Bayesian Methods for Ecological and Environmental Modelling
2-3.30 pm: Session 3 Linear modelling - part 1
3.30-4 pm: Tea & coffee break
4-5 pm: Session 3 Linear modelling – part 2
The core of the scientific process involves:
Inference is the process of estimating models, parameters, and their uncertainties, using data
The mathematically correct way to do inference with conditional probability
“Likelihood” and “probability” used interchangeably in common speech.
Likelihood has a specific meaning in statistics:
Next:
Bayesian approach applies to all types of systems and models (in principle)
We want to predict one thing (y) on the basis of another (x)
A function that describes a linear relationship between the response, \(y\), and the predictor, \(x\).
\[\begin{aligned} y &= \color{black}{\textbf{Model}} + \text{Error} \\[6pt] &= \color{black}{\mathbf{f(\theta, x)}} + \epsilon \\[6pt] &= \mathrm{intercept} + \mathrm{slope} \cdot x + \epsilon \\[6pt] &= \alpha + \beta x + \epsilon \\[6pt] \theta &= (\alpha, \beta) \\[6pt] \end{aligned}\]
A function that describes a linear relationship between the response, \(y\), and the predictor, \(x\).
\[\begin{aligned} y &= \color{black}{\textbf{Model}} + \text{Error} \\[6pt] &= \color{black}{\mathbf{f(\theta, x)}} + \epsilon \\[6pt] &= \mathrm{intercept} + \mathrm{slope} \cdot x + \epsilon \\[6pt] &= \beta_0 + \beta_1 x + \epsilon \\[6pt] \theta &= (\beta_0, \beta_1) \\[6pt] \end{aligned}\]
\[ \begin{aligned} y &= \color{purple}{\textbf{Model}} + \text{Error} \\[8pt] &= \color{purple}{\mathbf{f(\theta, x)}} + \epsilon \\[8pt] &= \color{purple}{\alpha + \beta x} + \epsilon \\[8pt] \end{aligned} \]
\[\begin{aligned} y &= \color{purple}{\textbf{Model}} + \color{blue}{\textbf{Error}} \\[8pt] &= \color{purple}{\mathbf{f(\theta, x)}} + \color{blue}{\boldsymbol{\epsilon}} \\[8pt] &= \color{purple}{\alpha + \beta x} + \color{blue}{\boldsymbol{\epsilon}} \\[8pt] \end{aligned}\]
\[\begin{aligned} y &= \color{purple}{\textbf{Model}} + \color{blue}{\textbf{Error}} \\[8pt] &= \color{purple}{\mathbf{f(\theta, x)}} + \color{blue}{\boldsymbol{\epsilon}} \\[8pt] &= \color{purple}{\alpha + \beta x} + \color{blue}{\boldsymbol{\epsilon}} \\[8pt] \end{aligned}\]
Regression slopes \(\beta\) are often referred to as effects
When the assumptions are not met …
Two practicals
rstanarm
Consider:
How does tree mass scale with stem diameter? Can we reliably estimate forest carbon stocks from simple measurements?
How does tree mass scale with stem diameter? Can we reliably estimate forest carbon stocks from simple measurements?
How do we assess risks from uncertain linear relationships? How do we combine data with prior knowledge?