7.1r Probabilistic Thinking

Probabilistic Thinking

Probabilistic Thinking is the practice of making decisions and predictions based on the likelihood of various outcomes, rather than on certainty. It treats knowledge as a distribution of beliefs that can be updated as evidence accumulates.

Core Principles

• Assign probabilities to possible outcomes (expressed as numbers between 0 and 1, or as odds).
• Update beliefs when new information arrives—this is the essence of Bayesian reasoning.
• Quantify uncertainty explicitly so that trade‑offs and risks become visible.
• Use expected value (probability × payoff) to compare alternatives when stakes differ.

How to Use

1. List mutually exclusive scenarios for the situation at hand.
2. Assign prior probabilities based on existing knowledge or base rates.
3. Gather evidence and evaluate how likely each piece of evidence is under each scenario (likelihoods).
4. Apply Bayes’ rule (see below) to compute posterior probabilities.
5. Make decisions using the posterior distribution (e.g., choose the action with highest expected utility).
6. Iterate as more data becomes available.

Application Areas

• Investing & finance – portfolio allocation, option pricing, risk assessment.
• Science & research – hypothesis testing, experimental design, meta‑analysis.
• Medicine – diagnostic testing, treatment efficacy, screening programs.
• Everyday life – estimating commute times, evaluating news claims, personal goal setting.
• Artificial intelligence – Bayesian networks, spam filtering, recommendation systems.

Bayes’ Update (Bayesian Reasoning)

Bayes’ theorem provides the formal mechanism for updating probabilities in light of new evidence:

$$P(H\mid E)=\frac{P(E\mid H)P(H)}{P(E)}$$

• $P(H)$ – Prior: probability of hypothesis $H$ before seeing evidence.
• $P(E\mid H)$ – Likelihood: probability of observing evidence $E$ if $H$ is true.
• $P(E)$ – Marginal likelihood (or evidence): total probability of $E$ across all hypotheses; acts as a normalizing constant.
• $P(H\mid E)$ – Posterior: updated probability of $H$ after incorporating $E$.

Intuitive Steps

1. Start with your prior belief (how plausible you thought $H$ was).
2. Ask: “How surprising would the evidence be if $H$ were true?” → gives the likelihood.
3. Combine them; the more expected the evidence under $H$, the stronger the boost to $H$’s posterior.
4. Normalize across all competing hypotheses so probabilities sum to 1.

Why It’s Inextricably Linked

Probabilistic thinking without a mechanism for belief change is static—you can assign numbers but never learn. Bayes’ update supplies the learning rule that turns a probability assignment into a dynamic, evidence‑driven process. In practice, most real‑world probabilistic reasoning (forecasting, risk management, machine learning) relies on some form of Bayesian updating, even if approximated.

Related Notes

• 7.1u Second-Order Thinking – thinking about the consequences of consequences.
• 7.1q Opportunity Cost – evaluating what you forego by choosing one probabilistic outcome over another.
• 7.1i Confirmation Bias – a common pitfall where likelihoods are skewed to favor priors.
• 7.1p Occam’s Razor – preferring simpler hypotheses; often reflected in prior choices.
• 7.1b Base Rate Neglect – ignoring priors, leading to faulty Bayesian updates.
• 7.1k Expected Value – the decision‑theoretic companion to probabilistic thinking.

Quick Reference Cheat Sheet

Step	Action	Formula / Tip
1	Define hypotheses	$H_1,H_2,\dots ,H_n$
2	Set priors	$P(H_i)$ (use base rates or expert judgment)
3	Compute likelihoods	$P(E\mid H_i)$ (ask: how likely is evidence if $H_i$ true?)
4	Calculate unnormalized posteriors	$\tilde{P}(H_i\mid E)=P(E\mid H_i)P(H_i)$
5	Normalize	$P(H_i\mid E)=\frac{\tilde{P}(H_i\mid E)}{{\sum }_j\tilde{P}(H_j\mid E)}$
6	Decide	Choose action maximizing expected utility: ${\sum }_{i}U(a,H_i)P(H_i\mid E)$

---

Feel free to expand the “Application” bullet list with domain‑specific examples you encounter, and to link to any case‑study notes you create (e.g., [[Investing – Bayesian Portfolio]] or [[Medical Diagnosis – Bayes’ Rule]]).