DRAGENwiki · Article

How progress is measured

Sources checked. Clinical review pending.

In ABA, progress is measured by counting or timing real behavior, again and again, and putting it on a graph, so everyone can see whether the plan is working.

Measuring progress, in 60 seconds

  • Data are numbers about real behavior: how often it happens, how long it lasts, how fast it starts, or how much of the time it is going on. [2]
  • The measure has to fit the question. "The wrong measure hides the change you're looking for." [2]
  • Data must drive decisions. Behavior analysts "graphically display, summarize, and use the data to make decisions about continuing, modifying, or terminating services." [1]
  • Data get checked. Two people sometimes record the same session to see whether they agree. [3]
  • Real progress shows up in real life: with different people, in different places, and weeks later. [1][5]

Jump to: Why measure at all · Four ways to measure behavior · Choosing the right measure · Measuring new skills · Checking the data: two observers · From numbers to a graph · Beyond the graph: does it last and does it matter? · What you can measure at home · Myths · Words to know · Sources

Why measure at all

Memory is kind, and it is often wrong. A hard week can make a good month feel bad. A good day can hide a slow slide. Numbers taken the same way every session show what is really happening.

That is why the ethics code makes data part of the job. Behavior analysts "actively ensure the appropriate selection and correct implementation of data collection procedures." [1] When the data show a goal is not being reached, they must "actively assess the situation and take appropriate corrective action." [1]

For families, data does something else too: it lets you see the answer yourself, instead of taking someone's word for it.

Four ways to measure behavior

DRAGEN's free lesson How to count a behavior so the number tells the truth explains the main measures. [2] In short:

  • Frequency, or count. How many times it happened. Best for behavior with a clear start and end that comes and goes quickly: hitting, hand-raising, words. If sessions are different lengths, divide by the minutes to get a rate. "4 in 20 minutes is 0.2 a minute." [2]
  • Duration. How long it lasted, from start to finish. Best when length matters more than how often: tantrums, time on task, time in seat. [2]
  • Latency. How long it took to start, from the end of an instruction to the first move. Best for following directions or starting homework. [2]
  • Interval recording. The session is split into equal boxes, such as 10 seconds each, and each box is marked yes or no. It is used when a behavior is too frequent or too blurry to count, or the adult is busy teaching. "It estimates; it doesn't count." [2]

Interval recording comes in three kinds [2]:

  • Partial interval: mark the box if the behavior happened at any moment. "It reads high."
  • Whole interval: mark the box only if the behavior lasted the whole box. "It reads low."
  • Momentary time sampling: look only at the last moment of each box. The easiest while teaching; it can miss in either direction.

Choosing the right measure

The same behavior can tell two stories, depending on how it is measured.

The lesson's made-up example: a child has 3 tantrums a day in week 1, each about 20 minutes. In week 4, still 3 a day, but each about 2 minutes. [2]

  • Count says: nothing changed. 3 a day, then 3 a day.
  • Duration says: tantrums dropped from 60 minutes a day to 6. That is 90% less.

"Same child, same weeks; the measure decided the story." [2] If your child's report says "no change," it is fair to ask: "Are we measuring the thing that matters?"

Measuring new skills

Skills are often measured a little differently from problem behavior.

  • Trial by trial. Each chance to do the skill is one trial, marked correct on their own, an error, or with help. Only correct-on-their-own counts toward the percentage.
  • Prompt level. Help is part of teaching. "A prompt is extra help that makes the right response more likely: a spoken hint, a point, showing how, or guiding their hands." [4] Good data show how much help was needed, so the help can be faded on a plan. "Help that never fades becomes part of the skill." [4]
  • Cold probes. The first try of the day, before any teaching, shows what the child can do on their own.
  • Task steps. For long skills like handwashing or dressing, each step is scored, so you can see which steps are learned.

Ask how a skill is being measured, and whether the number means "alone" or "with help." It makes a big difference.

Checking the data: two observers

Data are only useful if they are true. People make mistakes, and researchers say so plainly: data collection by human observers "is likely to be more prone to error than machine transduction." [6]

So ABA teams check. In interobserver agreement (IOA), two people watch the same session, each records on their own, and they compare. The result is a percentage. [3] The free lesson How to check your data with interobserver agreement explains: "Every decision on a graph assumes the dots are true." [3]

How IOA is calculated matters. A simple total count can look perfect even when the two people counted different moments. [3] Research comparing methods found that "Total reliability was uniformly high," while "exact-agreement reliability was the lowest of the measures, especially for high-rate responding." [6] Stricter methods catch more problems.

A common target: "at least 80% agreement," checked "in at least 20% of sessions," spread across conditions and observers. The lesson adds: "Those are common research conventions, not a law; your payer or program may set its own." [3]

Low agreement is not someone's fault. It usually means the behavior's definition is fuzzy. The fix is a clearer definition, a short retraining, and another check. [3]

From numbers to a graph

Data become a graph: one dot per day or session, with a vertical line where something changed, like a new plan. [5] Read three things, in order: the level (where the dots sit), the trend (which way they are going), and the variability (how much they bounce). [5]

For a full walk-through, see reading a progress report and the free lesson How to read a behavior graph in sixty seconds.

Beyond the graph: does it last and does it matter?

A rising line in one room is a start, not the finish. Real progress has three more tests.

  1. Does it show up elsewhere? "Generalization means a skill shows up where it wasn't taught: with other people, in other places, with other materials." [7]
  2. Does it last? "Maintenance means it's still there weeks later." [7] The ethics code asks behavior analysts to design interventions "to produce outcomes likely to maintain under naturalistic conditions." [1]
  3. Does it matter to your child and family? This one is DRAGEN's question. A goal can be met on paper and still not change daily life. Ask your child, in whatever way they communicate, and ask yourself: is life better, easier, or more free?

As the generalization lesson puts it: "A skill that only happens with one adult, in one room, isn't really learned yet." [7]

What you can measure at home

You do not need special training to take useful data. The graph lesson suggests a simple home tally: pick one short stretch of time each day, like homework or bath time, and count one behavior the same way every day. [5]

Tips from DRAGEN:

  • Pick one behavior, described so a stranger could count it.
  • Same time, same length, every day. Twenty minutes is plenty.
  • Write the number down right away, with the date.
  • Bring it to your next meeting. Your home numbers are real data. They show what happens outside of sessions.

For RBTs and BCBAs

Use this at your next supervision.

  • Match the measure to the question. Count, rate, duration, latency or intervals. [2]
  • Define it so a stranger could score it. Fuzzy definitions cause low agreement. [3]
  • Use a strict IOA method, not only total count. [3][6]
  • Graph it the same week, not at the end of the month.
  • Measure generalization and maintenance on purpose. Probe with new people, places and materials, and check again weeks later. [7]

Common myths

  • "More data is always better." Not if it is the wrong measure. The right measure, taken well, beats piles of the wrong one. [2]
  • "A percentage tells the whole story." Not without the number of chances and how much help was given. [4]
  • "If the RBT says it's going well, it's going well." Ask for the graph. Data, not impressions, decide. [1]
  • "Data is just for insurance." Data is how your child's plan gets better. Behavior analysts must use it to make decisions. [1]

Words to know

  • Frequency (count): how many times a behavior happened. [2]
  • Rate: count divided by time. [2]
  • Duration: how long a behavior lasted. [2]
  • Latency: how long it took to start after an instruction. [2]
  • Interval recording: marking equal time boxes yes or no; an estimate. [2]
  • IOA (interobserver agreement): how often two independent observers agree, as a percentage. [3]
  • Prompt: extra help that makes the right response more likely. [4]
  • Generalization: a skill showing up where it was not taught. [7]
  • Maintenance: a skill still there weeks later. [7]

Need a PDF?

Create it here, for free: the DRAGEN app. Use it at your next session.

Or print one we already made, also free:

More from DRAGEN

Related DRAGENwiki pages: reading a progress report · writing behavior goals · data collection for teachers · what an RBT does · when to end or change services · extended school year

What remains unknown

  • IOA targets like 80% and 20% are research conventions, not law. Programs and payers may set their own. [3]
  • The best measure for a given behavior is a judgment call made from the goal. Ask your BCBA why they chose theirs.
  • This page was last checked on November 12, 2026.

Sources

  1. Behavior Analyst Certification Board. Ethics Code for Behavior Analysts (effective January 1, 2022; updated August 2024). https://www.bacb.com/wp-content/uploads/2022/01/Ethics-Code-for-Behavior-Analysts-240830-a.pdf
  2. DRAGEN Academy. How to count a behavior so the number tells the truth. https://dragen.cloud/learn-measurement
  3. DRAGEN Academy. How to check your data with interobserver agreement. https://dragen.cloud/learn-ioa
  4. DRAGEN Academy. How to help without doing it for them: prompting and prompt fading. https://dragen.cloud/learn-prompting
  5. DRAGEN Academy. How to read a behavior graph in sixty seconds. https://dragen.cloud/learn-graphs
  6. Rolider, N. U., Iwata, B. A., & Bullock, C. E. (2012). Influences of response rate and distribution on the calculation of interobserver reliability scores. Journal of Applied Behavior Analysis, 45(4), 753. https://doi.org/10.1901/jaba.2012.45-753
  7. DRAGEN Academy. How to make a skill work everywhere: generalization. https://dragen.cloud/learn-generalization

Pages that link here