Previous Page

|

Next Page

Analyzing Saturated Designs

If a design is saturated, there are only enough degrees of freedom to estimate the parameters specified in the master model, including the overall mean. It is impossible to estimate the error variance without making additional assumptions, such as the effect sparsity assumption used in Lenth's (1989) method.

Furthermore, it is impossible to detect outliers or the need for a response transform. Therefore, the Check Fit Assumptions window is not available. When you analyze responses in a saturated design, you have to assume that there are no outliers and that your data will require no transformation.

The analysis of a saturated design differs from that of an unsaturated design as follows:

1. By default, ADX displays a normal plot instead of the effect selector, and it highlights the effects determined to be active with Lenth's (1989) method. The error estimate, called a pseudostandard error (PSE), arises from the assumption that approximately 20% of the effects in any given situation will be active, and the rest of the effect estimates are zero-mean jointly normal random variables.
2. The effect selector will not show any values for a test, since these tests are unavailable.
3. You can select active effects through any of the automatic selection methods except ANOVA.
4. You cannot open the Check Assumptions window.