Day 8 Applied examples

June 10th, 2026

8.1 Announcements

  • Homework due this Friday
  • Project proposal due this Friday

8.2 Applied example for today

R script

8.2.1 Discussion

  • What is the risk of making inference over a single treatment factor when the estimated interaction seemed to be relevant to explain variability in the data?
  • Where do the degrees of freedom come from?

8.3 Some topics that will arise today

Consider the general linear model for a specific case of an experiment with a two-wat factorial in a CRD:

\[y_{ijk} \sim N(\mu_{ijk}, \sigma^2), \\ \mu_{ijk} = \mu + A_i +C_j + AC_{ij},\]

where \(A_i\) is the effect of the \(i\)th level of \(A\), \(C_j\) is the effect of the \(j\)th level of \(C\), and \(AC_{ij}\) is the interaction between the \(i\)th level of \(A\) and the \(j\)th level of \(C\).

Oftentimes, the objective of our study is to make mean comparisons: essentially, we’re interested in the mean difference as much (or more) as the treatment mean estimate. We can consider the mean difference is \(\mu_i - \mu_{i'}\), where \(i \neq i'\). We could find different approaches to comparing said means:

  • Evaluate the mean difference \(\mu_i - \mu_{i'}\) and its 95% confidence interval.
  • Do a hypothesis test where \(H_0: \mu_i - \mu_{i'} = 0\) and \(H_a: \mu_i - \mu_{i'} \neq 0\)

The objective of our study can also justo be obtaining treatment mean estimates. To describe them, we probably want to have point estimates and 95% confidence intervals.

Considering all the above:

  • How to obtain a confidence interval: \(CI_{\hat{\theta}, 95\%} = \hat{\theta} \pm se(\hat{\theta}) \cdot t_{\frac{\alpha}{2}, dfe}\).
  • Let’s discuss what that \(dfe\) may be.

8.4 Tomorrow & Friday

  • No class.
  • Submit proposal and homework.