Responsible Behavioral Diagnosis: From Observed Behavior to a Testable Hypothesis
9 min read
Lesson objectives
- Define a target behavior and outcome measure without assigning psychological traits to people.
- Build competing hypotheses and specify evidence that distinguishes them.
- Integrate privacy, consent, fairness, and transparency into test design.
Diagnose the behavior, not the person
Behavioral diagnosis here is not psychological or clinical diagnosis. It is a structured formulation of a decision or action problem: What is the target behavior? Who performs it? In what context and time? What measurable outcome matters? What hypotheses could explain the gap? This language matters scientifically and ethically. Saying “users are lazy” labels people and hides alternatives; saying “among users who reach the document-upload step, 54% begin uploading during the session” defines behavior that can be measured and then investigated.
The OECD BASIC framework offers a practical sequence that begins with behavior and analysis, then develops strategies and tests interventions before broader change. The point is not that the framework guarantees a solution; it helps prevent starting with a favored intervention and then searching for a problem to fit it. At this foundational level, apply the logic: define the behavior, analyze context and candidate mechanisms, formulate an intervention that tests a hypothesis, and measure the outcome against an appropriate comparison.
| Diagnosis element | Practical question | Weak wording to avoid |
|---|---|---|
| Target behavior | What specific action are we trying to understand or measure? | We want better users. |
| Outcome measure | What count, rate, or time will be measured? | We want a nicer experience. |
| Context | When, where, and under what conditions does the behavior occur? | People always do this. |
| Hypotheses | What two or more mechanisms predict different patterns? | It is a known bias. |
| Discriminating evidence | What data or comparison separates the hypotheses? | We will see whether things improve. |
Start with the behavioral funnel and locate the gap
Choose the denominator to match the question. If we want the upload-start rate among those reaching the document step, it is 420 ÷ 780 ≈ 53.8%. If we want the upload-start rate among everyone who began the application, it is 420 ÷ 1,000 = 42%. Both are arithmetically correct, but they answer different questions. Naming the denominator explicitly prevents a common behavioral-analysis error: presenting a rate without defining the population to which it belongs.
Turn an explanation into a hypothesis with a prediction
A practical hypothesis template is: “Among [defined population] in [defined context], we expect [specific change] to affect [outcome measure] relative to [comparison condition] because [proposed mechanism]; we would reduce confidence in the hypothesis if [specified contrary pattern] appears.” This forces us to specify the population, intervention, outcome, comparison, and mechanism, and makes room to change our view when evidence does not support the prediction.
Ethics is part of design, not an afterthought
OECD guidance notes that ethical issues in applied behavioral science can arise from scoping through scaling, including privacy, consent, and fairness. In practice, ask before collecting data: Do we actually need this variable? Can participants understand what they should be told? Does the intervention distribute burdens or benefits unfairly across groups? Does its success depend on hiding material information or making exit difficult? An intervention that increases a short-term metric through deception is not responsible behavioral success.
Recap: responsible behavioral diagnosis defines behavior, context, and outcome before choosing an intervention, compares hypotheses that make different predictions, and treats ethics, privacy, and fairness as design conditions. The goal is to learn what explains behavior and what works, not to label people or push them at any cost.
Lesson summary
Behavioral diagnosis begins with behavior, measurement, and context, then turns explanations into competing hypotheses with predictions. Before intervention, review data necessity, transparency, fairness, and appropriate consent; the ability to influence behavior does not justify manipulation.