Decision Analysis Project
Theory without practice is inert. This lesson is a structured project: you will apply every major framework from this module — decision structure, uncertainty analysis, expected value, judgment trap audit, AI-assisted research, and retrospective planning — to one real decision that actually matters to you. The goal is not a clean answer, but a rigorous documented reasoning process.
Choosing Your Decision
A good decision analysis project requires a decision that is: Genuine: a real choice you actually face or recently faced — not invented. The engagement that comes from real stakes makes the analysis far more valuable. Uncertain: the outcome should not be certain in advance. Decisions with obvious answers do not require analysis; decisions where reasonable people could disagree are the productive territory. Multi-dimensional: the decision should involve tradeoffs across at least two values or dimensions — not purely financial, not purely immediate. Analyzable: you should be able to identify meaningful options, plausible outcomes, and rough probability estimates. 'What should humanity do about climate change' is too vast. 'Should I apply to this specific program this year or wait one more year?' is analyzable. Examples that work well: choosing between two academic paths, deciding whether to pursue or abandon a personal project, evaluating a commitment to an organization or activity, deciding how to handle a recurring conflict, choosing a major area of study.
Before beginning analysis, identify everyone affected by this decision — not just yourself. Sometimes the most important uncertainty is not about the world, but about how other stakeholders will respond. Include their likely responses as explicit outcomes in your analysis.
Full Decision Analysis
- Complete all six phases of this analysis in writing. Each phase should be substantive — this is a project, not a worksheet.
- PHASE 1: DECISION STRUCTURE
- Write a clear statement of the decision problem. Then list every option you can identify — including 'do nothing' or 'wait.' For each option, list the plausible outcomes (at least two per option). Assess which outcomes are within your influence and which depend on external factors. Identify who else is affected by each outcome.
- PHASE 2: UNCERTAINTY ANALYSIS
- For each outcome you listed, classify the uncertainty: is this a risk (you can estimate probabilities from reference class data or experience) or Knightian uncertainty (you genuinely cannot form a reliable probability estimate)? For the risk-category items, assign probability estimates and briefly justify each estimate. What is your primary source of uncertainty — your own capabilities, other people's decisions, or external events?
- PHASE 3: EXPECTED VALUE CALCULATION
- For at least two of your options, assign a value score (use a 0-100 scale if not monetary) to each outcome, then calculate expected value. Show your work. Now run a sensitivity analysis: identify the two probability estimates you are least confident in, vary each by plus or minus 15 percentage points, and check whether the ranking of your options changes. What does this tell you about where to focus your remaining uncertainty?
- PHASE 4: JUDGMENT TRAP AUDIT
- Examine your analysis for each of the five traps from Lesson 5: sunk cost fallacy, anchoring, framing effect, loss aversion / status quo bias, and overconfidence. For each, write one sentence: either 'this trap may be present because...' or 'I do not see this trap here because...' Be honest. Where you identify a potential trap, describe how you have attempted to correct for it.
- PHASE 5: AI CONSULTATION
- Query an AI assistant with your decision. Ask it specifically: (a) What options have I not considered? (b) What is the strongest argument against my currently preferred option? (c) What information would most change this analysis? Document what the AI added, what it missed, and any factual claims you verified. Note whether it reflected your actual values or a proxy metric.
- PHASE 6: DECISION AND JOURNAL ENTRY
- Make your decision. Write a one-paragraph explanation of your reasoning that a thoughtful skeptic could evaluate and challenge. Then write a decision journal entry (as in Lesson 8): your decision, key uncertainties, prediction, confidence, and date. Plan your retrospective: when will you evaluate the outcome and by what standard?
- Submit all six phases. Evaluation criteria: rigor of uncertainty analysis, honesty of the trap audit, quality of AI consultation (did you actually steelman, not just consult?), and clarity of final reasoning.
What Good Analysis Looks Like
A common mistake in decision analysis is confusing thoroughness with quality. Length is not rigor. A good analysis is precise: it names specific outcomes with specific probability estimates, not vague gestures toward 'might work out.' It is honest: the trap audit identifies real potential traps, not just 'I do not think any traps apply.' It is connected to values: it explicitly states what matters and why, not just what is most efficient by an arbitrary metric. Another common mistake is treating the analysis as a machine that produces an answer. Decision analysis is a tool for structuring thinking — the output is better reasoning, not a guaranteed optimal outcome. You may complete a rigorous analysis and still face genuine uncertainty about which option is best. That is fine. The analysis should clarify what you know, what you do not know, and what matters most — making the remaining uncertainty visible rather than hiding it under a false sense of certainty.
Decision analysis is a tool, not a substitute for deciding. Analysis paralysis — perpetually seeking more information or more analysis to avoid commitment — is itself a poor decision strategy. At some point, the cost of additional analysis (in time and opportunity cost) exceeds the expected benefit of better information. Good decision-makers know when the analysis is good enough and pull the trigger.
During Phase 4 of the decision analysis project, a student writes 'no judgment traps apply to my analysis' for all five traps without elaboration. What is the most likely problem with this assessment?
A student's sensitivity analysis in Phase 3 shows that if her estimate of a key probability drops from 40% to 25%, her preferred option switches from Option A to Option B. What is the correct interpretation?