There is no single best measurement procedure. There is only the best fit for this goal, this behavior, and this setting.
Why it matters
A measure can be accurate and reliable and still be the wrong measure. If a goal is to shorten tantrums and you only count them, the data can sit flat while the tantrums get much shorter. If you pick a perfect procedure that nobody in the classroom has time to run, you end up with gaps, guesses, or nothing at all.
A selection problem usually comes with a goal, a description of the behavior, and a constraint (one teacher, many students, sessions of different lengths, a behavior with no clear start or end). The right procedure fits all three.
Key ideas
- Start with the goal. Ask which dimension of the behavior the intervention is meant to change. That is the critical dimension, and your measure must capture it.
- How often it happens (repeatability): count or rate.
- How long it lasts (temporal extent): duration.
- When it happens relative to a cue or another response (temporal locus): latency or interresponse time.
- Then look at the behavior. Counting works best for responses with a clear beginning and end that occur at a pace an observer can track. Very high-rate responses, or behavior without clear boundaries (such as engagement or stereotypy that runs on and off), are often better captured by duration or a time-sampling method.
- Then look at the real-world limits. Who is observing, how many clients they are watching, how long they can watch, and what tools they have all matter. A teacher running a lesson cannot watch one student without a break.
- Prefer continuous measurement when you can. Event recording, duration, and latency capture every instance within the observation. When limits make that impossible, a discontinuous method gives an estimate instead.
- If you must estimate, choose the error you can live with. Partial-interval recording tends to overestimate how much of the session a behavior filled; whole-interval recording tends to underestimate it. Momentary time sampling can err in either direction and usually tracks duration better when intervals are short.
- Match the unit to how sessions vary. If observation periods differ in length, convert counts to rate so sessions can be compared. If learning opportunities are fixed, as in a set of 10 trials, percentage correct or trials to criterion may fit better.
- Make it sensitive enough to show change. A measure should be able to move when the behavior moves. Long intervals, a ceiling, a vague scoring rule, or one yes/no score for a whole multi-step routine can hide real change.
Definitions
- Dimensional quantity: a measurable property of behavior, such as how often, how long, or when it occurs.
- Event recording: tallying each occurrence of a behavior during an observation, which yields count or, with the observation time, rate.
- Partial-interval recording: scoring an interval if the behavior occurs at any point during it.
- Whole-interval recording: scoring an interval only if the behavior occurs throughout the entire interval.
- Momentary time sampling: scoring whether the behavior is occurring at the moment each interval ends.
- Sensitivity: how well a measure detects real changes in the behavior it is meant to track.
Examples from a session
Same client, different goals
Mateo, 7, has three goals at a clinic.
- Requesting more often: Each request is brief and has a clear start and end, so his RBT counts requests. Sessions run 30 to 50 minutes, so she reports requests per minute.
- Starting work sooner: The goal is about when he begins, so she records latency from the instruction to his first step.
- Shorter crying episodes: He cries only twice a day, but each episode can last 20 minutes, so duration per episode is the measure that matches the goal.
A few more examples with practical limits in play:
- School: Ms. Okafor teaches 24 second graders and wants data on Lena's engagement during independent work. She cannot watch continuously, so she uses momentary time sampling. A vibrating timer prompts her to glance at Lena every 3 minutes.
- Group home: Staff want to know how many residents are engaged in the evening activity. A planned activity check, a form of momentary time sampling for groups, counts how many people are engaged at each check.
- Home: Sam, 16, flaps his hands in rapid bursts that are hard to count one by one. His BCBA uses duration of bursts, recorded from a short video, rather than trying to tally each flap.
Tip
When a discontinuous method is your only option, use the shortest intervals observers can manage and check agreement often. Short intervals bring the estimate closer to what continuous measurement would show.
Common mistakes / exam traps
- Comparing raw counts across sessions of different lengths. Ten responses in 20 minutes is not the same as ten in 60 minutes. Use rate.
- Counting behavior without clear boundaries. If observers cannot agree on where one response ends and the next begins, count data will be unreliable.
- Choosing whole-interval recording for a busy teacher. Both partial- and whole-interval recording require watching for the full interval. Momentary time sampling asks the observer to look only at the moment each interval ends.
Exam trap
Count vs. duration. When episodes are few but long, counting them is the easy choice, yet the count can stay flat while each episode gets shorter. The goal's wording decides: "less often" points to count or rate, "end sooner" points to duration, and "start sooner" points to latency.
Exam trap
Partial-interval vs. whole-interval. They err in opposite directions. Partial-interval scores any occurrence, so it tends to overestimate and is the conservative choice for a behavior you want to decrease. Whole-interval needs the behavior throughout, so it tends to underestimate and is conservative for a behavior you want to increase. Momentary time sampling can err either way.
Quick check
Noah, 15, often waits a long time after a parent's prompt before he begins his homework. The team's goal is for him to start sooner. Which measure best captures this goal?