BCBA Measurement and Data Display: Choosing the Method and Reading the Graph

What Domain C of the BCBA exam covers

The BCBA Test Content Outline (6th ed.) gives Measurement, Data Display, and Interpretation 21 of the 175 scored questions (12%), across 12 tasks. The outline stays in force for 2027: the BACB's 2027 BCBA Requirements keep the exam on the 6th edition, in effect since January 1, 2025. Read in order, the tasks follow the life of a data set, from defining the behavior to judging what the graph says:

Four of these tasks start with "select," "design" or "evaluate," so expect scenarios that ask for the best choice under constraints, not only definitions. The task numbers and topics are the BACB's. The outline lists the tasks without defining them, so the explanations below follow the published sources listed with this guide.

Operational definitions and what to measure

An operational definition (C.1) describes the behavior in objective, observable terms, clearly and completely enough that someone who doesn't know the client could identify and measure it. Good definitions say what counts and name clear exclusions (a definition of spitting, for example, can exclude saliva leaving the mouth while eating). When the behavior will be timed, they also say where an instance starts and ends. Revisit a definition as the case goes on rather than treating it as final.

Task C.2 separates three kinds of measure. A direct measure records the behavior of interest itself, observed live or on video. An indirect measure relies on something else, such as an interview, a rating scale or a self-report, standing in for the behavior. A product measure records a lasting outcome the behavior leaves behind, such as completed worksheets. Product measures help when the behavior can't be watched, but only if the behavior reliably produces the product and nothing else produces it.

Tasks C.3, C.4 and C.7 cover the dimensions. Pick the one that answers the question being asked:

The outline lists trials to criterion, cost-benefit analysis and training duration as examples of efficiency measures (C.7). A trap to avoid: comparing raw counts from a 20-minute session and a 60-minute session. Without a time base, the comparison means little.

Continuous and discontinuous recording, and their bias

Continuous measurement (C.5) records every instance, as in counting, duration and latency recording. Discontinuous methods (C.6) sample the behavior, so they estimate, and the two interval methods err in a known direction:

These biases concern how much of the session the behavior takes up. Interval methods also blur count: an interval with five call-outs is scored the same as an interval with one, so partial-interval data can understate how often a frequent behavior occurs. Partial-interval and momentary time sampling errors grow with longer intervals, so shorter intervals generally give closer estimates, at the cost of more intervals to score.

The direction of the bias matters for choosing a method. Partial-interval recording is a poor fit for a behavior you want to increase, such as time on task, because its overestimate can make a small gain look large. Whole-interval recording is a poor fit for a behavior you want to reduce, because its underestimate can make a reduction look larger than it is. Momentary time sampling has no built-in direction and is generally the most accurate of the three for estimating duration, though it can miss low-frequency behavior.

Task C.9 adds the real-world test: choose a procedure that captures the behavior's critical dimension and still works in the setting. A teacher running a class of 25 usually can't record one student's behavior continuously, so momentary time sampling may be the practical choice; a brief, discrete behavior is usually counted rather than timed; a behavior that matters because it lasts too long calls for duration, not count. For practice at this kind of choice, 108 planned questions on Measurement, Data Display, and Interpretation are part of full prep at launch.

Validity, reliability and interobserver agreement

Task C.8 uses three ideas that are easy to blur:

• Validity. The procedure measures what it is meant to measure: here, the behavior of interest, in the dimension that matters for the question.
• Accuracy. The recorded values agree with an accepted reference value for what actually happened.
• Reliability. The procedure gives consistent values when the same events are measured again. A measure can be reliable and still consistently wrong.

Interobserver agreement (IOA) is the usual check: two observers record the same session independently and their records are compared. High agreement makes data more believable, but it doesn't prove accuracy, since two observers can agree and both be wrong. Three common calculations:

• Total count IOA. Smaller count divided by larger count, times 100. If one observer records 18 responses and the other 20, IOA is 90%. It can hide disagreement about which responses each observer counted.
• Interval-by-interval IOA. Intervals in agreement divided by agreements plus disagreements, times 100.
• Total duration IOA. Shorter duration divided by longer duration, times 100.

Watch for threats such as observer drift (an observer's use of the definition shifting over time), definitions too vague to apply the same way twice, and reactivity, in which being observed changes the behavior being recorded. Formulas and terms like these suit recall practice, and 70 flashcards on Measurement, Data Display, and Interpretation are planned for launch.

Graph types on the BCBA exam

Task C.10 names three examples of graphs:

The cumulative record is the one to read most carefully. A line that keeps rising doesn't mean behavior is getting worse; it only means responses are still occurring, and the slope tells you how fast.

Reading graphed data

Task C.11 is visual analysis. Describe each phase before judging any change:

• Level. Where the data sit on the vertical axis, summarized by the mean of the phase.
• Trend. The direction and steepness of the data path, the slope of a best-fitting straight line.
• Variability. How much the data spread around that trend line, for example their range.

Then compare adjacent phases: how much the level changed, how quickly the change appeared after the phase line, how much the data in the two phases overlap, and whether phases of the same condition look alike. A change counts as convincing evidence only when the baseline was stable enough to predict what would have happened without intervention. If baseline was already trending in the direction of improvement, a further improvement after the phase line is hard to attribute to the intervention. The BCBA experimental design guide takes this into design logic.

Procedural integrity data

Task C.12 asks you to measure the procedure, not just the client's behavior. Procedural integrity data show whether an intervention was implemented as designed, and the outline names accuracy and dosage as example dimensions: whether the steps were done correctly, and how much of the intervention was actually delivered. The data need to be representative, sampled across implementers, times and settings, and practical to collect. Without them, a flat graph can't tell you whether the intervention failed or was never really delivered, which is why Selecting and Implementing Interventions has its own task on decisions about procedural integrity.

A worked Domain C example written by Passy

This item was written by Passy for practice. It is not a BACB question.

Observers use 10-second partial-interval recording for a student's brief call-outs during a lesson. The summary shows calling out in 60% of intervals. Which interpretation is most accurate?

• A. The student called out during 60% of the lesson.
• B. Calling out occurred in 60% of intervals, which likely overstates how much of the lesson it took up.
• C. The student called out 60 times.
• D. Calling out probably took up more than 60% of the lesson.

Keyed answer: B. Partial-interval recording scores an entire interval if the behavior happens at any point in it, so a brief call-out marks 10 seconds that were mostly quiet. That is why the method tends to overestimate. A treats the percentage as time, C treats it as a count, and D reverses the method's bias, which belongs to whole-interval recording.

Studying Domain C for the BCBA exam

• Draw graphs by hand from small data sets, including one cumulative record, until reading slope is automatic.
• Describe before you judge. For every graph you meet, say level, trend and variability aloud for each phase first.
• Pair Domain C with Domain D. They share graphs, which is why the BCBA study plan guide schedules Measurement and Experimental Design in the same week.
• Calculate IOA by hand on a few invented records so the formulas stick.

Your next step for Domain C

Take one graph from your coursework or fieldwork and write a three-line description of each phase. Then try the 20 free BCBA practice questions, which include measurement items, and check whether you identified the dimension and the bias before reading each explanation. Once you're practicing at volume, accuracy tracked separately for each domain is planned for launch, to show whether Domain C is keeping pace with the others.

Passy is not affiliated with or endorsed by the BACB. The example item is original and isn't a BACB exam question.