What a measurement system is
A measurement system is the rule for turning behavior into a number. It specifies what counts as an occurrence, what the observation window is, and how occurrences are aggregated into a session-level value. Two clinicians watching the same child will produce the same data only if they are using the same system with the same definitions.
The systems are not interchangeable. A behavior measured as a frequency answers "how often"; the same behavior measured as a duration answers "for how long"; measured by partial interval it answers "roughly how much of the time" - and those three numbers can move in opposite directions during the same treatment. Choosing the system is therefore a clinical decision that precedes data collection, not a setting to be revisited later.
The eight systems
The table below covers all eight, in the order they are conventionally taught. "Metric" is the per-session quantity the system produces - the number that gets graphed and compared against a mastery criterion.
| System | What it measures | When to use it | Per-session metric |
|---|---|---|---|
| Discrete trial trainingDTT | Massed, structured teaching trials, each with a clear instruction, a response, and a consequence, scored one trial at a time. | Acquiring a new discrete skill that benefits from many closely-spaced, controlled opportunities. | Percent independent (unprompted correct) per session, or percent correct. |
| Natural environment teachingNET | Teaching embedded in play and daily routines, capturing opportunities as they arise rather than in massed blocks. | Building generalization and spontaneity, or working with a learner for whom table-based drilling suppresses motivation. | Percent independent across captured opportunities per session. |
| Task analysis | A chained skill broken into ordered steps, with each step scored for the prompt level the learner needed. | Teaching a multi-step routine - handwashing, dressing, a vocational sequence - where the useful signal is which step breaks down. | Percent of steps completed independently per session. |
| Frequency / rate | A count of how many times a behavior occurs within a time window. | The behavior is discrete and countable, and how often it happens is the clinical question. | Occurrences per session, evaluated against an at-or-below mastery rule. |
| Duration | How long a behavior lasts, timed per occurrence and totaled. | The length of an episode matters more than the count - a tantrum, an engagement bout, time on task. | Total seconds per session, evaluated against an at-or-below rule. |
| Interval recording | Time sampling across an observation split into intervals - partial interval (any occurrence counts), whole interval (must occur throughout), or momentary time sampling (scored at the interval's end). | The behavior is continuous, high-rate, or hard to count discretely, so an estimate of how much of the time it occurs is more practical than a true count. | Percent of scored intervals containing the behavior. |
| Latency | The elapsed time between the instruction and the onset of the response. | The learner responds, but not promptly - compliance delay is the target, not accuracy. | Mean seconds to respond, evaluated against an at-or-below rule. |
| Antecedent-behavior-consequenceABC | Narrative incident records capturing what preceded a behavior, the behavior itself, and what followed it. | Forming or testing a hypothesis about function, typically feeding a functional behavior assessment. | Incident count per session, with each record retained for analysis. |
How to choose between them
Start with the dimension of the behavior that matters clinically. If the concern is how often something happens and each instance has a clear start and end, use frequency. If instances vary in length and the length is the problem, use duration. If the behavior is so continuous or high-rate that counting it accurately is impractical, interval recording trades precision for feasibility on purpose.
For skill acquisition, the question is usually the teaching format rather than the dimension. Discrete trial training gives you many controlled opportunities and the cleanest percent-independent data. Natural environment teaching gives you fewer, less uniform opportunities but far better evidence of generalization. Task analysis is the right choice whenever the useful signal is which step in a chain breaks down, because a single percentage for the whole routine hides exactly that.
Latency is the system reached for least often and misused most: it applies when the learner does respond but too slowly, and it is the wrong choice when the real issue is accuracy. ABC recording is not a rate measure at all - it is a structured narrative feeding a hypothesis about function, and it usually runs alongside a quantitative system rather than instead of one.
Why the choice constrains everything downstream
A mastery criterion has to be written in the same units as the metric. "80% independent across three consecutive sessions" is meaningful for a DTT target and meaningless for a duration target, where the criterion has to be an at-or-below threshold in seconds. Software that stores every target as a generic number will happily evaluate a behavior-reduction target as though a higher score were better.
Graphs inherit the same constraint. Equal-interval line graphs suit percentage and rate data; cumulative records make acquisition trends legible; neither is interpretable if the underlying metric changed units partway through a program. This is the practical reason measurement systems belong in the data model rather than in a free-text field.
In Cadence, all 8 systems are native to the schema, and a target's metric is derived from its measurement system and mastery rule together, so the mastery threshold and the graphed value are always in the same units.