Key Points
- Every opportunity is scored as it happens, so the session's number is assembled from what occurred rather than recalled once the session is over.
- Recording the prompt level, not just right or wrong, is what distinguishes a skill being acquired from one being propped up by help.
- The order of the results carries information a percentage hides - where the errors clustered, and whether the strong trials came first or last.
Trial-by-Trial Data Explained
In a teaching program the unit of measurement is the trial: an instruction, the learner's response, and what followed it. Trial-by-trial data records each one - correct, incorrect, or prompted, and at what prompt level - so a session's accuracy is derived from the individual results rather than estimated as a whole.
The alternative is summary data: one figure per target per session, written down at the end. It is cheaper to collect and is the right choice for some programs, particularly probe-based ones, but it discards the sequence. Eight correct out of ten describes several very different sessions - two errors at the start while the learner warmed up, two at the end as attention faded, or two in the middle when the materials changed - and only one of those is a session that went well.
The field that earns this format its keep is the prompt level. A target can look mastered on accuracy alone while every correct response was partially prompted, and separating independent from prompted responses on the graph is what shows a prompt actually being faded - or a learner settling into dependence on it. Criteria written around independent responding cannot be checked at all without that detail.
The cost is the therapist's attention. Scoring every trial while running the session is real work, and a paper datasheet is usually where it gives way: trials get filled in from memory in batches, or after the session, at which point the data are a reconstruction. This is the clearest case for capture that takes one tap per trial and timestamps it, and for datasheets that follow the program instead of being redrawn by hand each week.
For a supervising clinician the payoff is diagnostic. Trial-level records show which discriminations a learner is failing rather than only that accuracy is low, whether errors track a particular instructor or time of day, and whether a target has sat at the same prompt level for weeks. A session percentage answers none of those, and they are the questions that decide what changes next.
Not every goal wants this resolution. Behaviors measured by frequency, duration, or interval sampling are not made of trials, and a program with dozens of targets can generate more detail than anyone will read. The judgment is the one measurement always demands: collect the detail a decision actually turns on, and the record stays usable.