Data collection in applied behavior analysis falls into two broad categories: continuous measurement, which captures every instance of a behavior during an observation, and discontinuous measurement, which samples behavior at set intervals. Within those categories sit a handful of specific methods, each suited to different behaviors and practical constraints. Choosing the wrong one can paint a misleading picture of a client’s progress, so the differences matter more than they might seem at first glance.
Continuous Measurement Methods
Continuous measurement means you are watching for and recording every single occurrence of a target behavior during the entire observation period. Nothing is sampled or estimated. This family of methods gives you the most complete picture of what is happening, but it demands steady attention from the observer. The most common continuous measures in ABA are frequency (or rate), duration, latency, and inter-response time.
Frequency is simply a count of how many times a behavior occurs. If a child raises their hand seven times during a 30-minute class, the frequency is seven. But a raw count by itself can be misleading if you don’t account for how long you were watching. Researchers in ABA have argued that the term “frequency” should really refer to rate, meaning the count divided by the observation time, and that reporting counts without noting the time window should be avoided.1PubMed Central. On Terms: Frequency and Rate in Applied Behavior Analysis Rate lets you compare across sessions of different lengths. If one session lasted 20 minutes and another lasted 45 minutes, the raw counts are not directly comparable, but the rate per minute is.
Duration measures how long a behavior lasts once it starts. If you are tracking how many minutes a student stays on task or how long a tantrum lasts, duration is the natural fit. You start a timer when the behavior begins and stop it when it ends. This works well for behaviors that stretch over time, like sustained engagement, self-stimulatory behavior, or crying episodes, where the concern is not how many times it happens but how long it persists.
Latency captures the gap between a prompt or signal and the start of the behavior. If a teacher gives an instruction and you want to know how quickly a student begins to comply, you measure from the moment the instruction is delivered until the student starts responding. Latency has practical value when repeated instances of a behavior would be undesirable or impractical to arrange during an assessment. Research has shown that latency and rate tend to move in opposite directions: when response rates are high, latencies are short, and vice versa, making latency a useful alternative measure in situations where you cannot or should not let a problem behavior repeat itself many times.2PubMed Central. Response latency as an index of response strength during functional analyses of problem behavior
Inter-response time, or IRT, measures the elapsed time between two consecutive instances of the same behavior. If a student calls out in class, IRT captures how long passes before they call out again. This is particularly helpful when the goal is to space out a behavior rather than eliminate it entirely. For example, if a client’s self-injurious behavior occurs every few seconds, an effective intervention might push that IRT from five seconds to several minutes. Tracking IRT lets you see that kind of progress clearly even when the overall count of behaviors hasn’t dropped much yet.
Discontinuous Measurement Methods
Continuous recording is the gold standard in terms of completeness, but it is not always feasible. A therapist working one-on-one with a client while also delivering instruction cannot realistically mark down every behavior in real time. Discontinuous methods solve this by breaking the observation period into intervals and sampling whether the behavior occurred during each one. These methods trade some accuracy for practicality.
The three main discontinuous approaches are partial-interval recording (PIR), whole-interval recording (WIR), and momentary time sampling (MTS). Each handles the trade-off between accuracy and practicality differently, and each introduces its own kind of bias.
With partial-interval recording, you divide the session into equal time intervals and mark “yes” if the behavior occurred at any point during the interval, even briefly. This makes PIR sensitive to behaviors that happen often but don’t last long. The catch is that PIR tends to overestimate how much a behavior is actually occurring, because a single brief instance in a 15-second interval gets scored the same as a behavior that filled the whole interval.
Whole-interval recording flips that logic. You score “yes” only if the behavior occurred throughout the entire interval. This approach consistently underestimates the actual level of behavior, because any pause or break during the interval results in a “no” score.3PubMed Central. An evaluation of time-sample measures of behavior That built-in underestimation makes WIR a poor choice for measuring behaviors you want to increase, since the data will always make progress look worse than it really is. But it can be useful when you want a conservative estimate of how consistently a client is engaged in a desired behavior.
Momentary time sampling asks the simplest question: is the behavior happening right now, at this exact instant? You glance at the client at the end of each interval and record what you see. Research suggests MTS is often the best interval-based system for estimating how long a behavior occupies a session, particularly when interval lengths and session durations are chosen thoughtfully to minimize error.4PubMed Central. Minimizing and Reporting Momentary Time-Sampling Measurement Error in Single-Case Research MTS is also the easiest of the three methods for a busy practitioner because you only need to attend at the moment of the observation, not during the entire interval.
A direct comparison of these discontinuous methods against continuous recording found that both PIR and MTS often produced treatment analysis conclusions that matched those from continuous data. MTS was slightly more likely to agree with duration-based continuous measures, while PIR was slightly more likely to agree with frequency-based continuous measures.5PubMed Central. A comparison of momentary time sampling and partial-interval recording for evaluating functional relations That finding is useful for practitioners deciding between methods: if you care most about how long something is happening, MTS is the safer bet; if you care about how often, PIR may track more closely.
Choosing the Right Method for the Behavior
Picking a data collection method is not arbitrary. The characteristics of the target behavior itself, along with the practical constraints of the setting, should drive the decision. A model proposed for selecting measurement procedures emphasizes matching the method to features of the behavior and the realities of the therapeutic environment.6PubMed Central. A Proposed Model for Selecting Measurement Procedures for the Assessment and Treatment of Problem Behavior
A few practical guidelines emerge from the research:
- Discrete, countable behaviors: If a behavior has a clear start and end and happens in distinct episodes (hitting, hand-raising, requesting), frequency or rate is usually the first choice.
- Extended-duration behaviors: If the main concern is how long the behavior lasts (time on task, stereotypy, sleeping), duration recording fits best.
- Compliance and responsiveness: When you want to know how quickly someone responds after a cue, latency recording captures what matters.
- High-frequency or hard-to-count behaviors: If a behavior happens so rapidly that counting each instance is impractical (e.g., fast-paced stereotypic movements), discontinuous methods may be the only feasible option.
- Resource-limited settings: When the person collecting data is also the one delivering the intervention, discontinuous methods, particularly MTS, reduce the burden substantially.
Guidelines for practitioners recommend that when designing a discontinuous measurement system, you should consider the dimension of behavior you are most interested in (frequency versus duration) and the expected direction of behavior change (increase versus decrease).7PubMed Central. Use of discontinuous methods of data collection in behavioral intervention: guidelines for practitioners For instance, if you expect a behavior to decrease and are using WIR, you might see floor effects sooner than is realistic because WIR already underestimates. That can distort your picture of how well an intervention is working.
In non-complex situations with limited resources, some researchers have suggested that estimation data may be reasonable for a portion of measurement efforts, supported by periodic continuous or sample data to check accuracy.8Research in Autism Spectrum Disorders. A comparison of data collection techniques used with discrete trial teaching Continuous recording remains the protocol of choice when accuracy is most critical, such as when data are driving complex treatment decisions or when precise documentation is required.
Probe and First-Trial Data in Discrete Trial Teaching
Discrete trial teaching (DTT) is a structured instructional format used heavily in ABA, especially with children on the autism spectrum. Each skill is practiced across multiple trials in a session, and one common shortcut is to record only the response to the first trial of each session rather than every trial. The rationale is that the first trial best reflects what the learner retained from previous sessions, since no new teaching has occurred yet.
Research comparing first-trial data against continuous data across all trials found that for most children and targeted skills, first-trial data provided a rough estimate of overall performance. However, it came with two notable drawbacks: it frequently led to premature indications that a skill had been mastered, and it was relatively insensitive to initial changes in performance.9PubMed Central. A comparison of methods for collecting data on performance during discrete trial teaching In other words, first-trial data may tell you a learner has “got it” before they truly have, and it might miss the early signs that they are starting to learn something new. If you are making decisions about when to move on to new skills, relying solely on first-trial data could mean advancing prematurely and having to reteach later.
Interobserver Agreement and Why It Matters
No matter which measurement method you use, the data are only as good as the observer producing them. Interobserver agreement (IOA) is the standard way ABA checks data quality. Two independent observers record the same behavior during the same session, and then their records are compared. If they largely agree, confidence in the data rises.
One investigation into how much IOA is enough found that, at least with well-trained observers in a structured clinical setting, the total number of sessions with IOA checks did not significantly affect the IOA scores obtained. However, IOA was sensitive to the overall rate of responding during the observation: when behavior occurred at very high or very low rates, agreement scores could shift.10PubMed. Interobserver agreement: A preliminary investigation into how much is enough? That finding is encouraging for practitioners with limited staff, because it suggests that taking IOA data on a smaller proportion of sessions may still give reliable results, as long as observers are well trained and working under structured conditions.
The way you calculate IOA also matters. A comparison of the three algorithms most commonly used in behavior-analytic research found meaningful differences in how they performed across behavior types. Block-by-block and exact agreement algorithms tended to produce inflated agreement estimates at lower rates and durations. The exact agreement method was overly strict at higher rates and durations, while time-window analysis inflated accuracy estimates at relatively high rates but not at low ones.11PubMed Central. Assessing observer accuracy in continuous recording of rate and duration: three algorithms compared The upshot is that no single IOA formula is universally best. The choice should depend on the characteristics of the behavior being measured.
Sources of Observer Error and Bias
Even with IOA procedures in place, systematic errors can creep into behavioral data. A classic review identified several factors that compromise observer accuracy, including reactivity (observers behave differently when they know they are being checked), observer drift (gradual changes in how an observer applies a definition over time), complexity of the observation code, and observer expectancies.12PubMed Central. Artifact, bias, and complexity of assessment: the ABCs of reliability If an observer knows what the intervention is supposed to do, they may unconsciously begin scoring in a direction that confirms their expectations.
Recommendations for reducing these influences include keeping observers unaware of experimental conditions when possible, conducting periodic retraining to prevent drift, using simple and clearly defined behavioral codes, and scheduling unannounced IOA checks so that reactivity is minimized.13Exceptional Children. Direct Observation: Factors Affecting the Accuracy of Observers These are not exotic fixes. They are basic hygiene that separates trustworthy data from data that look reliable but are quietly distorted.
Procedural Fidelity as a Form of Data Collection
ABA data collection is not limited to tracking client behavior. A parallel and equally important stream of data concerns procedural fidelity: the degree to which the intervention itself is being delivered as designed. If a therapist is supposed to deliver a prompt within two seconds of a demand but is routinely waiting five seconds, the behavioral data from that session reflect a different intervention than the one planned.
Measuring procedural fidelity typically involves direct observation of the implementer’s actions, often using a checklist of key steps for the procedure. A recent guide for practitioners walks through the process of creating and using fidelity measurement systems, emphasizing that direct observation of implementation is the preferred method.14PubMed Central. A Practitioner’s Guide to Measuring Procedural Fidelity This matters for quality control, data-based decision making, and achieving the best possible outcomes for clients.15PubMed Central. A Survey of Procedural-Fidelity Data Collection in Behavior-Analytic Practice
Fidelity reporting in the research literature has been improving. A detailed review of studies published in the Journal of Applied Behavior Analysis between 2006 and 2021 found increasing trends in describing how fidelity was calculated, reporting multiple fidelity values rather than a single summary number, and stating how often fidelity checks occurred.16PubMed. A detailed examination of reporting procedural fidelity in the Journal of Applied Behavior Analysis For clinicians, the practical lesson is straightforward: if you are only collecting data on what the client does and never on how accurately the treatment is being delivered, you are missing half the picture. A behavior that is not improving could signal a poor intervention choice, or it could signal that a good intervention is being delivered inconsistently.
Electronic Versus Paper-Based Recording
The shift from pen-and-paper data sheets to tablets, phones, and dedicated apps has been one of the more visible changes in ABA practice over the past decade. The appeal of electronic data collection is obvious: automatic graphing, built-in timers, cloud storage, and the ability to share data across a clinical team without physically passing around binders.
Research on whether the switch to electronic tools changes data quality or session flow gives a mixed but generally reassuring picture. One study found that electronic recording did not significantly affect IOA scores or treatment integrity compared to traditional methods, and that practitioners ranked electronic recording more favorably in terms of preference.17Behavioral Interventions. A Comparison of Electronic and Pen‐And‐Paper Recording in ABA Sessions
An earlier comparison in discrete trial training found a different wrinkle. Electronic data collection consistently required more time per session than pen-and-paper for all four participants studied. The accuracy of pen-and-paper recording was also slightly higher, averaging near-perfect levels compared to electronic accuracy that ranged more widely across participants.18Research in Autism Spectrum Disorders. A comparison of electronic to traditional pen-and-paper data collection in discrete trial training for children with autism On the other hand, producing graphs from electronic data was significantly faster than drawing them by hand. So the efficiency gain does not come during the recording itself but in what happens afterward with the data.
The practical takeaway is that electronic tools are not inherently better or worse for the actual recording step. They tend to save time on the back end (graphing, sharing, storage) while sometimes adding a few seconds per data entry during live sessions. For teams that need to review data quickly or share it across locations, the trade-off usually favors electronic systems. For a lone practitioner running fast-paced DTT trials, the slight speed cost per entry is worth knowing about.
Visual Analysis of Graphed Data
Collecting data is only half the job. Once data are recorded, ABA relies overwhelmingly on visual inspection of single-case graphs to judge whether an intervention is working. This is the primary method of data interpretation in the field, rather than the statistical tests common in group-design research.19PubMed Central. Consistent visual analyses of intrasubject data
A clinician plots the data on a line graph, usually with sessions along the horizontal axis and the behavioral measure on the vertical axis. Changes in level (how high or low the data sit), trend (whether the data are moving up, down, or staying flat), and variability (how much the data points bounce around) across phases tell the story. If a behavior was high during baseline and drops after an intervention is introduced, and that pattern reverses when the intervention is withdrawn, the visual case for a functional relationship is strong.
The strength of visual analysis is that it keeps the clinician close to the raw data, reducing the temptation to trust a single summary statistic. Its weakness is that it relies on human judgment, which can be inconsistent. Two analysts looking at the same graph may disagree, especially when effects are moderate or data are highly variable. Training and explicit decision rules help, but some subjectivity is baked in. This is one reason why the quality of the underlying data collection matters so much: noisy input from a poorly matched measurement method produces a graph that is harder to read.
Social Validity Assessments
Not all data in ABA come from direct observation of behavior. Social validity assessments collect information about whether the goals, procedures, and outcomes of an intervention are meaningful and acceptable to the people most affected: the client, their family, teachers, or other stakeholders. The idea is that an intervention that works on paper but is rejected by the family, or that targets a behavior nobody actually cares about changing, has limited real-world value.20PubMed Central. Social validity assessments: is current practice state of the art?
These assessments typically take the form of rating scales or questionnaires given to caregivers, teachers, or clients themselves. A systematic review of social validity assessments published between 2010 and 2020 found that most measured the acceptability of intervention procedures and outcomes, with fewer evaluating whether the goals themselves were appropriate. The most common collection method was Likert-type rating scales, followed by other types of questionnaires.21PubMed. A systematic review of social-validity assessments in the Journal of Applied Behavior Analysis: 2010-2020
Behavioral practitioners and researchers use these assessments to check that target behaviors are appropriately selected, that intervention procedures are acceptable to the people involved, and that the outcomes produced are satisfactory.22PubMed Central. Social Validity in Behavioral Research: A Selective Review Social validity data rarely drive session-by-session treatment decisions the way direct behavioral data do, but they serve a different purpose: keeping the intervention aligned with what the people it serves actually want and need. An intervention that produces impressive graphs but leaves parents frustrated with the process, or targets a skill the client never uses outside the clinic, has a social validity problem that no amount of frequency data can solve.
When Measurement Systems Interact With Behavior
One underappreciated aspect of data collection in ABA is that the act of measurement can itself influence behavior, both the client’s and the observer’s. If a child notices that a therapist is clicking a counter every time they engage in a target behavior, the awareness alone can change how often the behavior occurs. This is a well-known phenomenon in observational research generally, and ABA is not immune to it.
On the observer side, the phenomenon of reactivity means that observers may record differently when they know their data are being checked for agreement compared to when they believe they are recording privately. Classic work on this topic identified reactivity as one of several factors that can distort agreement scores, alongside observer drift and expectancy effects.23PubMed Central. Artifact, bias, and complexity of assessment: the ABCs of reliability The implication is that high IOA numbers in a setting where observers always know when they are being checked may not reflect the quality of data collected during unchecked sessions.
For practitioners, the antidote is not to abandon IOA but to build in variety: unannounced reliability checks, periodic retraining on behavioral definitions, and data collection systems that are as simple and intuitive as possible to minimize judgment calls in the moment. The less room a measurement system leaves for interpretation, the less vulnerable it is to the human factors that distort it.