ABA data collection methods: how to record and calculate each one

01 / Two families

Continuous and discontinuous measurement

The BACB’s RBT Test Content Outline (3rd ed.), which the RBT examination follows from 2026, opens with a Data Collection and Graphing domain worth 13 of the 75 scored questions. Its first tasks ask an RBT to implement continuous measurement (for example frequency, duration, latency, and interresponse time), discontinuous measurement (partial and whole interval, momentary time sampling), and permanent product recording, then to enter data, update graphs, and calculate and summarize data as rate, mean duration, or a percentage.

Whichever family you use, the method is not your choice to make in the moment. The operational definition says what counts as the behavior; the plan says how to measure it. An RBT’s job is to apply both the same way every session, and to tell the supervisor when the definition and what actually happens in the room stop matching.

Need paper? Our free printable ABA data sheets cover every method in this guide, one page each, with the summary formula printed on the sheet.

Domain names and task wording follow the BACB’s RBT Test Content Outline (3rd ed.), updated 09/2024.

02 / Frequency and rate

Frequency: count every occurrence, then turn it into a rate

Frequency recording is a tally. Each time the behavior meets its definition, it counts once. It suits behaviors with a clear start and end that do not last long, such as hand raises, requests, hits, or answers called out. It fits poorly for behaviors that run on for minutes, like crying or staying in a seat; duration or interval methods work better there.

A raw count depends on how long you watched, so it is usually reported as a rate: count divided by observation time.

Divide by the whole time you were observing, not by the gap between the first and last occurrence. If the client leaves the room for five minutes and you cannot observe, follow your team’s rule for that time; many plans subtract it from the observation time and ask you to note why.

03 / Duration

Duration: time each bout from onset to offset

Duration recording measures how long the behavior lasts. Start the timer at the onset the definition describes and stop it at the offset. When the behavior happens several times, each one is a separate bout, and the plan may ask for the total, the mean per bout, or the percentage of the session.

The classic duration error is a timer left running. If you realize you forgot to stop it, do not guess a new number silently: correct it the way your team documents corrections and add a short note. The second classic error is deciding onset and offset by feel; if the definition says the bout ends after 3 seconds of no crying, wait the 3 seconds.

04 / Latency

Latency: time from the cue to the response

Latency is the time between a cue, such as an instruction, and the start of the response. It is useful when the question is not whether the client responds but how quickly, for example following a transition instruction. Start timing when the cue is delivered, stop when the response begins, and repeat for each cue.

Do not enter a made-up time for a cue that got no response, and do not quietly drop it either. Plans handle it differently (some cap the time, some count it as a separate outcome), so ask how yours wants it recorded. Interresponse time, the time between two responses, is measured the same way, from the end of one response to the start of the next.

05 / Percent correct

Trial-by-trial data and percent correct

For skill targets, each learning opportunity is a trial, and each trial is scored correct or incorrect under the program’s rules. The summary is a percentage: correct trials divided by total trials, multiplied by 100.

Both targets show 80%, but 4 of 5 rests on far fewer trials than 16 of 20, so write the fraction next to the percentage. A session with no opportunities is “no data”; recording it as 0% would suggest the client got every trial wrong.

Many ABA and speech programs score more than correct or incorrect: the prompt level used, whether a response was independent, or the cueing an SLP set. If your program uses a prompt hierarchy, score exactly the categories on your data sheet. The same logic applies to speech-language pathology assistants. ASHA’s SLPA Scope of Practice lists “collecting data and calculating percentages for the SLP to use” and preparing charts, records, and graphs as SLPA responsibilities, and places interpreting results and modifying the plan of care with the supervising SLP. For oral reading fluency and running records, which use their own formulas, see our guide to calculating WCPM and reading accuracy.

SLPA responsibilities are quoted from ASHA’s Speech-Language Pathology Assistant Scope of Practice (2022).

06 / Interval recording

Whole-interval, partial-interval, and momentary time sampling

Interval methods divide the observation into equal intervals, for example twenty 30-second intervals in 10 minutes, and give each interval one score. The three methods differ only in the rule for scoring “yes.”

These biases are measured, not folklore. Powell, Martindale, and Kulp (1975) compared the three methods with a continuous record of in-seat behavior: whole-interval sampling consistently underestimated it, partial-interval sampling consistently overestimated it, and momentary samples missed high and low about equally often. The more often samples were taken, the closer every method came to the continuous record.

The summary is the percentage of intervals scored yes. The detail that trips people up is the interval you could not score because you were blocking, prompting, or looking away.

Whether unobserved intervals are excluded, repeated, or handled another way is your supervisor’s rule. What matters is that you never fill them in from memory and that the count of unscored intervals appears next to the percentage. For momentary time sampling, score at the sample moment itself; a look two seconds late is a different sample.

Bias findings: Powell, J., Martindale, A., & Kulp, S. (1975). An evaluation of time-sample measures of behavior. Journal of Applied Behavior Analysis, 8(4), 463–469.

07 / Recording errors

Common data collection errors and how to avoid them

The Test Content Outline asks RBTs to describe the risks of unreliable data collection and poor procedural fidelity. A supervisor can only make good decisions from data that reflects what happened. These are the errors that most often creep in during a busy session:

1. Recording from memory after the session

   Counts made at the end of a session drift toward what you expected to see. Record each occurrence, trial, or interval as it happens.

2. Definition drift

   Over weeks, “a hit” can quietly start to include a touch. Reread definitions regularly and ask when an occurrence does not clearly fit.

3. Timers left running or never started

   Check that every duration and latency timer is in the state you think it is before you look away, and stop open timers before ending the session.

4. Guessing missed intervals

   An interval you did not observe is unscored, not “probably no.” Report how many there were.

5. Percentages without denominators

   80% of 5 and 80% of 20 are not the same evidence. Always keep the count behind the percentage, and record “no data” rather than 0% when there were no opportunities.

6. Silent corrections

   If you tapped twice or scored the wrong target, correct it in the way your team documents corrections so the record shows a change was made.

7. Missing context

   The outline also asks RBTs to report variables that might affect progress, such as illness, medication, or schedule changes. A one-line objective session note gives your supervisor that context.

08 / Privacy

Keep client information to the minimum

Session data is confidential client information. The Test Content Outline expects RBTs to identify and comply with requirements for collecting, using, storing, protecting, and disclosing confidential information, and ASHA’s SLPA Scope of Practice lists compliance with HIPAA and FERPA among an SLPA’s responsibilities and bars sharing confidential information with anyone the SLP has not approved. In practice, your employer’s policy decides where data may be recorded, and its official system is usually where the record belongs.

A code or initials limits what a lost phone or a misdirected file reveals, but it does not make the record anonymous. HHS guidance on HIPAA de-identification says a data set containing patient initials does not meet the Safe Harbor method, and session dates are identifying details too. Treat a coded record as confidential client information.

De-identification guidance: HHS, Guidance Regarding Methods for De-identification of Protected Health Information (archived copy of 2026-08-25). Confidentiality task F.5: BACB RBT Test Content Outline (3rd ed.). SLPA duties: ASHA SLPA Scope of Practice.