Fin

Inference

Probability vs. Statistics

Generalized Linear Models

\[g(E(Y|x)) = \beta_0 + \beta_1 \cdot X_1 + \beta_2 \cdot X_2 \ldots\]

Linear: \(g(\cdot) = \cdot\)

Logistic: \(g(\cdot) = logit(\cdot)\)

Poisson: \(g(\cdot) = \ln(\cdot)\)

Interpreting variables

• Categorical
• Interaction
• Linear
• Multicollinearity

Survival analysis

• Censored observations
• Survival vs. hazard models
• Partial likelihood

Model Building

Multiple comparisons

• How many tests (in the wild) are null? how many are true?
• How do we control FWER?
• How do we control FDR?
• How do we control a single type I error over multiple looks at the observations?

AI

Statisticians are vital to the beginning and the end of the process.

• What questions to ask?
• How were the data generated?
• Can causation be claimed?
• Can you generalize to the population?
• What is the context within which the analysis is taking place?