situations where the observer is also teaching and cannot track every instance. It produces an estimate of how much of the time the behavior occurs, which is why it is called discontinuous measurement.
It is the wrong tool when a true count or a true duration is available. For a discrete behavior that happens a few times a session, frequency recording is simpler and more accurate; when the length of each episode is the clinical concern, duration recording measures it directly. The eight ABA measurement systems guide compares all the options.
Partial interval, whole interval, and momentary time sampling
All three methods divide an observation into equal intervals and score each one. What changes is the rule for scoring an interval as an occurrence, and with it the direction of the error.
| Method | An interval is scored when | Tends to | Often used for |
|---|---|---|---|
| Partial interval | The behavior occurs at any point during the interval, even once and briefly. | Overestimate how much of the time the behavior occurs, and undercount high rates, since several occurrences in one interval score once. | Behaviors targeted for reduction, and brief or high-rate behaviors. |
| Whole interval | The behavior continues throughout the entire interval. | Underestimate how much of the time the behavior occurs. | Behaviors targeted for increase, such as on-task behavior or engagement. |
| Momentary time sampling | The behavior is occurring at the moment the interval ends. | Overestimate or underestimate, with no consistent direction; the error shrinks as intervals get shorter. | Continuous, longer-lasting behaviors, observing a group, or observing while teaching. |
Why the direction of the error matters
Because partial interval recording overestimates, it is the conservative choice for a behavior a program is trying to reduce: the estimate errs high, so a reduction shown on the graph is unlikely to be an artifact of the method. Whole interval recording is the conservative choice for a behavior a program is trying to increase, for the same reason in reverse. Using the opposite pairing makes progress look better than it is.
Momentary time sampling has no built-in direction, which makes it neither conservative nor optimistic; its accuracy depends mostly on the interval length relative to the behavior. It also has a practical advantage: the observer only needs to look at the moment each interval ends, which is why it suits a clinician who is running a session at the same time.
Choosing an interval length
Shorter intervals produce estimates closer to the true value, and they demand more of the observer. Partial and whole interval recording are commonly run with intervals of around 10 seconds, sometimes split into an observe period followed by a short record period so the observer is never writing while the behavior could be occurring. Momentary time sampling is often run with longer intervals, but accuracy falls as the interval grows relative to the length of a typical episode.
Whatever length is chosen, keep it fixed for the life of the target, along with the length of the observation. Changing either changes the size and sometimes the direction of the error, so sessions recorded under different settings cannot be compared on the same graph. A timer that cues each interval, by vibration or a quiet tone, keeps scoring consistent across observers.
Scoring and graphing interval data
The session value is the percent of intervals scored: intervals in which the behavior was scored, divided by the total intervals observed, times 100. A 10-minute observation in 10-second intervals has 60 intervals; if the behavior was scored in 18, the session value is 30%. Intervals that could not be observed, because the learner left the room or the observer was interrupted, should be marked as not observed and left out of the denominator, not scored as non-occurrences.
Label the result as what it is. Percent of intervals is not a rate or a count, and a graph labeled "frequency" invites a reader to compare it with count data it cannot be compared with. When one target's data mix interval data and counts, the graph stops meaning anything.
Observing a group: the planned activity check
The planned activity check, or PLACHECK, is a form of momentary time sampling for groups. At the end of each interval the observer counts how many members of the group are engaged in the target behavior, such as participating in the activity, and records that number as a percentage of the group present. It answers questions about a classroom or a group session as a whole, rather than about one learner.
Checking reliability on interval data
Interval data lend themselves to interval-by-interval agreement: two observers score the same intervals independently, and agreement is the number of intervals scored identically divided by the total, times 100. This is the exact agreement method described in the interobserver agreement guide, and it is stricter than comparing the two observers' session percentages.
For behaviors that occur in very few intervals or in almost all of them, interval-by-interval agreement can look high simply because both observers scored most intervals the same way by default. Scored-interval agreement, which counts only intervals at least one observer scored as an occurrence, and unscored-interval agreement, which does the reverse, are the field's answers to that problem for low-rate and high-rate behaviors respectively.
How Cadence records interval data
Interval recording is one of the 8 measurement systems built into Cadence, covering partial interval, whole interval, and momentary time sampling. Each record keeps its method, its interval length, and the score for every interval, so the session's value - the percent of scored intervals containing the behavior - can be reconstructed exactly.
Because the per-interval scores are kept, a reliability check can compare two observers interval by interval with exact agreement IOA rather than comparing session totals.