MathIsimple
Point Estimation Theory

Point Estimation & Cramér-Rao Inequality

Master the theoretical foundations of statistical estimation, from basic concepts to advanced efficiency theory and optimal estimator construction

Theoretical FoundationPractical MethodsEfficiency Theory

Learning Objectives

What you'll master in point estimation theory and applications

1

Master fundamental concepts of point estimation theory and evaluation criteria

2

Understand Method of Moments, Maximum Likelihood Estimation, and Least Squares methods

3

Learn Uniformly Minimum Variance Unbiased Estimators (UMVUE) and construction methods

4

Apply Cramér-Rao inequality and Fisher information in estimation efficiency analysis

5

Analyze estimator properties: unbiasedness, efficiency, consistency, and asymptotic normality

6

Solve practical estimation problems in statistical inference and data analysis

Major Estimation Methods

Three fundamental approaches to parameter estimation

Method of Moments
Estimate parameters by equating sample moments to population moments

Key Formulas:

Population k-th moment: μₖ = E[Xᵏ]

Sample k-th moment: aₙ,ₖ = (1/n)∑Xᵢᵏ

Parameter equation: θⱼ = hⱼ(μ₁,...,μₖ)

Moment estimator: θ̂ⱼ = hⱼ(aₙ,₁,...,aₙ,ₖ)

Applications:

  • Exponential distribution parameter estimation
  • Uniform distribution boundary estimation
  • Binomial distribution parameter estimation
  • General parametric family estimation

Examples:

Example 1: Exponential E(λ) - Population moment μ₁ = 1/λ, so λ̂ = 1/X̄

Example 2: Uniform U(a,b) - μ₁ = (a+b)/2, ν₂ = (b-a)²/12, so â = X̄ - √3Sₙ

Example 3: Binomial B(k,p) - μ₁ = kp, ν₂ = kp(1-p), so p̂ = (X̄ - Sₙ²)/X̄

Example 4: Normal N(μ,σ²) - μ̂ = X̄ (sample mean), σ̂² = Sₙ² (sample variance)

Properties:

  • Consistency under mild conditions
  • Asymptotic normality when moments exist
  • Simple computational procedure
  • May not be efficient compared to MLE
Maximum Likelihood Estimation
Find parameter values that maximize the likelihood of observed data

Key Formulas:

Likelihood function: L(θ;x) = ∏p(xᵢ;θ)

Log-likelihood: ℓ(θ;x) = log L(θ;x)

Likelihood equation: ∂ℓ/∂θᵢ = 0

MLE: θ̂ = arg max L(θ;x)

Applications:

  • Normal distribution parameter estimation
  • Exponential family parameter estimation
  • Censored data analysis
  • Complex statistical models

Examples:

Example 1: Normal N(μ,σ²) - L(μ,σ²) = ∏(1/(σ√2π))exp(-(xᵢ-μ)²/(2σ²)), giving μ̂ = X̄, σ̂² = Sₙ²

Example 2: Exponential E(λ) - L(λ) = λⁿexp(-λ∑xᵢ), giving λ̂ = 1/X̄

Example 3: Poisson P(λ) - L(λ) = ∏(λˣⁱe^(-λ)/xᵢ!), giving λ̂ = X̄

Example 4: Uniform U(0,θ) - L(θ) = θ^(-n) (non-differentiable), giving θ̂ = X₍ₙ₎

Properties:

  • Invariance property under transformations
  • Asymptotic efficiency under regularity conditions
  • Consistency and asymptotic normality
  • Optimal large-sample properties
Least Squares Estimation
Minimize sum of squared residuals in regression models

Key Formulas:

Model: Yᵢ = μᵢ(θ) + εᵢ

Objective: Q(θ) = ∑(Yᵢ - μᵢ(θ))²

LSE: θ̂ = arg min Q(θ)

Normal equations: ∂Q/∂θᵢ = 0

Applications:

  • Linear regression analysis
  • Polynomial curve fitting
  • Nonlinear regression models
  • Time series analysis

Examples:

Example 1: Simple linear regression Yᵢ = β₀ + β₁xᵢ + εᵢ - β̂₁ = ∑(xᵢ-x̄)(Yᵢ-Ȳ)/∑(xᵢ-x̄)²

Example 2: Polynomial fitting Y = β₀ + β₁x + β₂x² + ε - minimize ∑(Yᵢ - β₀ - β₁xᵢ - β₂xᵢ²)²

Example 3: Exponential model Y = αe^(βx) + ε - linearize via log transform and apply LSE

Example 4: Multiple regression Y = Xβ + ε - β̂ = (X'X)⁻¹X'Y (matrix form)

Properties:

  • Best Linear Unbiased Estimator (BLUE)
  • Computational efficiency
  • Robustness to distributional assumptions
  • Geometric interpretation

Estimator Evaluation Criteria

How to assess and compare the quality of statistical estimators

Unbiasedness
Estimator expected value equals true parameter

Mathematical Definition:

E_θ[θ̂] = θ for all θ ∈ Θ

Examples:

  • Sample mean X̄ is unbiased for population mean μ
  • Sample variance S² is unbiased for σ² in normal populations
  • Uncorrected sample variance S²ₙ is biased for σ²

Why it matters: Ensures estimator does not systematically over or underestimate the parameter

Efficiency
Comparison of estimator variances among unbiased estimators

Mathematical Definition:

Var_θ[θ̂₁] ≤ Var_θ[θ̂₂] for all θ

Examples:

  • Sample mean vs sample median for normal populations
  • MLE vs method of moments estimators
  • Efficient estimators achieve Cramér-Rao lower bound

Why it matters: Lower variance implies more precise estimation with smaller confidence intervals

Consistency
Estimator converges to true parameter as sample size increases

Mathematical Definition:

θ̂ₙ →^P θ (weak consistency)

Examples:

  • Sample mean converges to population mean
  • MLE is consistent under regularity conditions
  • Sample variance converges to population variance

Why it matters: Guarantees estimator accuracy improves with larger samples

Mean Squared Error
Combined measure of bias and variance

Mathematical Definition:

MSE_θ[θ̂] = Var_θ[θ̂] + Bias²_θ[θ̂]

Examples:

  • Trade-off between bias and variance
  • Biased estimators may have lower MSE
  • James-Stein estimator phenomenon

Why it matters: Comprehensive measure balancing accuracy and precision

Cramér-Rao Theory & Efficiency

Fisher information and fundamental bounds on estimator variance

Fisher Information
Measure of information content about parameter in the data

Key Formula:

I(θ) = E_θ[(∂log p(X;θ)/∂θ)²]

Properties:

  • I(θ) = -E_θ[∂²log p(X;θ)/∂θ²] (second derivative form)
  • Sample Fisher information: Iₙ(θ) = nI(θ)
  • Higher information implies better estimation potential
  • Related to asymptotic variance of MLE
Cramér-Rao Lower Bound
Lower bound on variance of unbiased estimators

Key Formula:

Var_θ[ĝ] ≥ [g'(θ)]²/Iₙ(θ)

Properties:

  • Applies to unbiased estimators under regularity conditions
  • Equality achieved by efficient estimators
  • Lower bound increases with [g'(θ)]²
  • Inversely related to Fisher information
UMVUE Theory
Uniformly Minimum Variance Unbiased Estimators

Key Formula:

Var_θ[ĝ*] ≤ Var_θ[ĝ] for all unbiased ĝ

Properties:

  • Rao-Blackwell theorem for improvement
  • Lehmann-Scheffé theorem for uniqueness
  • Complete sufficient statistics play key role
  • Zero unbiased estimation approach

Advanced Topics & Extensions

Further developments in estimation theory

Asymptotic Properties
Large-sample behavior of estimators
  • Consistency and convergence modes
  • Asymptotic normality and central limit theorems
  • Delta method for function estimation
  • Asymptotic efficiency comparisons
Non-regular Cases
Estimation when regularity conditions fail
  • Support depends on parameter (uniform distribution)
  • Non-differentiable likelihood functions
  • Boundary parameter problems
  • Alternative bounds and efficiency measures
Robust Estimation
Estimation methods resistant to outliers
  • M-estimators and Huber functions
  • Breakdown point and influence functions
  • Least absolute deviation methods
  • Trimmed and Winsorized estimators

📚 Explore Related Resources

Deepen your understanding with practice problems, formulas, and applications

Practice Problems

Test your understanding with carefully designed problems covering all estimation methods

Practice Now

Formula Reference

Comprehensive formulas for estimation methods, bounds, and efficiency measures

View Formulas

Mathematical Statistics

Review fundamental concepts in mathematical statistics and inference theory

Review Fundamentals

Master estimation theory