A teacher wants to know whether a student's engagement is improving. Timing every second of engagement would answer that question precisely, but the teacher is also delivering instruction to twenty students. Could a brief observation every minute be enough?
Now replace engagement with brief, high-intensity aggression. Would checking once per minute still be acceptable? The answer changes because the clinical question and the cost of missing events have changed.
This is the final article in our measurement series. We have already examined how partial-interval, whole-interval, and momentary time sampling work. Today we tackle the decision that comes before choosing among them: when can an estimate support sound decision-making, and when do you need the behavioral record itself?
The BACB 6th Edition Test Content Outline explicitly includes C.5 (distinguishing continuous and discontinuous measurement), C.6 (designing and applying discontinuous procedures), and, most importantly for today's decision framework, C.9 (selecting a representative measurement procedure based on the critical dimension and environmental constraints). C.8 adds the duty to evaluate measurement validity and reliability (BACB, 2022; updated 2024).
A multiple-choice question may ask you to recognize a method. The tougher clinical question is whether that method preserves enough information to make the right decision.
| Continuous measurement | Discontinuous measurement |
|---|---|
| Attempts to capture every relevant occurrence or temporal value during the defined observation period | Produces scores for intervals or selected observation moments |
| Can yield count, rate, duration, latency, or interresponse time when appropriately arranged | Often yields percentage of intervals or moments scored positive |
| Preserves the dimension directly measured, subject to observer and instrument error | Trades completeness for feasibility; may miss events or distort estimates |
Discontinuous measurement does not necessarily mean the observer stops watching: whole- and partial-interval procedures may require observing throughout an interval while recording only an interval-level score.
A continuous record is not magically accurate. A poorly defined response, missed event, or inconsistent start/stop rule can corrupt it. What continuous measurement offers is the opportunity to preserve each relevant event or value rather than deliberately compressing observations into samples.
Cooper, Heron, and Heward (2020) distinguish repeatability (e.g., count), temporal extent (e.g., duration), and temporal locus (e.g., latency and interresponse time). The first question is therefore not “Which app or interval length should we use?” It is “What feature of the behavior matters to this decision?”
Fiske and Delmolino (2012) frame discontinuous-measurement selection around the behavioral dimension, expected direction of change, and practical constraints. That is a better starting point than selecting partial-interval recording simply because a data sheet already exists.
Consider a hypothetical 10-minute observation with three episodes of vocal protest, lasting 10, 40, and 10 seconds. Direct event and duration recording produce 3 episodes and 60 seconds total duration (10% of the observation). Mean duration per occurrence is 20 seconds.
Illustrative exact measurements, assuming onset and offset were defined and observed accurately.
Now imagine scoring 20 successive 30-second intervals instead. Depending on where the three episodes fall relative to interval boundaries, they may intersect different numbers of intervals. In this illustration, they intersect four intervals, so partial-interval recording reports 4/20 positive intervals = 20%. That is not four episodes, and it is not proof that protest occupied 20% of the time. The direct duration record still says 10%.
That does not mean a 20% partial-interval result is useless. It means the result answers a different question: in what proportion of the intervals was any protest detected? If you need the number of episodes or exact time occupied, the interval score cannot recover those facts.
“Twenty percent of intervals” is not automatically “twenty percent of time.” Nor is it a count of responses. A student could emit several responses within one scored interval, and the record would still contain only one positive interval.
Sampling can be defensible when continuous recording would seriously interfere with instruction, safety monitoring, or other clinical responsibilities; the chosen procedure is suited to the behavior's distribution; and the likely measurement error is acceptable for the decision at hand.
For example, a teacher observing sustained independent engagement at scheduled moments may obtain useful trend information without continuously timing every period of engagement. But that choice should be tested against the behavior and context, not assumed accurate simply because the data are easy to collect.
Research underscores why. Fiske and Delmolino (2012) reviewed how discontinuous methods differ in accuracy and sensitivity to change. In their evaluation of problem-behavior measurement in common practice, LeBlanc et al. (2020) found that shorter sampling intervals generally corresponded more closely with continuous data, and momentary time sampling performed better than the interval-recording methods they examined. These findings describe the conditions evaluated in that study; they do not establish that momentary time sampling is the best method for every behavior or clinical decision.
Suppose a treatment aims to reduce brief, dangerous aggression. A momentary sample could miss multiple short events occurring between checks. If even one missed event matters to safety or treatment evaluation, a count-based or otherwise appropriately direct system should be prioritized, subject to observation feasibility and staff safety.
Similarly, when a learner's tantrum episodes change from five minutes to thirty seconds, a coarse interval percentage might obscure a clinically important improvement. If episode duration is the key outcome, measure that dimension directly when feasible.
That last check is not a universal rule that everyone must run two systems forever. It is a practical validation step: if sampling is intended to stand in for a dimension you care about, gather evidence that it behaves like a useful proxy under your actual conditions.
Choose an answer, then read the explanation. These four items focus on what a measurement system permits you to conclude, not merely on memorizing procedure names.
During 10 minutes, a learner protests in three episodes totaling 60 seconds. Partial-interval recording using 20 thirty-second intervals produces four positive intervals.
What does the discontinuous score establish?
Why: Four of twenty intervals were scored positive, so the result is 20% of intervals with any protest. The direct record separately establishes 60 seconds across three episodes, or 10% of time and 20 seconds mean duration per occurrence.
Boundary: A partial-interval score cannot reconstruct the precise number or duration of episodes.
A BCBA must evaluate whether aggressive episodes are occurring less often and getting shorter. Each episode has identifiable onset and offset, and a trained observer can record them accurately.
Which system best preserves both outcomes?
Why: Event recording yields the count; onset-to-offset timing yields each duration. Interval or momentary scores compress the record and cannot give both exact dimensions.
Boundary: Direct recording is only as good as the episode definition and observation integrity.
A teacher checks for on-task behavior at 20 scheduled moments across a lesson. On-task behavior is observed at 15 of the moments.
What is the most defensible report?
Why: Fifteen divided by twenty gives 75% of samples positive. The measure can estimate time occupied under suitable sampling conditions, but it does not establish exact duration or count.
Boundary: The accuracy of an estimate depends on sampling interval, behavior distribution, and the purpose of the data.
A learner engages in very brief aggression that can cause injury. A proposed momentary time-sampling system checks the behavior every two minutes; episodes often last less than five seconds.
What should the BCBA prioritize?
Why: A point check every two minutes can miss brief episodes between moments. When individual events matter for safety, direct event measurement is more suitable if it can be implemented without compromising safety.
Boundary: Feasibility and safety are part of measurement selection; a nominally precise procedure that staff cannot implement is not a solution.
Part 1 asked whether measurement could be trusted. Part 2 showed why different interval percentages mean different things. Parts 3 and 4 distinguished count, rate, duration, latency, and IRT. This final installment adds the practical decision: does your measurement procedure preserve the information needed to act responsibly?
The best system is not automatically the most detailed or the easiest. It is the one that can be implemented accurately, captures the critical dimension, and provides data representative enough to support the decision at hand.
Don't ask only, “Can we collect this?” Ask, “If the behavior changes in a clinically important way, will these data show it?”
BA Prep's practice questions ask you to select and interpret a measurement system—not just name it. Practice connecting the behavioral dimension, observation constraints, and the conclusion the data can support.
Start Free TrialBehavior Analyst Certification Board. (2022). BCBA test content outline (6th ed.; updated September 2024). Tasks C.5, C.6, C.8, and C.9. Official TCO (PDF).
Cooper, J. O., Heron, T. E., & Heward, W. L. (2020). Applied behavior analysis (3rd ed.). Pearson.
Fiske, K., & Delmolino, L. (2012). Use of discontinuous methods of data collection in behavioral intervention: Guidelines for practitioners. Behavior Analysis in Practice, 5(2), 77–81. Read article.
LeBlanc, L. A., Lund, C., Kooken, C., Lund, J. B., & Fisher, W. W. (2020). Procedures and accuracy of discontinuous measurement of problem behavior in common practice of applied behavior analysis. Behavior Analysis in Practice, 13(2), 411–420. https://doi.org/10.1007/s40617-019-00361-6.