Study Questions, Chapter 15
1. Why do we call these designs 2k designs?
- Because they are multiple-factor designs, and each factor has 2
levels. Here, "k" refers to the number of factors.
2. Higher-order designs are _______________________, often used
in __________ ______________ experiments.
- factor screening
3. If there are replicates (Gs instead of Ss), what do we use as our MSE?
- We use our ordinary estimate of sigma-squared, the sum of squared
4. If there are no replicates, what is the problem?
- Our ordinary estimate of sigma-squared equals zero
5. What can we do about it (four methods)?
- remove small effects, use as error term
- remove interactions, use as error term
- remove the one smallest factor and all its interactions - "hidden replication"
- hidden replication but with more than one factor
6. In that case, what extra assumption(s) do we have to make?
- That the effects that have been pooled together into creating the
error term really are zero.
7. How can we graphically assess effect sizes? How does this help us?
- Make a quantile plot of q vs. sorted effect sizes
- The central, linear part contains effects thought to be noise-only,
and candidates for pooling into the error term
- The outliers, negative and positive, are the potentially
significant effects that should be tested