Confounding variable

A confound creates a rival explanation

An independent variable is the condition deliberately manipulated. A dependent variable is the measured outcome. A confound is another factor that covaries with the planned manipulation closely enough to offer a credible explanation for the outcome.

Imagine that a clinician introduces a visual schedule at the first session conducted in a quiet room. Task initiation then improves. The change may reflect the schedule, the room, their combination, or another event. The data can support a claim about the whole bundled change if that package was implemented consistently. They cannot isolate the schedule's effect.

The BACB BCBA Test Content Outline, 6th edition covers independent and dependent variables, internal and external validity, single-case design, procedural integrity, and data interpretation. It is examination content rather than a research protocol.

An extraneous variable becomes a confound through alignment

Many features vary during services: sleep, traffic, weather, staff, noise, health, motivation, materials, and time of day. An extraneous variable is any feature outside the question that could affect measurement or behavior. It becomes a confound when its variation aligns with conditions in a way that threatens the intended comparison.

If every baseline observation occurs in the afternoon and every intervention observation occurs in the morning, time of day is confounded with condition. If morning and afternoon observations occur throughout both conditions, time may still add variability, yet it is no longer perfectly aligned. The analyst should still examine whether unequal distribution or interaction remains a plausible rival explanation.

A review applying mainstream validity threats to ABA single-case experiments describes history, maturation, testing, instrumentation, attrition, ambiguous temporal precedence, and selection-related concerns. These labels help locate the problem; the practical question is whether another change could account for the observed pattern.

Common service-delivery confounds are easy to miss

• Staff: One clinician implements baseline and a more experienced clinician implements treatment.
• Medication or health: A prescriber changes a medication, sleep improves, pain resolves, or illness begins with the treatment phase.
• Setting and schedule: Baseline occurs in a crowded room before lunch while treatment occurs in a quiet room after lunch.
• Materials or task sets: One teaching method receives easier items or more familiar examples.
• Measurement: The operational definition, observer, device, session length, or opportunity rule changes with condition.
• Concurrent support: Caregiver coaching, AAC access, reinforcement, or another service starts on the same date as the target procedure.
• Condition order: A later condition benefits from practice, carryover, fatigue, or exposure created by an earlier one.

Some simultaneous changes are clinically necessary. Record them accurately and narrow the conclusion. A physician-directed medication change stays under medical authority. The analyst can adapt design and interpretation without delaying or reversing needed care for cleaner data.

Repeated demonstrations reduce alternative explanations

A single A-B phase change leaves history and maturation especially plausible. The SCRIBE reporting statement places A-B designs below stronger experimental designs because they do not adequately control such threats. Reversal, multiple-baseline, changing-criterion, and alternating-treatments arrangements use different forms of replication to test whether outcome changes repeatedly correspond with manipulation.

The WWC Single-Case Design Technical Documentation explains that repeated, systematic manipulation and repeated outcome measurement support causal inference. The useful logic is prediction, verification, and replication. If the same outside event accompanies every manipulation, however, repetition may reproduce the confound rather than eliminate it.

Concurrent measurement matters in a multiple baseline. When several participants share a system change but only the participant receiving the staggered intervention changes at each step, history becomes less plausible. Nonconcurrent baselines leave calendar-time events harder to rule out.

Prevention begins before the first observation

Write a replicable definition of the independent variable and list features that must stay constant. Predefine the outcome, opportunity, observation window, prompts, exclusions, phase-change rule, and condition order. Then:

• log staff, setting, schedule, health, medication, access, equipment, and other service changes
• balance or randomize eligible condition order when the design supports it
• match task sets and report any remaining differences
• measure procedural integrity for every condition
• train observers and sample agreement across phases and staff
• keep missing data, protocol deviations, and participant withdrawal visible
• label a bundled intervention as a package until its components are separated experimentally

A modern overview of single-case design families notes that systematic environmental change, maturation, dropout, and uncontrolled events can obscure treatment-outcome relations. Design features reduce threats; transparent records reveal what the design could not control.

Stable baseline data help with prediction, yet stability cannot prove that the next phase is unconfounded. High treatment integrity shows that the planned condition occurred. It does not show that an outside change was absent. Observer agreement addresses consistency between observers, while an unchanged measurement system addresses instrumentation. Each control answers a different question.

A fictional ABA example

Leah is a fictional teenager who uses speech and AAC. She chooses a goal of starting a preferred after-school planning routine within two minutes of an accessible cue. Across eight baseline opportunities, Leah starts within two minutes in 2 of 8.

On the date a new visual cue begins, three other changes occur: a new staff member leads the routine, the family moves it to a quieter room, and a prescriber changes the timing of a medication. Leah starts within two minutes in 7 of 8 later opportunities. The arithmetic is clear; the cause is not. Staff, setting, medication timing, the visual cue, or their combination could account for the difference.

The team keeps medical decisions with the prescriber and records the full timeline. It reports the first comparison as a bundled change rather than evidence for the visual cue alone. Once health and service conditions are stable, a qualified clinician can select a safe design to answer any remaining clinical question. Leah's preference, AAC access, benefit, and unwanted effects remain part of the decision regardless of the experimental result.

Interpret evidence at the level tested

When two features always occur together, the defensible conclusion concerns their combination. A later component analysis may separate package elements. A comparative analysis may contrast complete interventions. A parametric analysis may vary one dimension. Clear limits strengthen the record because readers can see which causal claims the design supports.

The current BACB Ethics Code applies to BCBA and BCaBA certificants and people who completed an application. It addresses competence, collaboration, medical needs, consent and assent when applicable, assessment, risk, data, and evaluation. Experimental neatness never supplies authority to alter prescribed care, remove communication, or continue an unacceptable condition.