5.5 The Three Levels of Scientific Understanding: Description, Prediction, and Control

5.5 The Three Levels of Scientific Understanding: Description, Prediction, and Control

Blueprint Anchor: QABA lists "scientific understanding: description, prediction, and control" as sub-topic 12 of Domain C, Core Principles of ABA. It pairs with the six attitudes of science and the seven dimensions covered in Section 5.1, and it is the concept that most directly governs what a supervisor may claim about their own data.

Science produces understanding at three increasingly powerful levels. They are cumulative: prediction presupposes description, and control presupposes prediction. A QASP-S who can place a claim at the right level will answer a large share of Domain C conceptual items and, more importantly, will stop making causal claims that the data do not support.

---

Level 1: Description

Description is the systematic observation and quantification of events, producing a collection of facts that can be organized and classified. Descriptive knowledge is real knowledge - taxonomy in biology and epidemiological counts in public health are descriptive - and in behavior analysis it is the foundation everything else rests on.

Descriptive activities in QASP-S practice include:

• Counting frequency, duration, latency, or interresponse time.
• ABC narrative recording of what preceded and followed each occurrence.
• Scatterplot recording across days and times of day.
• Rating scales and structured interviews about behavior.

The limit: description says nothing about relationships. A month of impeccable frequency data tells you the rate of aggression and nothing whatsoever about why it occurs or how to change it. A supervisor who has only descriptive data has not yet earned any statement containing the word because.

---

Level 2: Prediction

Prediction becomes possible when repeated observation reveals that two events reliably co-vary - when one changes, the other tends to change too. Covariation is expressed as correlation, and correlation supports forecasting.

Prediction is genuinely valuable to a QASP-S. If setting-event data show that aggression is elevated on days a client sleeps under five hours, the team can adjust demand density on those days without ever having established causation. Predictive knowledge is enough to act preventively.

The limit is the one every science student learns and every clinical team forgets under pressure: correlation does not establish causation. Two events can co-vary for three reasons other than one causing the other:

1. Reverse direction. Aggression and short sleep co-vary, but distress may be disrupting sleep rather than poor sleep producing aggression.
2. A third variable. Both may be produced by an untreated ear infection, a medication change, or a chaotic morning routine.
3. Coincidence. With enough variables tracked over enough days, some pairs will co-vary by chance alone.

Descriptive assessment - ABC recording, scatterplots, interviews - produces predictive knowledge only. This is precisely why a descriptive FBA yields a hypothesis about function rather than a demonstration of it, and why QABA distinguishes descriptive assessment from experimental functional analysis in Domain H.

---

Level 3: Control

Control is achieved when a functional relation has been demonstrated: a specific manipulation of one variable reliably produces a specific change in another, and the change can be attributed to that manipulation rather than to confounding variables. Cooper, Heron, and Heward describe the standard as a change in one event that reliably produces a change in another, with the demonstration replicated and unlikely to be the result of extraneous factors.

Three requirements distinguish control from prediction:

1. Systematic manipulation. The analyst deliberately changes the independent variable rather than waiting to observe natural variation.
2. Replication. The effect is shown more than once - reversed and reinstated, or staggered across baselines, or repeated across participants or settings.
3. Ruling out confounds. Other plausible explanations are held constant or made implausible by the design.

Control is what the analytic dimension of ABA requires. Baer, Wolf, and Risley defined an analysis as complete when the experimenter has demonstrated believable control over the occurrence and non-occurrence of the behavior. Nothing less licenses the claim that the intervention worked.

How Single-Case Designs Deliver Control

• Reversal (A-B-A-B): Behavior changes when the intervention is introduced, returns toward baseline when it is withdrawn, and changes again when it is reinstated. Three demonstrations of effect.
• Multiple baseline: The intervention is staggered across behaviors, settings, or participants; change occurs in each tier only when the intervention reaches that tier, making a common external cause implausible.
• Alternating treatments: Two or more conditions rapidly alternate, and differential responding across conditions demonstrates control.
• Changing criterion: Behavior tracks stepwise changes in the criterion, and the close correspondence between criterion shifts and behavior change demonstrates control.

An A-B design - baseline followed by intervention, with no replication - is not an analysis. It permits prediction at best. Almost every quasi-experimental clinical decision is made on A-B data, and that is often acceptable practice, but the supervisor should know exactly what level of claim their data support.

---

Why This Governs Supervisory Judgment

The three levels convert directly into a discipline about language and decisions.

Practical rule for the QASP-S: Match your verb to your design. Descriptive data support occurred and co-varied. Only a design with systematic manipulation and replication supports caused, produced, or worked. This is not academic fussiness - it determines whether you keep an ineffective plan running because a coincidence looked like a success, and it protects families and funders from claims your data cannot carry.