Direct observation

Direct observation measures events, not labels

An observer records what a defined response and relevant environmental event look like in a stated period. Depending on the question, the measure may be count, rate, duration, latency, interresponse time, steps completed, or responses per eligible opportunity.

The CASP ABA Practice Guidelines Version 3.0 public summary places assessment and treatment planning within standards of care for ABA behavioral health treatment for people diagnosed with autism. This editorial overview uses only that public scope; the licensed details remain outside it.

Observation differs from an interview, record review, or rating scale because the observer scores sampled events. A descriptive assessment records behavior and environmental events under naturally occurring conditions; observed patterns are correlational and do not demonstrate function. A functional analysis systematically manipulates environmental variables within an experimental design to evaluate their effects on behavior. It requires appropriate competence, client-specific rationale, consent and assent when applicable, risk safeguards, and stop criteria. Direct observation may provide measurement in either process, but the word direct does not make descriptive data experimental.

Plan the sample around a decision

The current BACB BCBA Test Content Outline, 6th edition covers definitions, direct and indirect measures, occurrence and time, representative measurement and integrity data, graphing, descriptive assessment, and functional analysis. It is exam content, not a clinical protocol. Before collecting, specify:

1. the client-informed question and decision
2. behavior and event definitions, measure, period, eligible opportunity, and missing-data rule
3. contexts and support conditions sampled, with the rationale
4. consent, assent when applicable, privacy, safety, and stop procedures
5. observer training, calibration, drift checks, and double-scoring plan
6. graphing, sharing, interpretation, and integration with other evidence

A clinic visit cannot represent a week at home or school. Record expected and missed opportunities and why each period was chosen.

Watch the environment and partners too

Observe partners and the environment too: whether instructions were accessible, AAC was within reach, materials were ready, break messages received a response, and scheduled reinforcement or safety supports were available. Scoring only the client's response can hide access or implementation problems.

Use observable definitions rather than intent guesses or value-laden labels. Instead of “noncompliant,” define what occurred, such as “did not begin the first listed step within 30 seconds after an accessible direction and materials.” The team must still decide whether the task, timing, and response requirement are appropriate. If pain, illness, fatigue, distress, a communication barrier, or a hazard appears relevant, pause when needed, meet immediate health or safety duties, and seek appropriate referral. Record only observable, decision-relevant information within the observer's role; observation does not replace medical or safety assessment.

Observer agreement is one quality check

Independent observers can score the same prespecified periods. Report the formula, unit, double-scored share, contexts, and numerator and denominator. Agreement shows similar coding in that sample; it does not establish accuracy, validity, representativeness, absence of shared bias, or causation.

In a 1980 videotape analogue with 36 observers, Farkas and Tharp found systematic sampling more accurate on average than unsystematic sampling, with observer-gender and behavior-valence interactions. In a laboratory model with 61 individuals, Lerman and colleagues distinguished misses from false alarms in simulated videos and showed that feedback and scoring consequences could bias responding. Neither study supplies a clinical accuracy rate.

Training needs definitions, examples and nonexamples, practice, feedback, and a task-specific criterion. Dempsey and colleagues sequentially placed 59 undergraduates into in-vivo or video training. Video-test scores were nearly identical, but video trainees reached criterion sooner. No pretest, random assignment, or single isolated training variable supports naming one method best for clinical use.

People may respond to being observed

An observer, camera, unfamiliar clinician, or changed routine can alter behavior. Ask what feels acceptable, make the observer's role predictable, allow familiarization when appropriate, sample multiple occasions, and note visible reactivity.

Codding and colleagues compared three school staff implementing one classwide plan with the observer in the room or behind a one-way window. They found no differentiation, but conditions were not independent and participants were not blind to the purpose. The result cannot rule out reactivity elsewhere.

Live observation and recording are separate privacy choices. Before recording, identify applicable authority and required consent or authorization. Explain its purpose, content, access, storage and transmission safeguards, retention or deletion schedule and exceptions, and bystander handling. Follow applicable health-record, school, workplace, payer, and state rules. Provide an accessible way to pause or stop when safe. Record a stopped period as stopped or missing under the preset rule, not as a behavior zero or compliance failure. Immediate safety duties remain.

Preserve communication and ordinary support

ASHA's AAC practice portal says AAC users should always have access to their communication tools or devices. Observation should include speech, AAC, sign, gesture, writing, and other reliable forms in the response definition. Do not remove AAC, mobility tools, prescribed care, sensory supports, or an established safety plan to obtain an “unsupported” sample.

For people covered by it, the current BACB Ethics Code addresses competence, consent and assent when applicable, assessment, risk, data, confidentiality, client involvement, and evaluation. BACB does not separately govern organizations; other applicable rules remain in effect.

A fictional observation example

Eli is a fictional fifteen-year-old who wants a reliable way to move to a quieter space during community groups. Across four observations, 14 predefined transitions occur while Eli is present. AAC is within reach during 12 of 14, or 86% when rounded, and unavailable during 2 of 14, or 14%. Report access as adult or environmental implementation; the two failures are not Eli zero messages.

With planned access, an independent quiet-space message is speech, AAC, sign, or a pre-agreed gesture before a communication prompt. Eli communicates during 4 of 12 opportunities, or 33% when rounded. Partners acknowledge it and make the agreed route or support available within 30 seconds after 2 of 4 messages, or 50%. Other refusal or distress signals are honored. Safe dissent ends observation and is not a behavior failure.

Before collection, two complete sessions are selected for independent double scoring. Both observers identify every scheduled transition and code access, messages, and partner responses without comparing notes. If those sessions contain seven scheduled slots, report each code's numerator and denominator, retain disagreements under the preset rule, and state coverage as 2 of 4 sessions and 7 of 14 slots. Do not select only primary-observer events or exclude disagreements.

The sample raises questions about message access and partner response; it does not prove a cause or treatment. Eli's preference, broader observation, caregiver input, and clinical judgment guide the next step.

Questions families can ask

• Which decision will the observation inform?
• How are the response and eligible opportunity defined?
• Which settings, people, times, and supports will be sampled?
• How will observers train and check agreement or drift?
• How will the team record adult actions and access failures?
• Can the person pause, decline, or communicate discomfort?
• Will anything be recorded, and how will privacy and deletion work?
• Which conclusions remain outside what this sample can support?