Framing Effect

Activate when: user says 'how you say it matters', 'they're spinning the numbers', 'same data but sounds different', 'is this just framing?', wants to write a persuasive message, suspects a statistic is one-sided, or is evaluating options where the language feels loaded. Do NOT activate when: the different descriptions convey genuinely non-equivalent information (one frame omits real facts); the decision is trivial and multi-frame analysis exceeds the stakes.

Overview

The framing effect: logically equivalent descriptions of the same decision produce different choices depending on whether outcomes are cast as gains or losses. Frame determines whether the brain enters gain-mode or loss-mode; the choice follows from the mode.

Three types: risky-choice (gain vs. loss on probabilities), attribute ("95% fat free" vs. "5% fat"), goal ("do X to gain Y" vs. "skip X, lose Y"). Real frames often compound all three.

Corrective: force both framings. If the decision is stable across frames, the frame is not driving it. If it flips, inspect why.

Composes with loss-aversion-prospect-theory, anchoring, pricing-strategy, signaling-games, critical-thinking.

When to Use

• Decision presented with strong gain-only or loss-only language
• Statistic shown in one form only (survival rate without mortality, or vice versa)
• Evaluating persuasive communication: ad, political message, medical recommendation, pitch
• Team moving toward a decision whose framing was not chosen neutrally
• You suspect manipulation by frame choice

Not when: the alternative frame is genuinely misleading (not logically equivalent); analysis cost exceeds decision stakes.

Coaching Novices (Adaptive Front Door)

• Engine mode: user has a concrete framed decision → run The Process directly.
• Coach mode: user is unfamiliar → guide step by step.

In Coach mode, respond one step at a time. Each [WAIT] is a hard stop — output only that step's question, then stop.

1. One-line: before accepting a framed decision, restate it in the opposite frame — do you choose differently?
2. Check fit: if the alternative frame is not equivalent, single-frame may be appropriate.
3. Elicit their real case: what's being decided? What gain/loss/do/don't language is in use?

[WAIT — do not advance until user responds]

4. Run the audit one step at a time with their input.

[WAIT — do not advance until user responds]

5. Close by naming what the frame was doing and what the frame-independent choice is.

[WAIT — do not advance until user responds]

The Process

Step 1 — Identify the frame: gain/loss/attribute/goal; verbatim language; who chose it. Step 2 — Construct the equivalent alternative frame: same math, opposite wording. Step 3 — Re-evaluate: does your choice change under the alternative frame? If yes, the frame is doing the work. Step 4 — Compute frame-independent EV: strip the language; which option is rationally best? Step 5 — Diagnose framer intent: who benefits from the framed choice? Was the frame chosen to steer? Step 6 — Choose response: take framed option (if aligned with EV), take frame-independent option, or expose the frame.

Output: Framing Audit

text

Frame in use: [type] | Verbatim: [quote] | Framer: [self/counterparty/advertiser]
Alternative frame: [restatement] | Mathematically equivalent: Y/N
Cross-frame decision: Original → [choice] | Alternative → [choice] | Stable: Y/N
EV (frame-independent): [best option + rationale]
Framer intent: [who benefits, is frame steering]
Decision taken: [choice + justification]

→ Method in Action: Tversky and Kahneman's 1981 "Asian Disease" Study

Pack: Framing Patterns

Domain	Manipulation	Use or counter
Medical	"X% survive" vs "Y% die"	Demand both framings of the same data
Pricing	"Save $X" vs "Cost $Y"	Recognize loss-avoidance frame when buying
Politics	"Death tax" vs "Estate tax"	Identify framer's goal; switch frame to expose
Public health	Gain frame → prevention; loss frame → detection	Match frame to desired behavior type

Applying It Well

• Always construct the logically equivalent opposite frame before acting on any framed decision.
• The structural defense is multiple frames in the same environment — not vigilance alone.
• Match frame to behavior type: loss for detection behaviors, gain for prevention behaviors.

→ Primary sources: references/sources.md

Common Rationalizations

[D] = designed upfront | [O] = observed in real use. [O] entries are more valuable.

Fake move	Reality
[D] "The numbers speak for themselves"	They don't; the frame is doing the work. Same numbers, different frame → different choices.
[D] "I'm not affected by framing"	Self-rated immunity has near-zero correlation with measured immunity in the literature.
[D] "There's only one way to describe this"	Almost never true. Any decision with gains and losses has at least one equivalent inversion.
[D] "Reframing is just spin"	The original frame is also spin. Calling reframing "spin" is itself a frame defense.
[D] "The other frame would be misleading"	Test: is it mathematically equivalent? If yes, both frames are equally accurate or misleading.
[D] "We just need to communicate the facts"	Words frame. There is no frame-neutral way to state most facts; choose deliberately or by default.
→ Add [O] entries here after each real use — paste the actual failure pattern	What went wrong and why

Red Flags

• Decision presented with gain-only or loss-only language; no alternative frame offered
• Statistic shown in one form only; alternative form dismissed as "spin"
• Framer benefits from the choice they're steering you toward
• "We have to use this language for the audience" used to block reframing

Verification

• Frame in use identified explicitly
• At least one logically equivalent alternative frame constructed
• Decision re-evaluated under the alternative frame
• If decision flips, dependency examined and deliberate choice made
• Framer intent considered
• If communicating, frame matched to desired behavior type

---

Part of deciqAI Knowledge Skills — open-source thinking skills that make rigor executable for AI agents. Built by deciqAI · https://deciqai.com · Contributions welcome — see the template at the repo root.