Antecedent given response conditional probability equals valid response events preceded by the named antecedent within the look-back window divided by all valid response events. Define response anchor, antecedent code, window, and treatment of repeated or ongoing antecedents. Compare the result with antecedent background occurrence and preserve raw counts, observation coverage, and context before clinical interpretation.

Start with the conditional direction

Antecedent given response conditional probability asks which valid response events were preceded by the defined antecedent. The response events form the denominator. Write the direction as P(antecedent | response) before counting so it is not confused with P(response | antecedent), which uses all antecedent exposures as its denominator.

Define the response anchor, usually onset, and a look-back window. Also define whether the antecedent must begin within that window, may already be ongoing, or may end within it. These choices can place the same record in different cells. Use timestamps precise enough for the selected window and confirm that observer clocks align before matching events.

Prepare a response-anchored worklist

Create one row for every candidate response. Retain response onset, setting, opportunity, antecedent code and onset, look-back eligibility, competing or simultaneous antecedents, observer, source, and resolution status. Then classify each eligible response as preceded by the named antecedent or not preceded by it. The two cells must reconcile to all valid responses.

Responses too close to observation start may lack a complete history. Label these as left-truncated unless earlier records cover the full window. Lost visibility, imprecise timestamps, and unresolved event codes are missing states, not evidence that the antecedent was absent. This distinction protects the denominator from artificial inflation.

Calculate Amara's probability

Amara has 14 valid response events with complete look-back windows. A demand begins within the defined window before 8 responses. Six responses have no demand meeting that timing rule.

P(demand | response) = 8 / 14 = 0.5714, reported as 57.1%.

The denominator is 14 because the calculation conditions on response. It is not the 20 demand intervals or all 70 background intervals. The worksheet should retain 8 preceded and 6 not preceded so the 14-response cohort reconciles.

Demand is also recorded in 20 of 70 valid background intervals:

background demand occurrence = 20 / 70 = 0.2857, or 28.6%.

This background value helps readers understand ordinary exposure. Because it uses intervals while the conditional uses response anchors, it should not automatically be subtracted as a formal contingency difference. A signed comparison requires compatible units and a prespecified analysis.

Review opportunity and coverage

Report planned observation, valid response anchors, truncated anchors, missed response periods, and background-interval coverage. Fourteen valid responses may be distributed across many routines or concentrated in one short condition. The probability cannot represent times, people, or settings that were not observed. Examine whether missing look-back history clusters near session starts, transitions, high-risk routines, or access disruptions.

Background exposure must use the same demand definition and a compatible sampling frame. If staff record demands reliably only when a response follows, 20/70 is biased downward or upward depending on the missing pattern. Review event logs and a sample of nonresponse periods before treating the background rate as informative.

Resolve ongoing, repeated, and compound antecedents

An antecedent may begin before the look-back boundary and continue into it. Predeclare whether ongoing exposure qualifies. Repeated demands within one window may be scored as any demand, first demand, most recent demand, or count, depending on the question. Preserve raw timestamps so another rule can be evaluated without changing the original record.

When several antecedents occur, retain their component codes. A demand, denied item, noise, pain indicator, and communication-access problem should not be collapsed into one convenient label after reviewing the outcome. Predefined mutually exclusive compound categories can answer a focused question, but small compound cells quickly become unstable. Health or safety concerns require the appropriate qualified response and should not be reduced to an ABC code.

Handle zero and uncertain data

If there are valid responses and none has the antecedent, the conditional result is zero. If there are no valid responses, the probability is undefined. A response with an incomplete look-back window does not enter either numerator or valid denominator. Report it as truncated. When antecedent presence is disputed or the timestamp cannot support the window, preserve the uncertain state and adjudication trail instead of defaulting to absence.

One or two responses can produce an apparently decisive percentage. Always show the fraction, sessions, contexts, and distribution. Additional representative observation may be the appropriate next step when the data are sparse, systematically missing, or dominated by a single routine.

Interpret association within clinical boundaries

Amara's 57.1% result means that 8 of 14 sampled valid responses were preceded by a demand under the selected definition and window. It does not mean demands caused 57.1% of responses, establish a behavioral function, or show that changing demands is clinically appropriate. Demands may be common, correlated with another event, coded differently by observers, or more visible near responses.

Qualified interpretation uses Amara's perspective, health and safety information, direct and indirect assessment, implementation evidence, and other relevant data. Descriptive findings can refine questions or sampling. They do not authorize creating distress, withholding access, or delaying necessary care to obtain more sequences.

Graph and report both denominator systems

Display P(demand | response) with the fraction 8/14. Show background demand occurrence separately as 20/70, clearly labeled as an interval measure. A two-panel display prevents readers from assuming both bars use the same denominator. Add counts of truncated response anchors and invalid background intervals in a coverage note.

Report wording can read: "A demand began within the defined look-back window before 8 of 14 valid responses (57.1%). Demand was recorded in 20 of 70 valid background intervals (28.6%). These response-anchored and interval-based summaries use different denominator universes and describe the observed sample; they do not establish function." Include dates, contexts, timing rule, rounding, and missingness.

Review records accessibly with Amara

Use Amara's preferred communication method to review the operational definitions, examples, graph, uncertainties, and proposed next step. Keep AAC, interpreters, mobility, food, water, bathroom use, prescribed care, rest, relationships, and emergency help available throughout observation. ASHA guidance supports continued AAC access. Record Amara's corrections, assent, dissent, discomfort, and priorities with their source.

Protect identifiable ABC details and limit access to the authorized team. Label direct observation, partner report, system timestamps, and clinician inference separately. A communication or access barrier belongs in context and coverage, not in the antecedent-absent cell.

Keep source claims in scope

The BACB ethics information, CASP public guideline summary, and BCBA Test Content Outline provide the professional assessment and measurement frame. Studies of predictor events, contingency-space methods, and lagged response-stimulus sequences support explicit direction, timing, and comparison cells. The precursor study demonstrates how reciprocal conditionals answer different questions. Descriptive-method comparisons caution against causal overreach.

Clinician review checklist

For antecedent given response conditional probability, confirm the clinical question, response definition and anchor, antecedent code, ongoing-event rule, look-back length, timestamp precision, all response rows, complete-window denominator, preceded and not-preceded cells, truncated and uncertain states, 8/14 arithmetic, background 20/70 calculation, denominator labels, graph, report wording, accessible review, safety actions, qualified owner, next step, and review date.

Practical limits

The result depends on the sampled responses, look-back rule, event definitions, clock quality, and observation coverage. It cannot identify unseen antecedents, establish a functional relation, measure treatment integrity, show clinical importance, predict an individual response, or determine an intervention. Background intervals may not be directly comparable with response anchors. Preserve raw events and open a separately labeled series whenever definitions, windows, settings, or sampling methods change.

Related resources

Sources