📈Theoretical Statistics Unit 9 – Bayesian statistics

Foundations of Probability

• Probability quantifies the likelihood of an event occurring and ranges from 0 (impossible) to 1 (certain)
• Joint probability $P(A,B)$ represents the probability of events A and B occurring simultaneously
  • Calculated by multiplying the individual probabilities of A and B if they are independent events
• Conditional probability $P(A|B)$ measures the probability of event A occurring given that event B has already occurred
  • Calculated using the formula $P(A|B) = \frac{P(A,B)}{P(B)}$
• Marginal probability $P(A)$ represents the probability of event A occurring, regardless of the outcome of other events
  • Obtained by summing the joint probabilities of A with all possible outcomes of the other event(s)
• Independence of events occurs when the occurrence of one event does not affect the probability of another event
  • Mathematically, $P(A|B) = P(A)$ and $P(B|A) = P(B)$ for independent events A and B
• Random variables assign numerical values to the outcomes of a random experiment
  • Discrete random variables have countable outcomes (integers)
  • Continuous random variables have uncountable outcomes (real numbers)

Introduction to Bayesian Thinking

• Bayesian thinking involves updating beliefs (probabilities) about an event or hypothesis based on new evidence or data
• Prior probability represents the initial belief about an event or hypothesis before considering new evidence
• Likelihood quantifies the probability of observing the data given a specific hypothesis or parameter value
• Posterior probability represents the updated belief about an event or hypothesis after considering new evidence
  • Combines the prior probability and the likelihood using Bayes' theorem
• Bayesian inference draws conclusions about parameters or hypotheses based on the posterior distribution
• Bayesian thinking allows for the incorporation of prior knowledge and the updating of beliefs as new data becomes available
• Bayesian methods are particularly useful when dealing with limited data or when prior information is available

Bayes' Theorem and Its Components

• Bayes' theorem is a fundamental rule in Bayesian statistics that relates conditional probabilities
  • Mathematically, $P(A|B) = \frac{P(B|A)P(A)}{P(B)}$
• The components of Bayes' theorem include:
  • Prior probability $P(A)$: the initial belief about event A before considering evidence B
  • Likelihood $P(B|A)$: the probability of observing evidence B given that event A is true
  • Marginal likelihood $P(B)$: the probability of observing evidence B, regardless of the truth of event A
  • Posterior probability $P(A|B)$: the updated belief about event A after considering evidence B
• Bayes' theorem allows for the updating of beliefs by combining prior knowledge with new evidence
• The denominator $P(B)$ acts as a normalizing constant to ensure the posterior probabilities sum to 1
• Bayes' theorem is the foundation for Bayesian inference and parameter estimation

Prior Distributions: Types and Selection

• Prior distributions represent the initial beliefs about parameters or hypotheses before considering data
• Informative priors incorporate prior knowledge or expert opinion about the parameters
  • Conjugate priors result in posterior distributions that belong to the same family as the prior (mathematically convenient)
• Non-informative priors aim to minimize the influence of prior beliefs on the posterior distribution
  • Uniform prior assigns equal probability to all possible parameter values
  • Jeffreys prior is proportional to the square root of the Fisher information matrix
• Improper priors are not valid probability distributions but can still lead to proper posterior distributions
• Prior selection should be based on available prior knowledge, the nature of the problem, and the desired properties of the posterior distribution
• Sensitivity analysis can be performed to assess the impact of different prior choices on the posterior inference

Likelihood Functions and Their Role

• Likelihood functions quantify the probability of observing the data given specific parameter values
• The likelihood function is a key component in Bayesian inference and is combined with the prior distribution to obtain the posterior distribution
• For discrete data, the likelihood is the probability mass function evaluated at the observed data points
• For continuous data, the likelihood is the probability density function evaluated at the observed data points
• Maximum likelihood estimation (MLE) finds the parameter values that maximize the likelihood function
  • MLE provides a point estimate of the parameters but does not incorporate prior information
• The likelihood function is not a probability distribution over the parameters but rather a function of the parameters given the observed data
• The shape of the likelihood function provides information about the precision and uncertainty of the parameter estimates

Posterior Distributions and Inference

• Posterior distributions represent the updated beliefs about parameters or hypotheses after considering the data
• The posterior distribution is obtained by combining the prior distribution and the likelihood function using Bayes' theorem
  • Mathematically, $P(\theta|D) \propto P(D|\theta)P(\theta)$, where $\theta$ represents the parameters and $D$ represents the data
• Posterior inference involves summarizing and interpreting the posterior distribution
  • Point estimates: mean, median, or mode of the posterior distribution
  • Interval estimates: credible intervals (Bayesian confidence intervals) that contain a specified probability mass
• Posterior predictive distributions allow for making predictions about future observations based on the posterior distribution of the parameters
• Bayesian model selection compares different models based on their posterior probabilities or Bayes factors
• Bayesian decision theory combines the posterior distribution with a loss function to make optimal decisions under uncertainty

Bayesian vs. Frequentist Approaches

• Bayesian and frequentist approaches differ in their philosophical interpretation of probability and their treatment of parameters
• Bayesian approach:
  • Probability represents a degree of belief or uncertainty about an event or hypothesis
  • Parameters are treated as random variables with associated probability distributions (priors and posteriors)
  • Inference is based on the posterior distribution, which combines prior knowledge with observed data
• Frequentist approach:
  • Probability represents the long-run frequency of an event in repeated experiments
  • Parameters are treated as fixed, unknown quantities
  • Inference is based on the sampling distribution of estimators and the construction of confidence intervals and hypothesis tests
• Bayesian methods allow for the incorporation of prior knowledge and provide a natural framework for updating beliefs as new data becomes available
• Frequentist methods focus on the properties of estimators and the control of long-run error rates
• Bayesian and frequentist approaches can lead to different results, especially when dealing with small sample sizes or informative priors

Computational Methods in Bayesian Analysis

• Bayesian inference often involves complex posterior distributions that cannot be analytically derived
• Computational methods are used to approximate and sample from the posterior distribution
• Markov Chain Monte Carlo (MCMC) methods are widely used in Bayesian analysis
  • MCMC generates a Markov chain that converges to the posterior distribution
  • Metropolis-Hastings algorithm is a general MCMC method that proposes new parameter values and accepts or rejects them based on a probability ratio
  • Gibbs sampling is a special case of MCMC that samples from the full conditional distributions of the parameters
• Variational inference is an alternative to MCMC that approximates the posterior distribution with a simpler, tractable distribution
  • Minimizes the Kullback-Leibler divergence between the approximate and true posterior distributions
• Laplace approximation approximates the posterior distribution with a Gaussian distribution centered at the mode of the posterior
• Importance sampling and particle filtering are used for sequential Bayesian inference in dynamic models
• Software packages (JAGS, Stan, PyMC3) and probabilistic programming languages simplify the implementation of Bayesian models and computational methods