Likelihood Thinking

Uncertainty & Probability

Medium
Likelihood thinking compares explanations by asking which one would make the observed evidence more expected. It is especially useful when you are not ready to compute full probabilities but still want a disciplined way to compare competing hypotheses.
Reasoning type
Comparative probabilistic
Certainty level
Relative rather than absolute
Cognitive load
Medium
Formality
Medium to High

Core Idea

Definition

Likelihood thinking evaluates hypotheses by comparing how well each hypothesis would predict the evidence you actually observed.

In Plain English

Instead of asking only whether a story could explain the facts, ask which story would have made those facts more likely.

Framework Structure

Components

Observed Evidence
Competing Hypotheses
Predicted Evidence Under Each Hypothesis
Relative Support

Flow

List hypotheses -> Ask what each would predict -> Compare how expected the evidence would be -> Prefer the better predictor

How to Apply

  • 1.Define the evidence clearly before debating interpretations
  • 2.Write down at least two plausible competing hypotheses
  • 3.Ask what each hypothesis would lead you to expect if it were true
  • 4.Compare which hypothesis makes the observed evidence less surprising
  • 5.Use the result as comparative support rather than total proof

When to Use

  • Diagnosis and investigation
  • Scientific interpretation of evidence
  • Comparing strategic explanations for observed behavior
  • Evaluating rival narratives around a confusing event
  • Any situation where multiple hypotheses can fit the same facts

When NOT to Use

  • When only one hypothesis has been generated
  • When the evidence is too vague to discriminate between explanations
  • When prior odds dominate and cannot be ignored
  • When the exercise is being used to rationalize a favored story

Example

Problem

A team's churn rises sharply after a pricing change.

Application

  • 1.Compare hypotheses such as price sensitivity, onboarding failure, or seasonal customer turnover
  • 2.Ask what each hypothesis would predict in segment-level data, support conversations, and activation metrics
  • 3.Notice that the price-sensitivity hypothesis best predicts the pattern among lower-usage customers
  • 4.Treat that hypothesis as more strongly supported and investigate further

Conclusion

The evidence favors price sensitivity over the main alternatives because it better predicts the observed pattern.

Takeaway

Likelihood thinking rewards explanations that would have expected the evidence, not just ones that can explain it afterward.

Common Mistakes

  • Confusing possible with likely
  • Evaluating whether a hypothesis can fit the facts rather than whether it predicts them well
  • Ignoring the need for competing alternatives
  • Forgetting that evidence can support one hypothesis more while still not proving it true
  • Smuggling in post hoc explanations after the evidence is already known

How to Practice

prediction column

When comparing hypotheses, add a column for what each would predict before deciding which is stronger.

could vs would

Replace the question could this explain it with would this have expected it.

rival hypothesis drill

For puzzling outcomes, practice generating two rival explanations and comparing their predictions.

Related Cognitive Biases

confirmation bias

People often search for ways their preferred theory could fit the facts rather than comparing predictions fairly.

narrative fallacy

A coherent story can feel persuasive even if it predicted little in advance.

base rate neglect

Relative fit to evidence can still mislead if starting plausibility is ignored entirely.

Related Frameworks

Related Skills

hypothesis generation
comparing evidence
evaluating credibility
deriving conclusions

Variants & Extensions

Likelihood ratios
Diagnostic evidence comparison
Hypothesis discrimination
Model evidence intuition

Typical Failure Modes

  • No rival hypothesis
  • Post hoc storytelling
  • Ignoring priors entirely

Further Reading

  • The Theory That Would Not Die by Sharon Bertsch McGrayne
  • Superforecasting by Philip E. Tetlock and Dan Gardner
  • The Art of Statistics by David Spiegelhalter