Data Types Overview

Frequency, duration, latency, intensity, prompt level, permanent product, ABC — and how to know which to use

For paraprofessionals who collect data in classrooms

Why this brief

Data is the bread and butter of a para's day. Every IEP, behavior plan, and progress report rests on it. But the word "data" can mean a dozen different things — sometimes a tally on a clipboard, sometimes a stopwatch reading, sometimes a checked box on a prompt-level sheet. Choosing the wrong type of data, or collecting it the wrong way, makes the data useless and your time wasted.

This brief is the orientation: what each data type measures, when each is the right choice, what each looks like in practice, and what good versus bad data collection looks like for paras specifically. Other briefs in domain 06 dig deeper into individual methods (interval recording, prompt-level data, ABC narrative) — this one helps you see the whole map.

| |

| :-: |

| The data principleData exists to answer questions. Before collecting anything, the team should be able to say what question this data answers. "How often is the behavior happening?" is a real question. "Some kind of behavior data" is not. |

Who this brief is for

Paras who are collecting data on academic, behavioral, communication, or self-help goals

New paras still figuring out what "take data" actually means in their setting

Supervising teachers and case managers helping paras pick the right data type for the question

Anyone reviewing data sheets and wondering if they're useful or just noise

The seven main data types

Most school-based data falls into one of these categories. Each measures something different. Each has a question it answers well and a question it answers badly.

| Type | What it measures | Best for |

| :-: | :-: | :-: |

| Frequency / count | How many times the behavior happened in a defined period | Discrete behaviors with a clear start and stop, occurring at a manageable rate (e.g., raising hand, hitting, asking for help) |

| Rate | Frequency divided by time (e.g., 4 per hour, 12 per day) | Comparing observation periods of different lengths; tracking change over time |

| Duration | How long the behavior lasted | Behaviors where length matters more than count (e.g., on-task, tantrum, sleeping) |

| Latency | Time between a cue and the response | Compliance, processing speed, transition timing (e.g., "how long until she starts the work after the direction") |

| Intensity / magnitude | How big or strong the behavior was, on a defined scale | Behaviors that vary in seriousness (e.g., aggression where a tap and a punch should not be the same data point) |

| Prompt level | What level of help the student needed to do the skill | Teaching skills — the most important data type for fading prompts; tied directly to instruction |

| Permanent product | The thing the behavior produced (worksheet, drawing, completed task) | Academic work, vocational tasks, self-help products — anything that leaves evidence behind |

Plus one method that combines several of these into a narrative format:

ABC (antecedent-behavior-consequence) narrative recording: captures the context around behaviors that need understanding before counting can mean anything. Covered in detail in brief 06.04.

| |

| :-: |

| Match the data to the questionIf you can't articulate what question your data is supposed to answer, you're collecting noise. "How often does Marcus ask for a break?" — frequency. "How long does he stay engaged in math?" — duration. "Can he do the steps with less help over time?" — prompt level. The question dictates the type. |

Frequency and rate

What it is

Frequency is a count: how many times the behavior occurred during the observation period. Rate is frequency divided by time so you can compare across different-length observations. Example: 12 hand-raises in a 30-minute lesson is a frequency; 24 per hour is a rate.

When to use it

The behavior has a clear start and stop (each instance is countable)

It happens at a rate you can actually catch — somewhere between every-few-minutes and every-few-days

You care about how often more than how long

Examples: hand-raises, requests for help, hitting peers, asking off-topic questions, leaving the seat, words spelled correctly

When NOT to use it

The behavior is too fast or too constant to count reliably (vocal stims, fidgeting)

Length matters more than count (on-task vs. off-task — duration is better)

Behaviors blur into each other (one long meltdown vs. four small ones — intensity or duration is better)

How to record it

Tally marks on a sheet — simplest method

Click counter (golf counter on the wrist) — useful when you can't take eyes off the student

Wristband ticker, sticky notes for each instance, paperclips moved across pockets

Apps like BehaviorSnap, Catalyst — if your district uses them

Whatever you use, define the behavior precisely and consistently. "Yelled" is fuzzy. "Used a voice loud enough that the next desk could hear, when not called on" is not.

Duration

What it is

How long a behavior lasted. Total duration adds up all instances across the observation period; mean duration is the average length per instance.

When to use it

Length is what matters: on-task time, sleep time, tantrum length, time spent in self-stimulation

The behavior could last anywhere from a few seconds to a long stretch, and that variability matters

Examples: on-task during independent work, time engaged with a peer, length of meltdowns, time in cool-down

When NOT to use it

Behaviors that are essentially instantaneous (a hit, a yell — count instead)

You only have your eyes available occasionally — duration usually requires close watching

How to record it

Stopwatch — phone or kitchen timer works fine

Start/stop times noted on a sheet

Apps designed for duration recording

If you're tracking on-task or engagement, define what counts as on-task before you start. Eyes on the work? Hands on the materials? Producing output? Be specific or your data will drift between observers.

Latency

What it is

Time between a cue (often a teacher direction) and the student's response. Short latency = quick response; long latency = slow start.

When to use it

Compliance and following directions — "how long does it take to start after I ask?"

Transitions — "how long from the bell to seated and ready?"

Processing-related goals — measuring whether wait time and supports are reducing latency over time

Some communication goals — wait time before initiating a response on AAC

How to record it

Stopwatch starting at the cue, stopping when the student begins the response

Note when the teacher gave the direction; note when the student started

Watch for confounds: did the student hear the direction? Was a peer interrupting? Note these so the data isn't misread later

| |

| :-: |

| Latency is sensitive to the cueMake sure the cue is consistent across observations. A whispered redirection is not the same as a posted schedule cue. If you change the cue you've changed what you're measuring. |

Intensity / magnitude

What it is

How big the behavior was. Almost always measured on a defined scale rather than a precise number — because "intensity" usually means severity, not size.

When to use it

Aggressive or self-injurious behavior where a tap and a punch are not the same

Vocal behavior where loud and quiet matter (calling out vs. shouting)

Crying or distress where mild fussing and screaming-meltdown should be tracked separately

Stereotypy where some forms are interfering and some are not

How to record it

Use a defined scale that everyone agrees on. Common scales:

| Scale | Common version |

| :-: | :-: |

| 1–5 severity | 1 = barely there; 2 = mild, no impact; 3 = noticeable, some impact; 4 = significant, disrupts class; 5 = unsafe, requires response |

| 3-level | Mild / Moderate / Severe with examples for each |

| Specific descriptors | "Open hand" vs. "closed fist"; "redirected with words" vs. "required physical block" |

Whatever scale you use, write down what each level looks like with concrete examples. Otherwise observers drift, and your "3" today might be someone else's "4" tomorrow.

Prompt-level data

What it is

How much help the student needed to do the skill. This is arguably the most important data type for paras to understand because it directly drives instruction. Without prompt data, you can't fade prompts; without fading prompts, you can't build independence.

Common prompt hierarchies

| Most-to-least intrusive | Examples |

| :-: | :-: |

| Full physical (FP) | Hand-over-hand; you do the skill with the student |

| Partial physical (PP) | You guide the elbow or wrist; student finishes the motion |

| Modeling (M) | You demonstrate; student copies |

| Gestural (G) | Pointing or visual cue without speaking |

| Verbal direct (VD) | "Pick up the pencil" |

| Verbal indirect (VI) | "What do you do next?" |

| Independent (I) | Student does it without any help |

Different programs use slightly different scales. The principle is the same: rank from most help to least help, and track which level each opportunity required.

How to record it

During teaching, after each opportunity to perform the skill, mark the lowest level of prompt that worked. Sample data sheet for putting on a shirt across 10 trials:

| Trial | Prompt level |

| :-: | :-: |

| 1 | FP |

| 2 | FP |

| 3 | PP |

| 4 | PP |

| 5 | M |

| 6 | M |

| 7 | G |

| 8 | VD |

| 9 | VD |

| 10 | I |

This data tells the team: she's progressing, she's at gestural-to-verbal across opportunities, fade physical prompts entirely on next session. Without prompt-level data, you'd just have "she's working on dressing" — useless for adjusting instruction.

| |

| :-: |

| Be honest about promptsThe biggest data integrity problem with prompt levels is over-prompting then recording "independent." If you said the answer first or guided their hand, that's not independent — that's verbal direct or partial physical. Document the help you actually gave. |

Permanent products

What it is

The output of the behavior — a worksheet, a completed task, a wash-cycle of laundry, a journal entry. Permanent products let you score the data later, often without needing to watch the behavior happen.

When to use it

Academic work — completed problems, words written, reading passages decoded

Vocational tasks — items packaged, copies made, tables wiped

Self-help routines that produce something — laundry done, lunch eaten

Communication — words spoken (recorded with consent) or AAC outputs (logged by the device)

Advantages

Doesn't require you to watch every minute

Can be scored by anyone using the same rubric

Provides a defensible record (you have the actual work)

Can be scored after the fact, by the teacher rather than the para in real time

Cautions

Can't tell you about the process — was the work copied from a peer? Was it done with hand-over-hand support?

Should be paired with prompt-level data for skills being taught

Define what counts as "complete" or "correct" before scoring — vague rubrics produce vague data

ABC narrative recording (cross-reference)

ABC recording is its own brief (06.04). The short version: when a behavior is concerning but the team doesn't yet understand it, you record the antecedent (what happened just before), the behavior (what the student did), and the consequence (what happened next). Patterns across many ABC entries inform a function-based hypothesis (see brief 05.01).

ABC is a narrative method, not a count. Use it when you need to understand a behavior, not just track it. Once a function is hypothesized and a plan is in place, you usually shift to frequency, duration, intensity, or prompt-level data tied to the plan's measurable outcomes.

Choosing the right data type

This is the question that paras get wrong most often — usually because someone handed them a data sheet without explaining why this one. A few ways to think it through:

Start with the question

| Question | Data type | Why |

| :-: | :-: | :-: |

| How often is X happening? | Frequency / rate | Counts answer how-often |

| How long is he engaged in Y? | Duration | Length is the variable of interest |

| How fast does she respond after a direction? | Latency | Time-to-start is what we're tracking |

| How serious is each incident? | Intensity | Severity needs a scale, not just a count |

| How much help is he needing? | Prompt level | Independence is the goal; prompts measure progress |

| Did she finish the task correctly? | Permanent product | The work itself is the data |

| Why does this happen — what's the function? | ABC narrative | Context and pattern are the question |

Match data type to behavior characteristics

| If the behavior... | Use |

| :-: | :-: |

| Is rare and discrete (a few per day) | Frequency |

| Is fast and high-rate (several per minute) | Interval recording (see 06.02) or rate over short windows |

| Has a duration that varies and matters | Duration |

| Has different levels of severity | Intensity |

| Is a learned skill being prompted | Prompt level + opportunity count |

| Produces a tangible output | Permanent product |

| Isn't yet understood functionally | ABC narrative |

Two data types are often better than one

Many goals benefit from tracking two metrics. Examples:

Aggression: frequency + intensity. "He hit twice today, once a 2 and once a 5."

On-task: duration + permanent product. "She was on-task 22 minutes of 30, completed 8 of 10 problems."

Asking for breaks: frequency + latency. "He asked 6 times today; average wait 4 minutes after a difficult task."

Toileting: frequency + prompt level. "Three trips, two with verbal cue, one independent initiation."

Data quality — what makes data trustworthy

Bad data is worse than no data because it can lead the team to wrong conclusions. Some hallmarks of trustworthy data:

Operational definitions

Every behavior on a data sheet should be defined precisely enough that two people watching the same student would record the same thing. "Disruption" is not an operational definition. "Calling out without raising a hand, defined as words audible to the next desk over" is.

Inter-observer agreement

When more than one person collects data on the same behavior, they should agree on what they saw at least 80% of the time (some standards say 90%). If two paras watching the same hour come back with very different counts, the operational definition needs sharpening or training. The teacher or BCBA usually checks this periodically.

Honest, real-time recording

Data filled in from memory at the end of the day is rarely accurate. People misremember the bad days as worse and the good days as better. Record as it happens, even if it's a tally on a sticky note that gets transferred later.

Sustainability

Data that's too complex to actually take during a real classroom day will not get taken. Better to track one variable well than five variables half-way. The simplest data sheet that answers the question is usually the right one.

Respect for context

Sometimes the data isn't taken because something else mattered more — a student in crisis, a fire drill, a substitute teacher. Note it. "No data, school assembly" is honest. Inventing data to fill gaps is dangerous and dishonest.

| |

| :-: |

| If the data sheet is broken, fix itIf you can't make the data sheet work in your real day, that's information for the supervising teacher and BCBA. The fix is simplifying the sheet, changing the data type, or rethinking the timing — not white-knuckling through bad design. |

Privacy and confidentiality

Data sheets are educational records under FERPA (see brief 13.01). Treat them like the IEPs they're tied to.

Use student initials or codes on sheets visible in the classroom

File completed sheets in the student's secure folder; don't leave them out

Don't photograph data sheets to your personal phone or email

Don't discuss specific data with anyone who doesn't have an educational need to know — including other paras

Recording video or audio for data requires district-level consent and family consent; don't improvise

Pitfalls

| Try this | Watch out for |

| :-: | :-: |

| Pick the data type that answers the actual question | Use whatever data sheet was on the clipboard last year |

| Define the behavior operationally before counting it | Use vague terms like "disruption" or "non-compliance" |

| Record in real time as the behavior happens | Fill in the sheet from memory at the end of the day |

| Be honest about the level of help you provided | Mark "independent" when you actually said the answer first |

| Pair frequency with intensity for behaviors that vary in severity | Treat a tap and a punch as the same data point |

| Track only what you can sustainably collect | Try to track 8 variables and end up tracking nothing reliably |

| Note context — substitutes, assemblies, illness — when relevant | Pretend the day was normal when it wasn't |

| Ask the supervising teacher / BCBA when the sheet doesn't fit reality | Quietly modify the data system without telling anyone |

| File data sheets like the educational records they are | Leave clipboards on the table or in the hallway |

| Use prompt-level data when teaching skills | Just track "yes/no" on whether they did the skill — that misses the dose of help |

Scenarios

Scenario 1: A vague data sheet for hand-raising

You've been given a data sheet that says "Hand-raising — yes/no per period." The student raises his hand sometimes appropriately, sometimes after blurting, and sometimes for completely off-topic comments.

This sheet doesn't answer a useful question. What does "yes" mean — at least one hand-raise? Bring this back to the supervising teacher and ask what the goal actually is. If it's increasing appropriate hand-raising, you want frequency of appropriate hand-raises (operationally defined: hand raised before speaking, on-topic). If the goal is reducing blurting, that's a different behavior to count. The sheet should match the goal.

Scenario 2: Hard-to-count high-rate behavior

A student vocally stims roughly continuously throughout the morning. The team wants data on whether it's getting worse.

Frequency won't work — you can't count hundreds of vocalizations per hour reliably. Duration of total vocal-stim time would work but requires watching. Interval recording (see 06.02) is probably the best fit: divide the period into 30-second intervals and mark whether vocal stim occurred during each interval. This gives you a percentage that can be compared over time without burning out the para.

Scenario 3: Independence on a self-help skill

A student is learning to brush teeth. The IEP goal is "will independently complete tooth-brushing routine in 8 of 10 trials."

Permanent product alone won't tell you what's happening — you need prompt-level data per step. Break the routine into steps (get toothbrush, apply paste, brush all surfaces, rinse, spit, put away) and mark the prompt level for each step each trial. Now you can see: she's independent on getting the toothbrush and paste, but still at modeling for brushing molars and partial physical for rinsing. Instruction targets the steps still requiring help.

Scenario 4: Aggression that varies

A student is occasionally aggressive. The frequency hasn't changed but the team thinks the severity has.

You're tracking frequency only — that's why the change isn't showing up. Add intensity. Define a 1–5 scale (1 = open-hand tap, 5 = closed-fist or weapon use). Now you can see: same number of incidents, but average intensity has gone from 1.5 to 3.2 over a month. That's data that supports a plan revision.

Scenario 5: A behavior the team doesn't yet understand

A student has started slamming his head on the desk during math but not during reading or other subjects. The team has not yet figured out the function.

Frequency or duration alone will tell you it's happening but not why. Run ABC narrative recording (see brief 06.04) for a couple of weeks. What was happening just before each incident? What followed? Patterns will emerge — maybe demands above his level, maybe specific assignments, maybe times of day. Once a function is hypothesized, switch to a data system tied to the plan.

Scenario 6: Multiple paras, drifting data

Three paras rotate through your student's day. The data is wildly inconsistent — one para is reporting 4 hand-raises a day, another is reporting 30.

Either the operational definition isn't tight, or someone is counting differently. Get the team together with the supervising teacher and BCBA. Watch the same period together. Discuss what counts. Write the operational definition on the data sheet itself. Take inter-observer agreement data on a few sessions until everyone matches at 80%+.

Closing thought

Good data is one of the most powerful things paras contribute to a team. Teachers and BCBAs can't be everywhere — paras can. The data you take, if it's accurate and tied to a real question, drives instruction, behavior plans, IEP goals, and family communication. The data you take poorly — vague, after-the-fact, or unhonest about prompts — drives the team toward bad decisions for the student.

The hardest part is usually picking the right type for the question and writing operational definitions you can stick to. Once those are in place, the actual collection becomes part of the rhythm of the day.

| |

| :-: |

| Bottom lineData exists to answer questions. Match the type to the question. Define the behavior precisely. Record in real time. Be honest about prompts. Pair types when one isn't enough. Sustain it. File it carefully. Push back when the system isn't working. |

Related briefs

06.02 Interval Recording — for high-rate or long-duration behaviors

06.03 Prompt-Level Data — deeper dive on the most important data type for fading

06.04 ABC Narrative Recording — when you don't yet know the function

06.05 IEP Progress Monitoring — tying data to goals

05.01 Function-Based Thinking — what data is for, in behavior support

04.02 Prompting Hierarchies — the prompt levels you'll be coding

12.06 Working with the BCBA — your collaborator on data systems

13.01 FERPA and Confidentiality — data sheets are educational records

Page of