Things to know for the final exam

advertisement
Things to know for the final exam
Split plot design: Two (or more) factor design with experimental material of two
different sizes (whole plot – large and sub plot – small). Levels of one factor are applied
to whole plots and levels of a second factor are applied to sub plots. Sub plots can be
sorted to form the whole plot or the whole plot can be subdivided to form the subplots.
When experimental material is reused for the subplots this design is often referred to as a
repeated measures design.
2p factorial experiments: p factors each at 2 levels, Low (–1) or High (+1).
Estimated Full Effect = Y1  Y1
Estimated Half Effect = Y1  Y 
Y1  Y1 
2
JMP calls the estimated half effect the parameter estimate because it estimates the slope
parameter for the coded factor values.
Standard error of an estimated full effect =
MS Error
2
where n is the number of
n
values for a factor level mean.
If there is only one value for each treatment combination then you cannot estimate error
variation in the usual way. Below are several ways to analyze data from an un-replicated
2p factorial experiment.
1. Use a normal probability plot of estimated full effects to identify potentially
important effects.
2. If you can drop a factor, and all of its interactions, you can use pseudo-replication
to come up with an estimate of error variation.
3. You can combine the sums of squares associated with higher order interaction
terms to come up with an estimate of error variation.
4. You can run multiple trials at center points for the p factors and use the variation
in these center point trials as your estimate of error variation.
Download