Reasons designs don't work
Most experimental situations call for standard designs that can be
constructed with many statistical software packages. Standard designs
have assured degrees of precision, orthogonality, and other optimal
properties that are important for the exploratory nature of most
experiments. In some situations, however, standard designs are
not appropriate or are impractical. These may include situations where
- The required blocking structure or blocking size of the
experimental situation does not fit into a standard blocked
- Not all combinations of the factor settings are feasible, or
for some other reason the region of experimentation is
constrained or irregularly shaped.
- A classical design needs to be 'repaired'. This can happen due
to improper planning with the original design treatment
combinations containing forbidden or unreachable combinations
that were not considered before the design was generated.
- A nonlinear model is appropriate.
- A quadratic or response surface design is required in the
presence of qualitative factors.
- The factors in the experiment include both components of a
mixture and other process variables.
- There are multiple sources of variation leading to nested or
hierarchical data structures and restrictions on what can be
- A standard fractional factorial design requires too many
treatment combinations for the given amount of time and/or