Translated from the Portuguese original. Read the original (Português)
Recording data means turning what you observe into numbers someone else can check. In Applied Behavior Analysis, this involves three decisions: what will be counted (frequency? duration? how long until it starts?), how the recording will be done (every occurrence, by time samples, or by the traces the behavior leaves), and how much you can trust the result. Different methods applied to the same episode produce very different numbers — which is why the choice of method is not a bureaucratic detail: it decides what you will end up seeing.
Measurement is the process of assigning numbers and units to the characteristics of objects or events. Applied to behavior, it consists of observing a response previously described by an operational definition and quantifying it along one of its dimensions — repeatability (count, rate), temporal extent (duration), or location in time (latency, interresponse time). The recording procedure is the concrete method that produces that number: who observes, under what conditions, with what instrument, and according to what marking rule. In Cooper, Heron & Heward, the quality of that number is judged by three distinct criteria — validity (does it measure what matters), accuracy (how close it is to the true value), and reliability (does it repeat consistently).
Why recording is the heart of the work
Every behavioral intervention promises one thing: observable change. The only honest way to back up that promise is to measure. That is what Baer, Wolf, and Risley stated in the field's founding article, in 1968, when they said a practice is only behavioral if it measures what the person actually does — and only analytic if it demonstrates what produced the change.
In everyday practice, recording serves four purposes at once. It describes the starting point (it is what supports the baseline). It reveals patterns invisible to memory — that meltdowns cluster in the late afternoon, that duration has been falling even while frequency stays stable. It warns when a plan is not working, before months are lost. And it makes the work checkable by someone else: family, school, supervision.
There is also a less obvious gain. Whoever records is forced to decide, before starting, exactly what counts as the target behavior. That requirement — describing it in observable terms, without relying on interpretation — often clarifies a case more than weeks of conversation about what "the child is feeling."
What exactly gets counted
"Recording the behavior" is vague: behaviors have several measurable faces, and the wrong choice hides exactly what changed. A child might keep the same number of meltdowns per week while each one shrinks from thirty minutes to three — if the recording only counts episodes, that huge progress stays invisible. Flip the cards to see the most commonly used measures:
How recording works in practice
Once you've chosen the dimension, you still need a procedure. There are four families, and each trades precision for feasibility in a different way. Explore them:
Event recording — marking every occurrence
The observer notes every single time the behavior happens: a tally on paper, a click on a counter, a tap on an app. It is continuous measurement par excellence, and the most faithful when it works. It requires two conditions, though: the response must have a clear start and end, and it cannot happen too fast to be tracked. Counting "spontaneous requests" works well; counting every rocking movement in a long sequence does not.
Timing — measuring time
A stopwatch records duration, latency, or interresponse time. It is simple to describe and tricky to carry out: it requires starting and stopping the clock exactly at the defined instants, which gets difficult when the behavior has a gradual onset. That is why the operational definition must state, beyond doubt, which observation marks the start and which marks the end of the episode.
Time sampling — estimating from samples
When it is not possible to observe everything all the time, the session is divided into intervals, and a record is made only of whether the behavior occurred — without counting how many times. It is a discontinuous measure: it produces an estimate, not the real value. In the three classic variants, the marking rule changes everything: whole-interval (only marks if it occurred from start to finish), partial-interval (marks if it occurred at any moment), and momentary time sampling (marks if it was occurring at the instant the interval ended). The chart below shows the effect of that choice.
Permanent product — measuring the trace
Some behaviors leave lasting traces: dishes washed, exercises solved, toys put away, messages sent. Measuring the product frees you from observing during the episode, removes the interference of the observer's presence, and allows a check later, calmly — even by a second person. The limit is obvious: you measure the result, not how it was achieved. The exam may have been taken with help; the dishes, washed by someone else.
Each method errs in a predictable directionThe bias of interval-based methods is not an accident or a flaw in execution: it follows directly from the marking rule. Whole-interval recording tends to underestimate occurrence and duration; partial-interval tends to overestimate; momentary time sampling is usually the least biased of the three for estimating the proportion of time, especially when intervals are short. Because the bias is known, it can be used deliberately: for a behavior you want to increase, whole-interval recording sets a demanding criterion; for a dangerous behavior that must not go unnoticed, partial-interval recording errs on the safe side. What you cannot do is switch methods partway through monitoring and read the difference as progress.
Can this number be trusted?
A recorded data point is not automatically a good data point. Three different questions need to be answered, and confusing them is the most common mistake beginners make:
- Validity — does the number relate to what actually matters? Counting how many tokens a child earned is not the same as measuring whether they learned to ask for something; measuring the number of meltdowns does not capture a reduction in their intensity. The most frequent threat to validity is indirect measurement: measuring a convenient substitute (a questionnaire, a teacher's impression) in place of the behavior itself.
- Accuracy — does the number approximate the true value? This can only be discovered by comparing the recording with a more trustworthy standard, such as a video of the same session recorded with extra care.
- Reliability — does the measure repeat consistently? A stopwatch running fast is perfectly reliable and systematically inaccurate. Consistency without accuracy means making the same error every time.
Three problems silently corrode these qualities. Observer drift is the gradual, involuntary change in how a definition is applied — after a few weeks, what was counted as "aggression" is no longer quite the same thing. The observer's expectation pushes their markings toward the desired result, especially when the same person both applies and evaluates the program. And reactivity changes the very behavior being observed: people act differently when they know they are being watched — and observers record differently when they know they are being supervised.
The standard tool for checking whether the measurement system is being applied consistently is interobserver agreement (IOA): two people record the same session independently and compare results. It is especially useful for evaluating someone who is just starting to record, detecting drift over the weeks, and showing whether the operational definition is clear enough to produce the same number in different hands. Against expectation, however, IOA offers no protection: the remedy there is different — keeping whoever records unaware of the program's phase and the expected result, and not letting the same person both apply and evaluate their own intervention.
Agreement is not the same as truthTwo observers trained together, with the same imprecise definition and the same expectation, can make exactly the same mistake — and show 100% agreement. IOA attests that the measure is reproducible, not that it is accurate; for that, it must be compared with an external standard. The literature does not set a universal threshold: it is conventional to report values above 80%, with 90% or higher taken as desirable, but that is a convention of the field, not a theorem. And a high number obtained in just one session, or always with the same pair of observers, says very little.
Recording that fits real life
The best recording system in the world is useless if no one can keep it up. A teacher with twenty-five students is not going to time every episode; a parent in the middle of a meltdown is not going to fill out a spreadsheet. The practical rule is unforgiving: the easier it is to record, the greater the chance the record will actually exist. A simple tally kept every day is worth more than an elaborate protocol abandoned in the second week.
A few decisions usually solve this: reduce the number of monitored behaviors to the minimum that answers the question; keep the data sheet physically where the behavior happens; choose time sampling when continuous observation is impossible; use a permanent product whenever the behavior leaves a trace; and define who records what, without assuming "someone's keeping track." Training whoever will collect data — with examples and counterexamples, checking agreement before it counts — saves weeks of unusable data.
Behavior recording is health information, and that has consequences. In Brazil, the General Data Protection Law (Lei Geral de Proteção de Dados, LGPD) classifies health-related data as sensitive personal data, subject to heightened protection; the Federal Council of Psychology (Conselho Federal de Psicologia) makes documented records of care mandatory and regulates their storage and confidentiality; and the ethics codes of behavior analysis treat the collection, use, and confidentiality of data as a professional responsibility. In practice, this means: no spreadsheets with full names in messaging groups, no photos of data sheets on social media, and clarity with the family about what is being recorded and why. It is also worth remembering who owns the information — the person receiving services and their family have a right to know what is being written down about them.
From recording to decision
The cycle only closes when someone actually looks at what was collected. Data that turns into a dead archive is wasted work, and it is a more common failure than people imagine: collection continues for months, the spreadsheet grows, and no one turns it into a graph or a decision. Recording without analyzing gives the appearance of rigor without any of its benefits.
The full sequence is short to state and demanding to sustain: define the behavior in observable terms, choose the dimension that answers the question, choose a method that fits the routine, train whoever records and check agreement, turn the numbers into a graph regularly, and — the decisive step — change the plan when the data calls for it. Keep exploring with operational definition, baseline, and functional assessment.
Test your understanding
Quick quiz
Three questions to lock in the concept. You'll see the explanation right away.
Further reading
- Cooper, J. O.; Heron, T. E.; Heward, W. L. Applied Behavior Analysis. 3rd ed., 2020 — chapters on defining and measuring behavior, recording procedures, and measurement quality.
- Baer, D. M.; Wolf, M. M.; Risley, T. R. "Some current dimensions of applied behavior analysis". Journal of Applied Behavior Analysis, 1968.
- Johnston, J. M.; Pennypacker, H. S.; Green, G. Strategies and Tactics of Behavioral Research and Practice. Routledge — foundations of behavior measurement.
- Behavior Analyst Certification Board (BACB). Ethics Code for Behavior Analysts (2022) — data collection and use, documentation, and confidentiality.
- Conselho Federal de Psicologia (Brazil). Resolução CFP nº 001/2009 — mandatory documented record-keeping arising from the provision of psychological services.
- Brazil. Lei nº 13.709/2018 — Lei Geral de Proteção de Dados Pessoais (LGPD), art. 5, II: health-related data is sensitive personal data.
This entry is educational and does not replace assessment, diagnosis or treatment by a qualified health professional. Every person is unique — for a specific case, seek individual guidance.