DRAGENwiki · Article
Latency: Measuring How Long It Takes to Start
Customize your experience
Sources checked. Clinical review pending.
Latency is the time between a cue and the start of a behavior. You start the clock at the cue and stop it at the first move you defined. Choose latency when how soon a response begins matters more than how often it happens or how long it lasts.
Latency in 60 seconds
- Time the gap. Latency recording measures "the amount of time (usually in seconds) it takes each behavior to occur following a specific environmental event". [3]
- Cue to start. LeBlanc and colleagues put it simply: "Record the amount of time that passes between the discriminative stimulus and the onset of the behavior". [3]
- One usual summary. Latency data "are typically summarized as the mean latency to each behavior during an observation." [3]
- A close cousin. Interresponse time (IRT) is the time from one response to the next. It is explained below, on this page. [5]
- It costs attention. You watch two things at once: the cue and the behavior. [3]
Where latency fits
Latency is one of the temporal dimensions of behavior. The BCBA outline asks analysts to "Measure temporal dimensions of behavior (e.g., duration, latency, interresponse time)." [1] The technician outline lists latency in "Implement continuous measurement procedures (e.g., frequency, duration, latency, interresponse time)." [2]
- Latency asks how long it took to start after a cue.
- Interresponse time asks how long passed between one response and the next.
- Duration asks how long each episode lasted.
- Frequency and rate ask how many, and how many per minute.
Latency ends at the very moment duration begins: the onset. Both are forms of continuous measurement, because you need the exact moment each thing happens.
When latency is the right measure
LeBlanc and colleagues write that "Latency recording is a preferred measurement procedure when information about a behavior's latency is the dimension of interest." [3] Some questions where the start is the point:
- How soon does a person answer? A friend's greeting, or a question at the dinner table.
- How soon does a learner begin a chosen activity? Opening a book once it is in hand.
- How soon does a person move when an alarm sounds? Here a quick start is a safety skill.
- How long does a learner wait before taking a turn? Sometimes a longer latency is the goal, such as waiting for a turn in a game or pausing to think before answering.
A count cannot answer these. A learner can answer every question in a session and still take a long time to begin each answer.
The cue and the start
A latency record is only as good as the two moments that start and stop the clock. Write both into the operational definition:
- The cue. The event that starts the clock. LeBlanc and colleagues give "a discriminative stimulus" as an example. [3] For a spoken cue, DRAGEN's latency sheet says to say it once, in the same words each time, and to start timing when you finish speaking.
- The onset. The first move that counts. "Picks up the pencil" is clearer than "starts working."
- The wait limit. How long you will wait before you stop the clock with no start. The DRAGEN sheet suggests a limit such as 2 minutes. If nothing starts by then, write the limit and note it.
Research uses limits too. In one of Thomason-Sassi, Iwata, Neidert and Roscoe's studies, some sessions "were terminated immediately following the delivery of programmed consequences for the first instance of a target behavior or when 5 min elapsed, whichever came first." [4]
Two people reading the same definition should start and stop their timers at the same moments. If they would not, the definition needs more detail.
Mean latency, step by step
LeBlanc and colleagues write that "Latency-recording data are typically summarized as the mean latency to each behavior during an observation." [3] Their table lists "Mean latency" as the resulting measure for latency recording. [3]
A worked example. In one session, a learner hears the same cue four times. The latencies are 12, 8, 20 and 4 seconds.
- Total: 12 + 8 + 20 + 4 = 44 seconds.
- Mean latency: 44 ÷ 4 = 11 seconds.
When a trial reaches the limit. Say the limit is 60 seconds, and on a fifth trial nothing starts. Following the DRAGEN sheet, you write 60 and note it. The mean becomes (44 + 60) ÷ 5 = 20.8 seconds.
That number is a floor. The real wait on the fifth trial was longer than 60 seconds, so the real mean was higher than 20.8. Report the limit count next to the mean, every time:
- Mean latency: at least 20.8 seconds.
- Trials at the limit: 1 of 5.
A graph of the means alone would hide that fifth trial. The note keeps it visible.
Interresponse time: the time between responses
Interresponse time, or IRT, uses the same clock with a different starting point. Latency runs from a cue to a response. IRT runs from one response to the next response of the same kind.
In a tutorial on one data collection app, Slanzi and Fernand explain that "Calculating interresponse time is almost identical to calculating latency". [5] In that app, IRT needs a single button for the behavior: "only one key would be required per target response, which would be selected each time the event occurred." [5] On paper, the same idea means writing the clock time of every response.
A worked example. A learner answers math problems at 2:10, 2:40 and 3:30 into a session. The IRTs are 30 seconds (2:10 to 2:40) and 50 seconds (2:40 to 3:30).
- Short IRTs mean responses come close together, so the rate is high.
- Long IRTs mean long gaps, so the rate is low.
IRT is useful when the spacing is the point, such as the pace of work through a page of problems. Slanzi and Fernand advise against measuring IRT with the app's duration key, a timer switched on and off around each event, because "it may not be feasible to turn on and off the key repeatedly as events occur". [5] Write the moment each response happens instead.
Recording latency, step by step
- Write the cue, the onset and the wait limit. Keep them where every observer can see them.
- Have a timer ready. A phone stopwatch works.
- Give the cue the same way each time. A cue that changes can change the latency.
- Start at the cue, stop at the onset. Write the seconds on their own line.
- At the limit, stop. Write the limit and note it.
- Sum it up. Report the mean latency and the number of trials at the limit, the same way every session.
Paper works. The latency recording sheet has columns for the date, the time the cue was said, the time the response started, and the seconds between.
What makes latency hard
It takes two kinds of attention. "However, like duration recording, latency recording requires constant vigilance (for both the behavior and antecedent event) and a timing device." [3] An adult who misses the cue cannot time the gap.
The first response is all it shows. Thomason-Sassi and colleagues name this limit in their second study: their latency graphs "provided information on the first response that occurred in a session" but "provided no indication of how many subsequent responses occurred within that same session." [4] If you need to know how many, add a count.
The cue has to stay the same. A cue said twice, said louder or paired with a gesture is a different cue. The latency may change because the cue changed, not the learner.
Checking latency data
A second adult can time the same trials, and you compare the two records. Thomason-Sassi and colleagues used a simple formula: "Reliability for latency data was calculated by dividing the shorter latency (in seconds) by the longer latency and multiplying by 100%." [4]
A worked example. One observer writes 14 seconds and the other writes 15. Agreement is 14 ÷ 15 × 100 = 93%.
Latency agreement can vary. In their study, "Mean reliability scores across subjects ranged from 94% to 97% for response rate and from 78% to 99% for response latency." [4] Clear cue and onset rules help two observers start and stop together.
The interobserver agreement page and the lesson How to check your data with interobserver agreement (IOA) show how to compare two records.
Latency when repeating a behavior is not wise
Most assessment counts many occurrences. Thomason-Sassi and colleagues note that "Research on the assessment and treatment of problem behavior traditionally has used measures that reflect repeated occurrences of behavior, such as rate, duration, or the percentage of intervals during which responding occurs." [4]
They asked whether the first occurrence could be enough: "We examined response latency as the index of behavior during assessment." [4] Across three studies, they found "an inverse relation between rate and latency, indicating that latency might be a useful measure of responding when repeated occurrences of behavior are undesirable or impractical to arrange." [4] In plain words: in conditions where a behavior happened often, it also started sooner.
The results lined up well. "A high degree of correspondence was observed between outcomes based on latency and repetition measures." [4] In their second and third studies, latency predicted the results of standard functional analyses "with a high degree of accuracy", and in the third it did so "while requiring many fewer responses". [4]
This matters because some behaviors are risky or upsetting every time they happen. A measure that needs only the first occurrence asks less of everyone. This was research, and a functional analysis is a BCBA's job, done with consent and a safety plan.
Measuring yes and no
Latency can show willingness. Breaux and Smith's table of measures for assent and assent withdrawal includes "Latency to initiate a practitioner-directed task (i.e. work task)". [6] A quick start can be one sign that a learner is on board. A slow start can be a sign that the task, the setting or the relationship needs a look. "A learner will likely use a myriad of topographies to give or withdraw assent" [6], so learn each learner's own ways of saying yes and no before you time anything.
DRAGEN's rule. We never time a learner's "no" in order to make it shorter. A slow start is information about the task, the setting or the relationship. It is never "noncompliance" and never a target. We do not push for faster starts with pressure. When latencies grow, the team asks what changed. When there is a real safety need, adults act right away, and that support stays separate from the learner's choices during practice.
A made-up example: Student F
Student F is 9. Loud alarms are hard for F. F, the family and the school team agree to practice fire drills so that F can feel ready and get outside quickly.
- The definition. Cue: the first sound of a recorded alarm, played at a volume F chooses. Onset: F stands up and takes a first step toward the door. Wait limit: 60 seconds.
- The method. The teacher starts a phone stopwatch at the first sound and stops it at F's first step. Each practice run gets its own line.
- First week. Three runs: 25 seconds, 40 seconds, and no start by 60 seconds, written as 60 and noted. Mean latency: at least 41.7 seconds, with 1 of 3 runs at the limit.
- The plan. F controls the volume, wears headphones if F chooses, follows a picture route and walks with a buddy.
- Some weeks later. Three runs: 6, 4 and 8 seconds. Mean latency: 6 seconds, with no runs at the limit.
- Choice. F may skip any practice run. A skipped run is written as skipped, never as a slow latency.
- Safety stays separate. In a real alarm, adults help F outside every time. That is safety, not practice, and it is never counted as a practice run.
- The check. In week 3, a second adult timed the same run: 14 seconds and 15 seconds, or 93% agreement.
- Reading it honestly. Faster matters here because F practices willingly and feels more ready. If the times dropped because F felt rushed, the number would no longer measure what the team cares about.
Common myths
- "Shorter is always better." Not always. Some goals want a longer wait, like waiting for a turn. A fast start matters only when the learner is choosing it.
- "Latency and duration are the same thing." Latency ends where the behavior starts. Duration starts there.
- "A trial with no start counts as zero." It does not. Write the limit and note it, and report how many trials reached it.
- "You need special equipment." A timing device is all it takes, and a phone stopwatch works. [3]
- "Latency is only for instructions." Any clear event can start the clock: an alarm, a question or a greeting. [3]
- "A latency tells you how often." It tells you about the first response only. [4]
Words to know
- Latency: the time from a cue to the start of a behavior. [3]
- Cue: the event that starts the clock. A discriminative stimulus is one example. [3]
- Onset: the first moment of the behavior that counts.
- Trial: one cue, and the response or the limit that follows it.
- Mean latency: the total of the latencies divided by the number of trials. [3]
- Wait limit: how long you wait before you stop the clock with no start.
- Interresponse time (IRT): the time from one response to the next response of the same kind. [5]
Need a PDF?
Create it here, for free: the DRAGEN app. Use it at your next session or team meeting.
Or print one we already made, also free:
- Latency recording sheet (PDF)
- Which measure? card (PDF)
- IOA check sheet (PDF)
- Graph-reading checklist and blank graph (PDF)
- Every free PDF
More from DRAGEN
- Lesson: How to count a behavior: frequency, duration, latency and interval recording
- Lesson: How to check your data with interobserver agreement (IOA)
- Lesson: How to read a behavior graph in sixty seconds
- Free app for parents: DRAGENparent
- BCBA care: DRAGENtherapy
- Books: DRAGENbooks
- Talk with Chase: Contact
- Home: dragen.cloud
Related DRAGENwiki pages: measuring behavior · continuous measurement · duration · frequency and rate · operational definition · interobserver agreement · celeration
What remains unknown
- The sources on this page do not settle how to score a trial with no start. This page follows the DRAGEN sheet: write the limit, note it, and report how many trials reached it.
- Thomason-Sassi and colleagues found that "a history of reinforcement might influence response latency even when contact with the contingency is limited." [4] A latency may reflect a person's past, not only what is happening now.
- LeBlanc and colleagues write that their selection model "has not been empirically tested". [3]
- The certification outlines name latency and interresponse time, but they do not say how to summarize them. [1] [2]
- This page is general information, not medical or legal advice.
- This page was last checked on November 12, 2026.
Sources
- Behavior Analyst Certification Board. BCBA Test Content Outline (6th ed.). https://www.bacb.com/wp-content/bcba-outline-6thEd/
- Behavior Analyst Certification Board. RBT Test Content Outline (3rd ed.), effective January 1, 2026. https://www.bacb.com/rbt-test-content-outline-3rd-ed
- 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 (free full text: https://pmc.ncbi.nlm.nih.gov/articles/PMC4788644/)
- Thomason-Sassi, J. L., Iwata, B. A., Neidert, P. L., & Roscoe, E. M. (2011). Response latency as an index of response strength during functional analyses of problem behavior. Journal of Applied Behavior Analysis, 44(1), 51–67. https://doi.org/10.1901/jaba.2011.44-51 (free full text: https://pmc.ncbi.nlm.nih.gov/articles/PMC3050468/)
- Slanzi, C. M., & Fernand, J. K. (2024). On the use and benefits of electronic data collection systems: A tutorial on Countee. Behavior Analysis in Practice, 17(4), 1228–1237. https://doi.org/10.1007/s40617-024-00964-8 (free full text: https://pmc.ncbi.nlm.nih.gov/articles/PMC11707102/)
- Breaux, C. A., & Smith, K. (2023). Assent in applied behaviour analysis and positive behaviour support: Ethical considerations and practical recommendations. International Journal of Developmental Disabilities, 69(1), 111–121. https://doi.org/10.1080/20473869.2022.2144969 (free full text: https://pmc.ncbi.nlm.nih.gov/articles/PMC9897747/)