Proof: Maximum likelihood estimation for simple linear regression
Index:
The Book of Statistical Proofs ▷
Statistical Models ▷
Univariate normal data ▷
Simple linear regression ▷
Maximum likelihood estimation
Metadata: ID: P290 | shortcut: slr-mle2 | author: JoramSoch | date: 2021-11-16, 11:53.
Theorem: Given a simple linear regression model with independent observations
\[\label{eq:slr} y_i = \beta_0 + \beta_1 x_i + \varepsilon_i, \; \varepsilon_i \sim \mathcal{N}(0, \sigma^2), \; i = 1,\ldots,n \; ,\]the maximum likelihood estimates of $\beta_0$, $\beta_1$ and $\sigma^2$ are given by
\[\label{eq:slr-mle} \begin{split} \hat{\beta}_0 &= \bar{y} - \hat{\beta}_1 \bar{x} \\ \hat{\beta}_1 &= \frac{s_{xy}}{s_x^2} \\ \hat{\sigma}^2 &= \frac{1}{n} \sum_{i=1}^{n} (y_i - \hat{\beta}_0 - \hat{\beta}_1 x_i)^2 \end{split}\]where $\bar{x}$ and $\bar{y}$ are the sample means, $s_x^2$ is the sample variance of $x$ and $s_{xy}$ is the sample covariance between $x$ and $y$.
Proof: Simple linear regression is a special case of multiple linear regression with
\[\label{eq:slr-mlr} X = \left[ \begin{matrix} 1_n & x \end{matrix} \right] \quad \text{and} \quad \beta = \left[ \begin{matrix} \beta_0 \\ \beta_1 \end{matrix} \right]\]and weighted least sqaures estimates are given by
\[\label{eq:mlr-mle} \begin{split} \hat{\beta} &= (X^\mathrm{T} V^{-1} X)^{-1} X^\mathrm{T} V^{-1} y \\ \hat{\sigma}^2 &= \frac{1}{n} (y-X\hat{\beta})^\mathrm{T} V^{-1} (y-X\hat{\beta}) \; . \end{split}\]Under independent observations, the covariance matrix is
\[\label{eq:mlr-ind} V = I_n, \quad \text{such that} \quad V^{-1} = I_n \; .\]Thus, we can write out the estimate of $\beta$
\[\label{eq:slr-mle-b} \begin{split} \hat{\beta} &= \left( \left[ \begin{matrix} 1_n^\mathrm{T} \\ x^\mathrm{T} \end{matrix} \right] V^{-1} \left[ \begin{matrix} 1_n & x \end{matrix} \right] \right)^{-1} \left[ \begin{matrix} 1_n^\mathrm{T} \\ x^\mathrm{T} \end{matrix} \right] V^{-1} y \\ &= \left( \left[ \begin{matrix} 1_n^\mathrm{T} \\ x^\mathrm{T} \end{matrix} \right] \left[ \begin{matrix} 1_n & x \end{matrix} \right] \right)^{-1} \left[ \begin{matrix} 1_n^\mathrm{T} \\ x^\mathrm{T} \end{matrix} \right] y \end{split}\]which is equal to the ordinary least squares solution for simple linear regression:
\[\label{eq:slr-mle-b-qed} \begin{split} \hat{\beta}_0 &= \bar{y} - \hat{\beta}_1 \bar{x} \\ \hat{\beta}_1 &= \frac{s_{xy}}{s_x^2} \; . \end{split}\]Additionally, we can write out the estimate of $\sigma^2$:
\[\label{eq:slr-mle-s2} \begin{split} \hat{\sigma}^2 &= \frac{1}{n} (y-X\hat{\beta})^\mathrm{T} V^{-1} (y-X\hat{\beta}) \\ &= \frac{1}{n} \left( y - \left[ \begin{matrix} 1_n & x \end{matrix} \right] \left[ \begin{matrix} \hat{\beta}_0 \\ \hat{\beta}_1 \end{matrix} \right] \right)^\mathrm{T} \left( y - \left[ \begin{matrix} 1_n & x \end{matrix} \right] \left[ \begin{matrix} \hat{\beta}_0 \\ \hat{\beta}_1 \end{matrix} \right] \right) \\ &= \frac{1}{n} \left( y - \hat{\beta}_0 - \hat{\beta}_1 x \right)^\mathrm{T} \left( y - \hat{\beta}_0 - \hat{\beta}_1 x \right) \\ &= \frac{1}{n} \sum_{i=1}^{n} (y_i - \hat{\beta}_0 - \hat{\beta}_1 x_i)^2 \; . \end{split}\]∎
Sources: Metadata: ID: P290 | shortcut: slr-mle2 | author: JoramSoch | date: 2021-11-16, 11:53.