Index: The Book of Statistical ProofsGeneral TheoremsFrequentist statisticsLikelihood theory ▷ Log-likelihood ratio

Definition: Let $m_0$ and $m_1$ be two generative models describing the same measured data $y$ using different model parameters $\theta_0 \in \Theta_0$ and $\theta_1 \in \Theta_1$. Then, the logarithmized quotient of the maximized likelihood functions of these two models is denoted as $\log \Lambda_{01}$ and is called the log-likelihood ratio of $m_0$ relative to $m_1$:

\[\label{eq:llr} \log \Lambda_{01} = \log \frac{\operatorname*{max}_{\theta_0 \in \Theta_0} \mathcal{L}_{m_0}(\theta_0)}{\operatorname*{max}_{\theta_1 \in \Theta_1} \mathcal{L}_{m_1}(\theta_1)} = \log p(y|\hat{\theta}_0,m_0) - \log p(y|\hat{\theta}_1,m_1) \; .\]
 
Sources:

Metadata: ID: D199 | shortcut: llr | author: JoramSoch | date: 2024-06-14, 14:45.