A teacher finishes giving a direction at 9:00:00. The student begins working at 9:00:12, stops at 9:00:50 after 38 seconds of work, and starts again at 9:01:05. For this example, latency starts when the direction ends.
Three time-based measurements are hiding in those four timestamps. All three are measured in seconds. But they answer different behavioral questions.
How long did the student take to begin? How long did the work episode last? How long was the gap before working started again?
A stopwatch does not tell you which dimension you measured. Your start and stop events do.
The BACB 6th Edition Test Content Outline places duration, latency, and interresponse time together in C.4: Measure temporal dimensions of behavior (e.g., duration, latency, interresponse time). Selecting which dimension is relevant also draws on C.9: Select a measurement procedure to obtain representative data that accounts for the critical dimension of the behavior and environmental constraints.
As Cooper, Heron, and Heward (2020) distinguish, duration measures temporal extent, while latency and IRT locate responses in time (temporal locus). The count and rate measures covered in part 3 address response repeatability. The exam can ask for a definition. A stronger application question presents timestamps, a clinical target, or an observation constraint and asks which measure—or interpretation—actually answers the question.
Duration measures the temporal extent of a response or episode: the time from the behavior's defined onset to its defined offset.
In our example, the student starts working 12 seconds after the direction and stops at 50 seconds. That work episode lasts 38 seconds. The direction at zero seconds does not start the duration clock: the behavior has not yet begun.
You can summarize duration in different ways. Total duration sums time spent in the target behavior across an observation. Mean duration per occurrence divides total duration by the number of episodes recorded. Percentage of observation time expresses total time engaged relative to actual observation time when that measure answers the question.
If a learner engages in three episodes lasting 20, 40, and 60 seconds, the total duration is 120 seconds and the mean duration per occurrence is 40 seconds. Neither number tells you how long the learner waited after an instruction to begin.
“Off task” and “engaged” are not self-timing definitions. Decide what observable event counts as onset, what counts as offset, and whether a brief interruption ends the episode. Otherwise two observers can time different events while believing they used the same procedure.
Latency measures time from a defined antecedent event to the onset of the target response. A conventional definition measures from the onset of an antecedent stimulus (Cooper, Heron, & Heward, 2020), while some instructional recording protocols start after a spoken direction is completed. Either approach requires a precise, consistently applied operational rule; they are not interchangeable when the direction takes time to deliver. The IRIS Center offers a teaching resource on duration and latency recording.
Using the timeline, the direction finishes at zero seconds and working begins at 12 seconds. The latency is 12 seconds. The 38 seconds the student continues working are not part of the latency.
Why would a BCBA choose latency? Because the clinical concern might be delayed initiation rather than short engagement. A student can begin an assignment promptly but stop almost immediately; another may take a long time to begin and then work steadily. Those patterns call for different decisions.
Define the initiating event precisely. Does the measurement start at the onset or offset of an instruction? After a prompt, after a transition signal, or after a material is presented? Those choices are not interchangeable, and the observation must consistently follow the selected rule.
A trial that ends without the target response is not automatically a zero-second latency. Nor can it be treated as an observed latency equal to the trial length without an explicit measurement rule. The response onset was not observed within the opportunity.
Report the nonresponse and observation limit separately—for example, “no initiation within the 30-second opportunity.” If a protocol assigns a capped score for analysis, identify that convention rather than describing the cap as the observed response latency. Averages based only on trials with a response can also conceal important nonresponses.
Latency is a dimension, not a value judgment. Shorter latency to starting a requested task may be desirable; shorter latency to aggression after a demand may not be. First identify which response starts the clock and why that measure is clinically meaningful.
Interresponse time (IRT) measures the interval between successive responses in a defined response class. For responses or episodes with measurable duration, a common onset/offset convention is to time from the end of one response to the beginning of the next.
Our student stops working at 50 seconds and starts working again at 65 seconds. The gap is 15 seconds. That is the illustrated IRT between work episodes. The first response in a session can have a latency from an identified antecedent event, but there is no preceding within-session response from which to calculate its IRT.
Notice how differently the same observations could be described: the 38-second episode is duration; the 15-second pause before the next episode is IRT. Measuring the pause does not tell you how long either episode lasted.
With very brief, discrete responses, the distinction between a response's onset and offset may be negligible in everyday recording. But it is not safe to substitute onset-to-onset for offset-to-onset when episodes have appreciable duration. A graph, protocol, or research report must specify the intended IRT boundary.
IRT can be useful for studying response spacing, but it is not another name for rate. Rate summarizes the number of responses over observed time. IRT summarizes measured gaps between successive responses. Extended response durations, gaps in observation, or variable observation windows can make a simple reciprocal conversion misleading. For very brief, discrete responses observed continuously, mean IRT may be approximately the reciprocal of rate when response duration and session-edge effects are negligible; it is not an exact identity for arbitrary episodes or observation windows.
If a clinical question concerns time between instances of a discrete target response, measure that gap directly rather than assuming the overall rate tells you exactly how responses were spaced.
| Dimension | Start the clock | Stop the clock | Question answered |
|---|---|---|---|
| Duration | Target behavior begins | Target behavior ends | How long did it last? |
| Latency | Specified antecedent event | Target behavior begins | How long until it started? |
| IRT | Prior response ends | Next response begins | How long was the gap? |
The table uses an offset-to-onset IRT convention for episodes with measurable duration. For an actual measurement plan, make the event boundaries and observation rules explicit.
LeBlanc et al. (2016) emphasize selecting measurement procedures with attention to the response characteristics and practical assessment or treatment circumstances. Imagine a learner who leaves their seat during independent work. The team might want to know how many seconds each out-of-seat episode lasts (duration), how soon after work begins the first departure occurs (latency), or how long the learner remains seated between departures (IRT between episodes, if departures are defined as episodes).
None is automatically the “best” dimension. Each measures a different relation, and an appropriate choice depends on the target's operational definition, clinical significance, and observation conditions. A comprehensive plan may need more than one measure.
Similarly, if a learner takes 45 seconds to begin a task but then works for 15 minutes, the primary concern may be response initiation. If they begin in 3 seconds but work for only 20 seconds, duration may reveal the more relevant problem. A single “on-task percentage” may not reveal either pattern clearly.
Continuous timing of each episode may be difficult when an observer supervises several learners. That does not change what duration, latency, or IRT mean. It changes whether the observer can measure them accurately and whether an alternative or supplementary procedure is appropriate. Do not select a measure solely because the stopwatch is available.
These questions deliberately mix time calculations with measurement selection. Identify the relevant start and stop events before you look at the units.
After a teacher finishes a direction, a student begins writing 9 seconds later, writes continuously for 42 seconds, pauses for 18 seconds, and begins writing again.
What is the duration of the first writing episode?
Why: Duration starts when writing begins and ends when that episode stops. The 9-second wait is latency, and the 18-second gap is an IRT between writing episodes using an offset-to-onset convention.
Boundary: Adding waiting and pausing time to the writing episode would misrepresent how long writing actually occurred.
A technician records latency to initiating a request after an opportunity is presented. The observation window lasts 30 seconds. On one trial, the learner never initiates the request. The protocol does not specify how to score a trial with no response.
Which record is most defensible?
Why: The event that would stop the latency clock never occurred during the observation window. Without a predefined rule, a zero, an exact 30-second observed latency, or an IRT would misstate what was observed. Report the nonresponse separately.
Boundary: Some protocols assign a maximum or capped analytic value; that is a reporting convention and should be disclosed as such.
A student's first vocalization begins at 10 seconds and ends at 13 seconds. The next begins at 21 seconds. The team defines IRT as the interval from the end of one vocalization to the start of the next.
What is the IRT between these two vocalizations?
Why: Under the specified convention, the prior response ends at 13 seconds and the next begins at 21 seconds, so the gap is 8 seconds.
Boundary: Onset-to-onset is 11 seconds, but that includes the first response's three-second duration. It is not the defined gap between responses here.
A student begins independent work promptly after each direction but stops within 15–25 seconds. The team's primary goal is to increase the amount of time the student remains engaged once work has begun. Continuous observation is available.
Which measure best addresses that goal?
Why: The outcome is how long engagement continues after initiation, so duration directly measures the critical temporal dimension. Latency addresses the delay before work begins; IRT addresses gaps between episodes.
Boundary: Total session time includes periods without engagement, so it does not directly measure the duration of each work episode.
In part 3, the denominator changed the interpretation of an occurrence measure. Here the crucial distinction is which events define the interval. A number such as “15 seconds” is incomplete until you can say 15 seconds between what and what?
Next in this series (part 5): continuous versus discontinuous measurement, focusing on how to choose between a complete record and an estimate under real-world observation constraints. For the scoring rules of momentary time sampling, partial-interval, and whole-interval recording, see part 2.
A stopwatch measures elapsed time. A behavior analyst defines which elapsed time matters.
BA Prep practice questions are built to make you identify the measurement boundary before you choose a formula. If you can explain exactly when the clock begins and ends, you're doing the hard part.
Start Free TrialBehavior Analyst Certification Board. (2022). BCBA test content outline (6th ed.; updated September 2024). Tasks C.4 and C.9. Official BACB TCO.
Cooper, J. O., Heron, T. E., & Heward, W. L. (2020). Applied behavior analysis (3rd ed.). Pearson.
LeBlanc, L. A., Raetz, P. B., Sellers, T. P., & Carr, J. E. (2016). A proposed model for selecting measurement procedures for the assessment and treatment of problem behavior. Behavior Analysis in Practice, 9(1), 77–83. https://doi.org/10.1007/s40617-015-0063-2.
Vanderbilt University, IRIS Center. Behavior assessment: Duration and latency recording. Measurement teaching resource.