An ABC conditional probability table organizes observable antecedents, responses, and consequences into sequence counts with explicit denominators. Define the event codes, direction, time window, observation unit, and valid exposure before calculating. Keep conditional probabilities beside unconditional background probabilities and raw cells. The table describes recorded relations under sampled conditions and supports questions for qualified clinical review.

Start with one sequence question

Write the exact probability before building the table. Examples include P(work request before response), P(response after work request), and P(break after response). Each needs a named conditioning event, direction, sequence window, and eligible denominator. A table that mixes these directions under headings such as “antecedent percentage” or “consequence percentage” is hard to audit.

Identify the person, settings, observation dates and times, response definition, event definitions, sampling method, and intended clinical review. The BCBA Test Content Outline includes descriptive assessment and measurement concepts as examination content. It does not provide a universal table design or case decision rule.

Create the codebook and coverage record

Define every antecedent, response, and consequence in observable terms. State whether categories are mutually exclusive or may overlap, how simultaneous events are handled, and how onset and offset are recorded. Preserve “event absent,” “not observable,” “interval missed,” “outside window,” and “not applicable” as different states.

Record planned observation units, observed units, valid units, and exclusions with reasons. If 50 intervals were planned, 45 were observed, and 40 were valid, table coverage is 40 of 50 planned intervals, or 80%. The conditional denominator comes from eligible sequence exposures within those valid records, not automatically from all 40 intervals.

Map direction and time windows

For an antecedent-before-response calculation, decide how far backward from response onset to search and whether the nearest or any qualifying antecedent counts. For a consequence-after-response calculation, choose response onset or offset as the anchor and define the forward window. A repeated event may start a new window, overlap an active window, or be ignored until the first closes. Declare the rule before coding.

Keep truncated windows separate. A response in the final seconds of an observation may lack a complete follow-up window. Scoring the consequence absent would add an unsupported zero to the numerator while retaining the response in the denominator. Mark the sequence incomplete unless the protocol supplies an observable alternate rule.

Build raw cells before percentages

For each sequence calculation, store the conditioning-event count, number with the target event inside the window, number without it, number truncated or missing, and the resulting valid denominator. Then add background counts for the target event across the declared observation unit.

A compact build-ready table can include:

RelationTarget countEligible denominatorProbabilityBackground countWork request before response610 responses60%12 of 40 intervalsBreak after response410 responses40%8 of 40 intervals

The labels must make clear that “before” and “after” use different sequence rules even when both happen to use ten response intervals.

Verify Mira’s calculations

Mira has 40 valid ten-second intervals. The response appears in 10, so its unconditional interval probability is 10 / 40 = 25%. A work request occurs in the defined antecedent position for 6 of the 10 response intervals: 6 / 10 = 60%. A break occurs in the forward consequence window after 4 of the 10 responses: 4 / 10 = 40%.

Across all valid intervals, work requests appear in 12 of 40, or 30%, and breaks in 8 of 40, or 20%. The four fractions answer different questions. The 60% and 40% values condition on response intervals. The 30% and 20% values describe background interval presence. Raw cells remain beside every percentage.

Do not infer unreported complements when event categories overlap. A response can have both a break and another consequence in its window. Marginal consequence probabilities may therefore sum above 100%. If exclusive patterns matter, add explicit cells such as break only, attention only, both, and neither.

Handle zero cells and missingness

A zero numerator with a positive denominator is calculable. If work request preceded 0 of 10 responses, the result is 0%. A zero conditioning denominator is different: with no observed response events, P(work request before response) is undefined because division by zero is impossible. Report “undefined, 0 eligible responses,” not 0%.

Missing observation can bias the table when it clusters in a setting, routine, staff arrangement, or time period. Report planned and valid coverage by those dimensions. Never treat an unobserved interval as an event-free interval. Recalculate after source corrections and retain the prior version.

Graph and report the pattern

Use a table or paired dot display that shows raw fractions and percentages, with conditional and background values clearly distinguished. Include coverage and small-denominator warnings. A sequence timeline for selected sessions can help reviewers see how windows were applied, but it should not replace the complete cell counts.

A report might say: “Across 40 valid ten-second intervals, 10 contained the defined response. Work request preceded 6 of 10 response intervals (60%) and occurred in 12 of 40 intervals overall (30%). Break followed 4 of 10 responses (40%) and occurred in 8 of 40 intervals overall (20%). These sampled associations depend on the declared interval and window rules.”

Keep interpretation descriptive

The table can identify patterns that merit further clinical inquiry. It cannot establish that an antecedent evoked a response, that a consequence reinforced it, or that the sampled setting represents other times. Common background events can appear frequently in sequences by chance. Correlated health, pain, task, access, communication, or social conditions may remain unmeasured.

Primary work on antecedent and consequent predictors and contingency space analysis shows why background probability and window selection matter. Comparative descriptive-method research found limited correspondence with experimental findings, and functional-assessment ambiguity research distinguishes observed association from experimental demonstration.

Review access and client meaning

Review the observation and proposed next step with Mira through accessible communication. Keep AAC continuously available under ASHA guidance. Ask about discomfort, relevant events, assent, dissent, priorities, and whether the coded outcome captures something meaningful.

Observation and calculation never authorize restriction of food, water, bathroom access, mobility, prescribed care, health support, rest, relationships, or emergency help. Medical, safety, privacy, payer, and legal questions go to their qualified owners. The BACB ethics hub and CASP public summary provide professional context for covered practice.

Clinician review checklist and limits

Before using the table, confirm:

  • observable definitions, category overlap, timestamps, direction, anchors, and windows are explicit;
  • planned, observed, valid, missing, excluded, and truncated units reconcile;
  • every conditional result has the correct eligible conditioning denominator;
  • background values use a compatible observation unit and preserve raw counts;
  • zero numerators and zero denominators are reported differently;
  • graphs and notes retain small cells, coverage, context, and client feedback; and
  • the review records limitations, next assessment step, qualified owner, and date.

Mira’s 40 intervals and 10 responses form a small descriptive sample. One reclassified sequence would move 6 of 10 to 5 of 10 or 7 of 10. Window length, observation timing, event overlap, and missingness can alter the pattern. Reopen the ABC conditional probability table when definitions, source data, sampling, settings, access, health context, observer, or software changes, and preserve earlier versions for audit.

Related resources

Sources