Meta Decision Analysis

Decision Analysis

Framework

IRON LAW: Make Criteria and Weights Explicit BEFORE Evaluating Options

Choosing criteria after seeing the options lets bias sneak in — you
unconsciously weight criteria that favor your preferred option.
Define criteria, assign weights, THEN score options.

Decision Matrix (Weighted Scoring)

• List alternatives (3-6 options including "do nothing")
• Define criteria (4-8 factors that matter)
• Weight criteria (must sum to 100%)
• Score each option per criterion (1-5 or 1-10)
• Calculate weighted total = Σ(score × weight)
• Sensitivity check: Does the winner change if you adjust the top-weighted criterion?

Decision Tree (Sequential Decisions Under Uncertainty)

For decisions with uncertainty and sequential steps:

• Map decision nodes (squares) and chance nodes (circles)
• Assign probabilities to chance outcomes (must sum to 1.0)
• Assign payoffs to terminal nodes
• Calculate Expected Value = Σ(probability × payoff)
• Choose the branch with highest EV (or best risk-adjusted outcome)

Multi-Criteria Decision Analysis (MCDA)

For complex decisions with competing stakeholder priorities:

• Each stakeholder defines their criteria and weights independently
• Aggregate into a combined weighted matrix
• Identify where stakeholders agree (easy decisions) and disagree (requires negotiation)

Output Format

# Decision Analysis: {Decision}

## Alternatives
1. {Option A}
2. {Option B}
3. {Option C}

## Decision Matrix
| Criterion | Weight | Option A | Option B | Option C |
|-----------|--------|----------|----------|----------|
| {criterion 1} | {X%} | {1-5} | {1-5} | {1-5} |
| **Weighted Total** | 100% | **{total}** | **{total}** | **{total}** |

## Sensitivity Analysis
- If {criterion} weight changes from X% to Y%, winner changes from {A} to {B}

## Recommendation
{Winner with rationale and key trade-offs acknowledged}

Gotchas

• "Do nothing" is always an option: Include it as a baseline. Sometimes the best decision is to wait.
• Scores are subjective: A score of "4" from one person ≠ "4" from another. Calibrate by defining what each score means before scoring.
• Expected value ignores risk preference: EV of $50 (certain) vs EV of $50 (50% chance of $0, 50% chance of $100) are equal by EV but feel very different. For high-stakes decisions, use risk-adjusted metrics.
• Analysis paralysis: Decision analysis should accelerate decisions, not delay them. Set a time limit for the analysis.

References

• For decision tree software tools, see references/decision-tools.md