ABA Data Collection Sheets: A Worked Example per Method

If you search for ABA data collection sheets, you want a page you can lift a sheet from. This post gives one worked sheet for each common data collection method, with example data filled in and the one calculation you do at the end. The learners in the examples are fictional and the numbers are invented to show the arithmetic.

First, a way to choose. A method is a match between the behavior and the question, so pick the dimension you care about before you pick a sheet:

You want to know	Use
How many times it happens	Frequency (event recording), reported as a rate if sessions vary in length
How long it lasts	Duration
How quickly it starts after a cue	Latency
Whether it happens during a time window, when you cannot count every instance	Partial or whole interval, or momentary time sampling
How a skill is performing across teaching trials	Trial-by-trial data
Which step of a chain is the problem	Task analysis data
What tends to come before and after a behavior	ABC data
What a behavior left behind	Permanent product

For the theory behind these choices, see continuous versus discontinuous measurement. Here we stay practical.

Sheet 1: Frequency (event recording)

Use it for behavior with a clear start and stop that is countable: hitting, requests, bites of food taken on request.

Columns: date, observer, start time, end time, tally, total, minutes observed, rate.

Date	Start	End	Tally	Total	Minutes	Rate per hour
10/5	9:00	10:00	||||	4	60	4.0
10/6	9:00	9:45	|||| |	6	45	8.0

Compare the two days. Raw counts say 4 versus 6, and the second looks worse. Per hour it is 4.0 versus 8.0, which is much worse, because the second session was shorter. Rate is count divided by observation time; use it whenever session lengths differ. More on this in frequency, rate, duration and latency.

Sheet 2: Duration

Use it for behavior where the length is the problem: tantrums, on-task engagement, time out of seat.

Columns: episode number, onset, offset, seconds. Define the onset and offset criterion in writing, for example "begins at first scream; ends after 10 consecutive seconds of quiet".

Episode	Onset	Offset	Seconds
1	9:12:00	9:14:10	130
2	9:40:20	9:45:05	285
3	10:20:00	10:21:05	65

Total duration is 480 seconds (8 minutes). Mean duration per episode is 160 seconds. If the session was 60 minutes, the behavior filled 480 of 3,600 seconds, or 13.3 percent. Report both the count of episodes and the mean length, because "fewer but longer" and "more but shorter" tell opposite stories.

Sheet 3: Latency

Use it for compliance and response speed: how long from the instruction to the start of the response.

Columns: trial, instruction given (time), response began (time), seconds.

Trial	Instruction	Response began	Latency (sec)
1	10:02:00	10:02:08	8
2	10:15:30	10:15:42	12
3	10:31:10	10:31:15	5
4	10:44:00	10:44:20	20

Mean latency is (8 + 12 + 5 + 20) / 4 = 11.25 seconds. Latency is the time before the response starts, not how long it takes to finish; time to finish is duration. If the instruction is repeated, decide in advance whether the clock restarts and say so on the sheet.

Sheet 4: Partial-interval recording

Use it for high-rate or hard-to-count behavior where you only need to know if it happened in the window. A "1" means the behavior occurred at any point in the interval.

Columns: interval number, interval length, occurrence (Y or N). Fix the interval length in advance, such as 10 seconds across 10 intervals.

Interval	1	2	3	4	5	6	7	8	9	10
Vocal outburst	Y	N	Y	Y	N	N	Y	N	Y	Y

Six of ten intervals contain the behavior: 60 percent of intervals. Partial interval tends to overestimate how much of the time the behavior occurs, so use it when the goal is reduction and you want the estimate to err on the high side. Details in partial versus whole interval recording.

Sheet 5: Whole-interval recording

Use it for behavior you want to increase and need to be present for the entire window, such as staying seated or on task. Score Y only if the behavior lasted the whole interval.

Interval	1	2	3	4	5	6	7	8	9	10
On-task	Y	Y	N	Y	Y	N	Y	Y	Y	N

Seven of ten intervals: 70 percent. Whole interval tends to underestimate occurrence, so the number is conservative for a behavior you want more of. Never compare a partial-interval percentage with a whole-interval percentage; they measure differently.

Sheet 6: Momentary time sampling

Use it for situations where one staff member watches several learners or cannot keep eyes on one learner continuously. You record only what is happening at the instant each interval ends.

Moment (every 60 sec)	1	2	3	4	5	6	7	8	9	10
Engaged with materials	Y	N	N	Y	N	Y	N	N	Y	N

Four of ten moments: 40 percent. Use a timer that signals the moment, such as a vibrating timer or interval beep, because looking up at random introduces bias. See momentary time sampling for when it is trustworthy.

Sheet 7: Trial-by-trial (discrete trial) data

Use it for acquisition programs run in trials. Code each trial by response and prompt level so that the sheet shows independence, not just correctness.

Columns: target, then one cell per trial coded +, P (prompted) or − (incorrect). Add a key at the top of the sheet and use the same key across the practice.

Target	1	2	3	4	5	6	7	8	9	10	% independent
Receptive: "touch cup"	+	+	P	+	-	+	P	+	+	P	60%

Six of ten were independent. Report percent independent, not percent correct, whenever prompts are in use; prompted correct responses are not the same as the skill. If you track a prompt hierarchy, record the level used on each prompted trial. See prompt hierarchies and prompt fading.

Sheet 8: Task analysis data

Use it for chained skills like handwashing, dressing or brushing teeth. List the steps in order and score each step, so the data shows where the chain breaks.

Step	Mon	Tue	Wed
1. Turn on water	I	I	I
2. Wet hands	I	I	I
3. Pump soap	P	I	I
4. Rub hands, 10 seconds	P	P	P
5. Rinse	I	I	I
6. Turn off water	I	I	I
7. Take towel	P	P	I
8. Dry hands	I	I	I
Steps independent	5 of 8	6 of 8	7 of 8

Across the three days, steps independent went from 5 to 7 of 8, and step 4 never moved. That is the information a single percent hides: the team should teach step 4 directly. Compare task analysis and chaining.

Sheet 9: ABC data

Use it for understanding function, usually when starting an assessment or when a behavior is new. It is descriptive, not a measure of how much behavior occurs, so do not use it as your outcome measure.

Time	Antecedent (what happened before)	Behavior (observable, no interpretation)	Consequence (what happened after)
9:12	Teacher gave a worksheet at the table	Pushed worksheet off desk, yelled "no" (about 20 seconds)	Teacher removed the worksheet and walked away
11:40	Adult talking to another student, learner sitting alone	Threw a block toward the floor	Adult turned and said "stop that" and sat beside learner
13:05	Snack was placed out of reach	Reached, cried	Adult handed the snack

Written as observable events ("pushed the worksheet") rather than conclusions ("was angry"), these three rows hint at three different possible functions: escape, attention and access. A few rows suggest hypotheses; the hypotheses are tested by the BCBA, not decided by the sheet.

Sheet 10: Permanent product

Use it for work that leaves a record: worksheets, completed chores, a made bed, written responses.

Date	Product	Items attempted	Items correct	% correct
10/12	Math worksheet	20	18	90%

The advantage is that you can score it after the session, which is why it suits settings where nobody can be observing at that instant. The disadvantage is that it tells you nothing about how the product was made unless you also observe.

Choosing interval length and session layout

Interval methods fail most often because the interval is wrong for the behavior. A rough guide that practitioners use: the shorter the typical behavior, the shorter the interval needs to be, because a long interval will almost always contain at least one occurrence and a partial-interval score will saturate near 100 percent. When that happens the data cannot show improvement. Pilot your interval for one or two sessions, and if almost every interval is a Y, shorten it. Your supervising BCBA sets the final choice; there is no universal interval length.

Lay out every sheet so the observer never has to hunt. Put the learner code, date, observer and session start at the top; the operational definition and the key under that; the data grid in the middle; and the summary calculation, filled in at the end of the session, at the bottom. Leave a notes line for anything unusual such as illness, a substitute staff member, or a fire drill, since those explain data points on the graph later. For multi-program sessions, give each program its own sheet rather than squeezing several onto one; mixed sheets are where codes get confused.

Making the sheets reliable

• Operational definitions on the sheet. If two people would score the same moment differently, the definition is not tight enough. Put it in the header.
• One key, everywhere. Pick codes (+, P, -, I) and never change them between learners.
• Record at the time. Memory-based data is guesses. The sheet should be fast enough to use during the session.
• Check agreement. A second observer scoring a share of sessions shows whether the data is believable. See how to calculate IOA for the formulas, and ABA IOA software if you want agreement scored for you.
• Graph. Data you never look at does not guide decisions. Summarize each sheet into a point on a graph the same day.

When paper stops scaling

Paper sheets are fine for a single learner and cheap to start. They get hard when you have a team, many programs per learner, and a requirement to keep records and graphs. If you are comparing electronic options, our ABA data collection software page describes how Wilma collects data on a phone or tablet. A phone is not required to collect good data, and a good paper sheet beats a poor app.