Duration in ABA is the amount of time a defined behavior lasts, measured from the behavior’s onset to its offset. If a learner begins crying at 12 seconds and stops at 27 seconds, the duration is 15 seconds. Duration answers the question, “How long did the behavior occur?” It is different from frequency, which counts episodes, and latency, which measures how long it takes a response to begin.
Duration is useful when the length of an episode matters. A behavior may happen only once but last for 20 minutes, or happen many times for only a few seconds each. A clear operational definition, consistent timing boundaries, and a recording rule that fits the behavior are necessary for the number to be meaningful.
What duration measures in ABA
Duration is a continuous measurement dimension because the observer records the time span of an eligible behavior episode as it occurs. The episode has a defined onset, a defined offset, and a measurable interval between them. The observer is not simply estimating whether the behavior occurred at some point; the observer is timing how long it lasted.
The behavior definition must make the boundaries observable. “Noncompliance” is too broad if two observers cannot agree on when it starts or ends. A more usable definition might state that noncompliance begins when the learner does not begin the specified task within 10 seconds of an instruction and ends when the learner starts the task or the instruction is withdrawn. For vocal behavior, the definition might specify whether a pause of a certain length separates one episode from the next.
Duration can describe a single episode, the total duration across a session, or the average duration of eligible episodes. Keep those summaries separate. “The episode lasted 15 seconds” is not the same claim as “the learner engaged in the behavior for 15% of a 10-minute observation.” The denominator and the unit must be visible whenever a summary is reported.
Shorter duration is not automatically better. For a learning target such as sustained independent engagement, a longer duration may indicate improvement. For an unsafe behavior, a shorter episode may be the goal. Interpret duration against the target, context, accuracy, intensity, and treatment objective.
Duration formula and a 15-second example
The basic duration formula is:
Duration = offset time − onset time
Suppose a defined behavior begins at 12 seconds and ends at 27 seconds from the start of the observation. The calculation is 27 s − 12 s = 15 s. The 27-second timestamp is not the duration because it includes the time before the behavior started. The 12-second timestamp is not the duration either; it is the onset boundary used in the subtraction.
Use the same time unit throughout the session. If the data system records minutes and seconds, convert them consistently before calculating. For example, an episode from 2 minutes 10 seconds to 2 minutes 42 seconds lasts 32 seconds. Do not subtract the minute and second fields independently without borrowing when the seconds value is smaller.
Worked duration data example
Imagine that the target is a clearly defined episode of off-task behavior. The observer records every episode during one 10-minute observation. Each row contains the onset and offset timestamps so another observer can reproduce the calculation.
| Episode | Onset | Offset | Duration | Calculation |
|---|---|---|---|---|
| 1 | 0:45 | 0:57 | 12 s | 57 − 45 |
| 2 | 2:10 | 2:42 | 32 s | 162 − 130 |
| 3 | 5:05 | 5:20 | 15 s | 320 − 305 |
| 4 | 8:00 | 8:41 | 41 s | 521 − 480 |
The total duration is 12 + 32 + 15 + 41 = 100 seconds, or 1 minute 40 seconds. The mean duration per episode is 100 ÷ 4 = 25 seconds. The behavior occupied 100 seconds of a 600-second observation, so total duration as a percentage of the observation was 100 ÷ 600 × 100 = 16.7%.
Those summaries answer different questions. The mean describes a typical episode, while the percentage describes how much of the observation was occupied by the behavior. The raw episodes also show variability: the fourth episode was much longer than the first. A graph of total duration or mean duration across sessions can reveal a treatment trend that a count alone would miss.
How to collect duration data step by step
Duration recording works best when the observer decides the timing rules before the session begins. A stopwatch, tablet data system, or event-timing application can be used, but the tool cannot fix an unclear behavior definition.
- Define the episode. State exactly what starts the behavior and what ends it. Include rules for pauses, repeated topographies, and overlapping behavior.
- Set the observation window. Record the session start and end times, setting, materials, and any planned interruption that changes the opportunity to respond.
- Start timing at onset. Begin the timer when the observable onset criterion is met, not when the instruction is delivered unless the target is defined that way.
- Stop timing at offset. End the episode when the offset criterion is met. Record the timestamp or elapsed duration using the agreed precision.
- Repeat for every eligible episode. Do not record only the first or most severe episode if the plan calls for all episodes.
- Summarize with the correct denominator. Report total duration, mean episode duration, or percentage of observation time as separate measures.
If the behavior is still occurring when the observation ends, record the episode as ongoing and document the boundary rule. Do not silently treat the end of the observation as the natural offset unless the protocol explicitly defines it that way. If behavior never occurs during a valid observation, the total duration may be zero, but the observation opportunity and data-collection conditions should still be documented.
Duration vs. frequency, rate, latency, and IRT
Choose the measurement dimension that matches the clinical or exam question. A behavior may be measured with more than one dimension, but the dimensions should not be treated as interchangeable.
| Measure | Question answered | Timing boundaries | Best fit |
|---|---|---|---|
| Duration | How long did one episode or all episodes last? | Onset to offset | Episodes whose length matters |
| Frequency | How many times did it occur? | Each eligible occurrence | Discrete, countable responses |
| Rate | How often did it occur per unit of time? | Count divided by observation time | Comparing unequal observation periods |
| Latency | How long until the response begins? | Antecedent or instruction to response onset | Task initiation or response delay |
| IRT | How much time passed between responses? | Prior response offset to next response onset | Spacing or fluency of repeated responses |
For related examples, review the site’s frequency recording guide, latency guide, and IRT guide. The continuous measurement guide explains how these dimensions fit into a broader measurement plan.
Continuous duration recording vs. interval sampling
Continuous duration recording times the actual episode boundaries. It preserves more information, but it may be difficult when the behavior is frequent, observers are responsible for several learners, or exact onset and offset timing is not feasible.
Interval procedures answer different questions. Partial-interval recording scores whether the behavior occurred at any point in an interval; whole-interval recording scores whether it occurred for the entire interval; momentary time sampling scores whether it is occurring at a selected check moment. These methods can estimate patterns efficiently, but they do not produce the same value as continuous duration.
Do not label the percentage of intervals as “duration” unless the protocol explicitly defines that derived measure. If a team changes from continuous duration to interval sampling, document the change and avoid treating the resulting numbers as directly equivalent. The method should be selected based on the decision, behavior pattern, observer capacity, and acceptable measurement error.
Data quality and treatment decisions
Reliable duration data depend on shared boundaries. Observers should practice examples and nonexamples, use the same timer precision, and record prompts, interruptions, and setting events that may change episode length. When an episode is ambiguous, the data sheet should preserve the uncertainty or follow the prewritten rule rather than silently guessing.
For total-duration observer agreement, one common summary is smaller total duration ÷ larger total duration × 100. That calculation is meaningful only when both observers used the same operational definition, observation window, and timing unit. For more detailed decisions, compare episode-by-episode agreement or graph both observers’ data rather than relying on one percentage.
Use the graph that matches the question: total duration for time occupied, mean duration for episode length, or percentage of observation time when session lengths differ. Pair duration with frequency or rate when a treatment could change both the number and length of episodes. For a replacement skill, also check accuracy and independence so a longer response does not look like improvement when it is actually prompted or incorrect.
BCBA exam traps and quick review
- Duration is not frequency. One long episode can have a low frequency but a high total duration.
- Duration is not latency. Latency ends at response onset; duration begins at response onset and ends at response offset.
- Duration is not IRT. IRT is the gap from one response offset to the next response onset.
- Use the boundaries in the definition. Subtract offset from onset in the correct order: offset minus onset.
- Do not confuse interval scoring with continuous timing. A sampled interval percentage does not automatically equal seconds of behavior.
- Match the denominator. Rate uses observation time; percentage of total duration uses the observation window; mean duration uses the number of episodes.
For a quick check, ask: “Am I measuring the length of each defined episode, the number of episodes, the delay before the response, or the gap between responses?” If the answer is “the length,” use duration and identify the onset and offset boundaries before calculating.
Want more timed practice? Try the free BCBA mock exam to practice selecting measurement procedures under exam conditions.
