AI summary by TwinMind

Live 2020-05-18!!! Bayes' Theorem

StatQuest with Josh Starmer

30m 6s

Introduction and Announcements

  • Josh Starmer hosting StatQuest livestream from Chapel Hill, North Carolina
  • Major announcement: StatQuest study guides now available at statquest.org for $3 each
    • Available guides: PCA (6 pages), Classification Trees (8 pages), Linear Regression, Naive Bayes
    • Working on new guides daily, taking requests from viewers
    • Part of "Josh Starmer needs money for groceries fund"
  • Global audience participation from: Dallas, India, Istanbul, South Korea, Mumbai, Vienna, Chennai, Karnataka, Helsinki, Turkey, Lima Peru, Russia, Honduras, Germany, Malaysia, Miami, Ireland, Mexico, Kolkata, Tunisia, Kenya, Borneo, Madras, Hyderabad, Taiwan, China, Ukraine, Saudi Arabia, Denmark, Indonesia, Brazil, Seattle, Bangalore, Italy, France

Conditional Probability Review

  • Following up on previous livestream about conditional probability
  • StatLand example with 14 people total surveyed about candy and soda preferences:
    • 2 people: love candy AND soda
    • 4 people: love candy only
    • 5 people: love soda only
    • 3 people: love neither
  • Calculated probabilities by dividing counts by total population (14)
  • Conditional probability formula: P(not candy but love soda | love soda) = 5/7 = 0.71
  • Alternative calculation: P(not candy but love soda | not candy) = 5/9 = 0.63
  • Key insight: conditional probability = probability of event scaled by knowledge we already have

Bayes Theorem Derivation

  • StatGuy's challenge: solve conditional probabilities without knowing P(not candy AND love soda)
  • Algebraic manipulation of conditional probability equations:
    • Multiply both sides by denominator terms
    • Set equal expressions equal to each other
    • Derive two forms of relationship
  • Successfully derived Bayes theorem through basic algebra
  • Bayes theorem allows converting between different conditional probabilities:
    • From P(A|B) to P(B|A) using known marginal probabilities

Practical Application Without Complete Data

  • StatGuy provides incomplete information:
    • P(not candy | love soda) = some value
    • P(love soda) ≈ 0.6 (estimated)
    • P(not candy) = 0.57
  • Using Bayes theorem with estimates: calculated P(love soda | not candy) = 0.75
  • Different from exact calculation (0.63) due to StatGuy's guess for P(love soda)
  • Real-world necessity: impossible to survey entire populations (e.g., all of India)

Bayesian Statistics Philosophy

  • Bayes theorem forms basis of Bayesian statistics
  • Philosophy centered on dealing with uncertainty and incomplete knowledge
  • Contrast with frequentist statistics (t-tests, ANOVA, linear models)
  • Bayesian approach: upfront admission of uncertainty, work with educated guesses
  • Complex topic involving approximation methods (similar to gradient descent in machine learning)
  • Future topic for detailed exploration

Standard Notation Review

  • Conventional notation differs from Josh's "clunky" but clear notation
  • Standard form omits redundant information:
    • P(love soda | not candy) instead of P(not candy AND love soda | not candy)
    • P(not candy | love soda) instead of P(not candy AND love soda | love soda)
  • Both conditional probabilities refer to same yellow cell in contingency table
  • Only difference: which marginal probability used for scaling (row vs column totals)

Q&A Session

  • Assumptions in Bayes theorem: none in the equation itself, but Bayesian statistics assumes incomplete knowledge
  • Relationship to maximum likelihood: likelihood ratio appears in numerator, denominator ensures probability distribution sums to 1
  • Connection to Naive Bayes: numerator used extensively, denominator often omitted as scaling factor
  • Turku, Finland viewer recognition

TwinMind

Get summaries like this for your own meetings

TwinMind takes notes in the background during meetings, lectures, and calls, then writes the summary for you.

Free plan to get started

No bot joins your calls

Phone, desktop & Chrome apps

Your next meeting, summarized like this

TwinMind listens in the background and turns meetings, lectures, and calls into searchable notes you can chat with. Start free.

Available on phone, desktop & Chrome

Your next meeting, summarized like this

TwinMind listens in the background and turns meetings, lectures, and calls into searchable notes you can chat with. Start free.

Available on phone, desktop & Chrome

Your next meeting, summarized like this

TwinMind listens in the background and turns meetings, lectures, and calls into searchable notes you can chat with. Start free.

Available on phone, desktop & Chrome