How to move from “something is wrong” to “this is probably why, and this is the least foolish next move.”

Causal analysis helps you understand what is producing a problem. Decision analysis helps you choose what to do when options have trade-offs. Together they keep you from solving the wrong problem with great confidence, which is one of humanity’s more reliable hobbies.

This page covers causal diagrams, confounders, mediators, feedback loops, counterfactual thinking, decision matrices, decision trees, sensitivity analysis, scenario comparison, and value-focused thinking. Use it when people disagree about the cause, when several options look plausible, or when the decision matters enough that “seems fine” is not a responsible strategy.

 

Quick navigation

 

Best used for

  • Disputed causes.
  • Hard choices with trade-offs.
  • Repeated problems where symptoms keep returning.
  • Choosing between several plausible actions.
  • Testing whether a preferred solution fits the actual cause.

 

5-minute version

Use this when the problem is pressing, and you need the fastest, most responsible version of the method. Not perfect, but better than sprinting into a decision while waving a flaming assumption.

  1. Write the problem and the outcome you care about.
  2. List possible causes, not just the most obvious one.
  3. Ask what mechanism connects each cause to the outcome.
  4. Look for confounders: other factors that might explain the pattern.
  5. List your options and decision criteria.
  6. Choose the option that best fits your goals, risks, and constraints.

 

 

30-minute careful version

Use this when the issue matters enough to deserve a slower look. Thirty minutes of structured thinking can prevent thirty months of cleanup, which is apparently a bargain humans keep trying to avoid.

  1. Draw a simple causal map: causes, outcome, links, and feedback loops.
  2. Separate direct causes from indirect causes.
  3. Identify confounders, mediators, and reverse causation possibilities.
  4. Ask counterfactual questions: what would likely happen if this cause were absent?
  5. List possible actions.
  6. Create a decision matrix with criteria such as impact, cost, reversibility, risk, time, fairness, and evidence strength.
  7. Run sensitivity analysis: which assumptions would change the decision if they were wrong?
  8. Choose a provisional action and define how you will test whether it worked.

 

 

Vignette: The expensive symptom cure

A small business sees declining sales and blames marketing. They spend more on ads. Sales keep declining. The ad budget gets bigger, because apparently the solution to a leaky roof is more buckets.

A causal map shows that customer churn increased after support response times doubled. Marketing was bringing people in, but bad follow-up was pushing them out. The decision matrix compares ads, support staffing, product fixes, and customer interviews. The best first action is not more promotion. It is a support bottleneck fix plus churn interviews. The problem was causal confusion wearing a marketing hat.

 

 

 

Practice: apply this to one of your three current problems

Write down your three most important current problems. Pick one. Then apply the prompts below. Do not merely admire the tool from a safe distance like a museum visitor staring at a fire extinguisher.

  1. Choose one recurring problem.
  2. Draw a cause map with at least five possible causes.
  3. Mark, which causes are supported by evidence and which are guesses?
  4. List three action options.
  5. Compare them using five criteria.
  6. Write what assumption would change your choice.

 

 

Common mistakes

  • Mistaking correlation for causation.
  • Blaming the most visible person instead of tracing the mechanism.
  • Using a decision matrix with fake criteria that do not matter.
  • Ignoring reversibility and downside risk.
  • Failing to define how you will know whether the decision worked.

 

 

 

AI Prompt Support Module

Use AI as a thinking partner, not as a priest, judge, or magical vending machine for certainty. First, write your own answer. Then ask AI to challenge, improve, and stress-test it.

Build a causal map

Help me build a causal map for this problem: [describe]. Identify possible direct causes, indirect causes, confounders, reverse causation possibilities, feedback loops, and evidence I need to test the map.

Create a decision matrix

I need to choose between these options: [list]. Create a decision matrix using criteria that fit my goals, risks, constraints, fairness concerns, and reversibility. Explain how changes in assumptions could change the decision.

Run sensitivity analysis

Here is my preferred decision and assumptions: [describe]. Identify which assumptions are most fragile and which, if wrong, would change the recommended action.

FAQ

What is the difference between cause and correlation?

Correlation means two things move together. Causation means one thing helps produce another through some mechanism. The mechanism matters.

What if I cannot prove the cause?

Then state your confidence level, test the most likely causes, and choose reversible actions when uncertainty is high.

Are decision matrices too mechanical?

They can be if used badly. Used well, they reveal trade-offs and assumptions instead of pretending a complex decision is just a vibe with a spreadsheet.

 

 

Glossary

  • Causal diagram: A visual map of possible cause-and-effect relationships.
  • Confounder: A third factor that may explain a relationship between two variables.
  • Mediator: A factor through which a cause produces an outcome.
  • Sensitivity analysis: Testing how much a conclusion or decision changes when key assumptions change.
  • Value-focused thinking: Starting with the values and goals that should guide a decision before comparing options.

 

References and bibliography

These sources are included so readers can go deeper, check the intellectual foundations, and avoid treating this guide like it descended from the clouds on a glowing clipboard.

  1. Amos Tversky and Daniel Kahneman, “Judgment under Uncertainty: Heuristics and Biases,” Science, 1974. PubMed record.
  2. Richards J. Heuer Jr., Psychology of Intelligence Analysis, CIA Center for the Study of Intelligence. CIA PDF.
  3. Philip E. Tetlock and Dan Gardner, Superforecasting: The Art and Science of Prediction. See also Good Judgment Open’s explanation of probabilistic scoring. Good Judgment Open FAQ.

 

Next: Big Data Set Errors and Model Validation

The next page deals with one of the great modern confusions: the belief that more data automatically means better truth. It does not. Big data can reveal patterns, but it can also industrialize error.

You will learn how large datasets can still be biased, incomplete, contaminated, badly measured, non-representative, or causally misleading, because apparently even millions of rows can still be wrong with confidence.